Update image URLs

For the num of requests serving charts always take the max value from the interval

.gitignore

@@ -12,7 +12,6 @@ test-reports
 .pytest_cache
 venv
 *.noseids
-build
 *.egg-info
 .cache
 .mypy_cache

LICENSE

@@ -1,7 +1,7 @@
                      Server Side Public License
                      VERSION 1, OCTOBER 16, 2018
 
-                    Copyright © 2019 allegro.ai, Inc.
+                    Copyright © 2025 ClearML Inc.
 
   Everyone is permitted to copy and distribute verbatim copies of this
   license document, but changing it is not allowed.

README.md

@@ -1,45 +1,41 @@
-# Trains Server
+<div align="center">
 
-##  Auto-Magical Experiment Manager & Version Control for AI - ε Devops Included!
+<img src="docs/clearml_server_logo.png" width="250px">
+
+**ClearML - Auto-Magical Suite of tools to streamline your ML workflow 
+</br>Experiment Manager, ML-Ops and Data-Management**
 
 [![GitHub license](https://img.shields.io/badge/license-SSPL-green.svg)](https://img.shields.io/badge/license-SSPL-green.svg)
-[![Python versions](https://img.shields.io/badge/python-3.6%20%7C%203.7-blue.svg)](https://img.shields.io/badge/python-3.6%20%7C%203.7-blue.svg)
-[![GitHub version](https://img.shields.io/github/release-pre/allegroai/trains-server.svg)](https://img.shields.io/github/release-pre/allegroai/trains-server.svg)
-[![PyPI status](https://img.shields.io/badge/status-beta-yellow.svg)](https://img.shields.io/badge/status-beta-yellow.svg)
+[![Python versions](https://img.shields.io/badge/python-3.9-blue.svg)](https://img.shields.io/badge/python-3.9-blue.svg)
+[![GitHub version](https://img.shields.io/github/release-pre/clearml/clearml-server.svg)](https://img.shields.io/github/release-pre/clearml/clearml-server.svg)
+[![Artifact Hub](https://img.shields.io/endpoint?url=https://artifacthub.io/badge/repository/clearml)](https://artifacthub.io/packages/search?repo=clearml)
 
-### Help improve Trains by filling our 2-min [user survey](https://allegro.ai/lp/trains-user-survey/)
+</div>
 
-## :rocket: Trains-Agent Services is now included, for more information see [services](https://github.com/allegroai/trains-server#services)
+## ClearML Server
+#### *Formerly known as Trains Server*
 
-## v0.16 Upgrade Notice
-
-In v0.16, the Elasticsearch subsystem of Trains Server has been upgraded from version 5.6 to version 7.6. This change necessitates the migration of the database contents to accommodate the change in index structure across the different versions.  
-
-Follow [this procedure](https://allegro.ai/docs/deploying_trains/trains_server_es7_migration/) to migrate existing data.
-
-## Introduction
-
-The **trains-server** is the backend service infrastructure for [Trains](https://github.com/allegroai/trains).
+The **ClearML Server** is the backend service infrastructure for [ClearML](https://github.com/clearml/clearml).
 It allows multiple users to collaborate and manage their experiments.
-By default, **Trains** is set up to work with the **Trains** demo server, which is open to anyone and resets periodically.
-In order to host your own server, you will need to launch **trains-server** and point **Trains** to it.
+**ClearML** offers a [free hosted service](https://app.clear.ml/), which is maintained by **ClearML** and open to anyone.
+In order to host your own server, you will need to launch the **ClearML Server** and point **ClearML** to it.
 
-**trains-server** contains the following components:
+The **ClearML Server** contains the following components:
 
-* The **Trains** Web-App, a single-page UI for experiment management and browsing
+* The **ClearML** Web-App, a single-page UI for experiment management and browsing
 * RESTful API for:
     * Documenting and logging experiment information, statistics and results
     * Querying experiments history, logs and results
 * Locally-hosted file server for storing images and models making them easily accessible using the Web-App
 
-You can quickly [deploy](#launching-trains-server)  your **trains-server** using Docker, AWS EC2 AMI, or Kubernetes. 
+You can quickly [deploy](#launching-the-clearml-server)  your **ClearML Server** using Docker, AWS EC2 AMI, or Kubernetes. 
 
 ## System design
 
 
-![Alt Text](https://github.com/allegroai/trains/blob/master/docs/system_diagram.png?raw=true)
+![Alt Text](docs/ClearML_Server_Diagram.png)
 
-**trains-server** has two supported configurations:
+The **ClearML Server** has two supported configurations:
 - Single IP (domain) with the following open ports
     - Web application on port 8080
     - API service on port 8008
@@ -50,11 +46,11 @@ You can quickly [deploy](#launching-trains-server)  your **trains-server** using
     - API service on sub-domain: api.\*.\*
     - File storage service on sub-domain: files.\*.\*
     
-## Launching trains-server
+## Launching The ClearML Server
 
 ### Prerequisites
 
-The ports 8080/8081/8008 must be available for the **trains-server** services.
+The ports 8080/8081/8008 must be available for the **ClearML Server** services.
    
 For example, to see if port `8080` is in use:
 
@@ -68,24 +64,23 @@ For example, to see if port `8080` is in use:
    
 ### Launching   
     
-Launch **trains-server** in any of the following formats:
+Launch The **ClearML Server** in any of the following formats:
 
-- Pre-built [AWS EC2 AMI](https://allegro.ai/docs/deploying_trains/trains_server_aws_ec2_ami/)
-- Pre-built [GCP Custom Image](https://allegro.ai/docs/deploying_trains/trains_server_gcp/)
+- Pre-built [AWS EC2 AMI](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_aws_ec2_ami)
+- Pre-built [GCP Custom Image](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_gcp)
 - Pre-built Docker Image
-    - [Linux](https://allegro.ai/docs/deploying_trains/trains_server_linux_mac/)
-    - [macOS](https://allegro.ai/docs/deploying_trains/trains_server_linux_mac/)
-    - [Windows 10](https://allegro.ai/docs/deploying_trains/trains_server_win/)
+    - [Linux](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_linux_mac)
+    - [macOS](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_linux_mac)
+    - [Windows 10](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_win)
 - Kubernetes    
-    - [Kubernetes Helm](https://allegro.ai/docs/deploying_trains/trains_server_kubernetes_helm/)
-    - Manual [Kubernetes installation](https://allegro.ai/docs/deploying_trains/trains_server_kubernetes/)
+    - [Kubernetes Helm](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_kubernetes_helm)
+    - Manual [Kubernetes installation](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_kubernetes)
 
-## Connecting Trains to your trains-server
+## Connecting ClearML to your ClearML Server
 
-By default, the **Trains** client is set up to work with the [**Trains** demo server](https://demoapp.trains.allegro.ai/).  
-To have the **Trains** client use your **trains-server** instead:
-- Run the `trains-init` command for an interactive setup.
-- Or manually edit `~/trains.conf` file, making sure the server settings (`api_server`, `web_server`, `file_server`) are configured correctly, for example:
+In order to set up the **ClearML** client to work with your **ClearML Server**:
+- Run the `clearml-init` command for an interactive setup.
+- Or manually edit `~/clearml.conf` file, making sure the server settings (`api_server`, `web_server`, `file_server`) are configured correctly, for example:
 
         api {
             # API server on port 8008
@@ -98,44 +93,44 @@ To have the **Trains** client use your **trains-server** instead:
             files_server: "http://localhost:8081"
         }
 
-**Note**: If you have set up **trains-server** in a sub-domain configuration, then there is no need to specify a port number,
+**Note**: If you have set up your **ClearML Server** in a sub-domain configuration, then there is no need to specify a port number,
 it will be inferred from the http/s scheme.
 
-After launching the **trains-server** and configuring the **Trains** client to use the **trains-server**,
-you can [use](https://github.com/allegroai/trains#using-trains) **Trains** in your experiments and view them in your **trains-server** web server,
+After launching the **ClearML Server** and configuring the **ClearML** client to use the **ClearML Server**,
+you can [use](https://github.com/clearml/clearml) **ClearML** in your experiments and view them in your **ClearML Server** web server,
 for example http://localhost:8080.  
-For more information about the Trains client, see [**Trains**](https://github.com/allegroai/trains).
+For more information about the ClearML client, see [**ClearML**](https://github.com/clearml/clearml).
 
-## Trains-Agent Services  <a name="services"></a> 
+## ClearML-Agent Services  <a name="services"></a> 
 
-As of version 0.15 of **trains-server**, dockerized deployment includes a **Trains-Agent Services** container running as 
+As of version 0.15 of **ClearML Server**, dockerized deployment includes a **ClearML-Agent Services** container running as 
 part of the docker container collection.
 
-Trains-Agent Services is an extension of Trains-Agent that provides the ability to launch long-lasting jobs 
+ClearML-Agent Services is an extension of ClearML-Agent that provides the ability to launch long-lasting jobs 
 that previously had to be executed on local / dedicated machines. It allows a single agent to 
 launch multiple dockers (Tasks) for different use cases. To name a few use cases, auto-scaler service (spinning instances 
 when the need arises and the budget allows), Controllers (Implementing pipelines and more sophisticated DevOps logic),
 Optimizer (such as Hyper-parameter Optimization or sweeping), and Application (such as interactive Bokeh apps for 
 increased data transparency)
 
-Trains-Agent Services container will spin **any** task enqueued into the dedicated `services` queue. 
-Every task launched by Trains-Agent Services  will be registered as a new node in the system, 
+ClearML-Agent Services container will spin **any** task enqueued into the dedicated `services` queue. 
+Every task launched by ClearML-Agent Services  will be registered as a new node in the system, 
 providing tracking and transparency capabilities.  
-You can also run the Trains-Agent Services manually, see details in [trains-agent services mode](https://github.com/allegroai/trains-agent#trains-agent-services-mode-)
+You can also run the ClearML-Agent Services manually, see details in [ClearML-agent services mode](https://github.com/clearml/clearml-agent#clearml-agent-services-mode-)
 
 **Note**: It is the user's responsibility to make sure the proper tasks are pushed into the `services` queue. 
 Do not enqueue training / inference tasks into the `services` queue, as it will put unnecessary load on the server.
 
 ## Advanced Functionality
 
-**trains-server** provides a few additional useful features, which can be manually enabled:
+The **ClearML Server** provides a few additional useful features, which can be manually enabled:
  
-* [Web login authentication](https://allegro.ai/docs/faq/faq/#web-auth)
-* [Non-responsive experiments watchdog](https://allegro.ai/docs/faq/faq/#watchdog)  
+* [Web login authentication](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_config#web-login-authentication)
+* [Non-responsive experiments watchdog](https://clear.ml/docs/latest/docs/deploying_clearml/clearml_server_config#non-responsive-task-watchdog)  
 
-## Restarting trains-server
+## Restarting ClearML Server
 
-To restart the **trains-server**, you must first stop the containers, and then restart them.
+To restart the **ClearML Server**, you must first stop the containers, and then restart them.
 
    ```bash
    docker-compose down
@@ -144,12 +139,12 @@ To restart the **trains-server**, you must first stop the containers, and then r
 
 ## Upgrading <a name="upgrade"></a>
 
-**trains-server** releases are also reflected in the [docker compose configuration file](https://github.com/allegroai/trains-server/blob/master/docker-compose.yml).  
-We strongly encourage you to keep your **trains-server** up to date, by keeping up with the current release.
+**ClearML Server** releases are also reflected in the [docker compose configuration file](https://github.com/clearml/clearml-server/blob/master/docker/docker-compose.yml).  
+We strongly encourage you to keep your **ClearML Server** up to date, by keeping up with the current release.
 
 **Note**: The following upgrade instructions use the Linux OS as an example.
 
-To upgrade your existing **trains-server** deployment:
+To upgrade your existing **ClearML Server** deployment:
 
 1. Shut down the docker containers
    ```bash
@@ -158,10 +153,10 @@ To upgrade your existing **trains-server** deployment:
 
 1. We highly recommend backing up your data directory before upgrading.
 
-   Assuming your data directory is `/opt/trains`, to archive all data into `~/trains_backup.tgz` execute:
+   Assuming your data directory is `/opt/clearml`, to archive all data into `~/clearml_backup.tgz` execute:
 
    ```bash
-   sudo tar czvf ~/trains_backup.tgz /opt/trains/data
+   sudo tar czvf ~/clearml_backup.tgz /opt/clearml/data
    ```    
 
    <details>
@@ -169,51 +164,51 @@ To upgrade your existing **trains-server** deployment:
 
    To restore this example backup, execute:
    ```bash
-   sudo rm -R /opt/trains/data
-   sudo tar -xzf ~/trains_backup.tgz -C /opt/trains/data
+   sudo rm -R /opt/clearml/data
+   sudo tar -xzf ~/clearml_backup.tgz -C /opt/clearml/data
    ```
    </details>
 
 1. Download the latest `docker-compose.yml` file.
 
    ```bash
-   curl https://raw.githubusercontent.com/allegroai/trains-server/master/docker-compose.yml -o docker-compose.yml 
+   curl https://raw.githubusercontent.com/clearml/clearml-server/master/docker/docker-compose.yml -o docker-compose.yml 
    ```
 
-1. Configure the Trains-Agent Services (not supported on Windows installation). 
-   If `TRAINS_HOST_IP` is not provided, Trains-Agent Services will use the external 
-   public address of the **trains-server**. If `TRAINS_AGENT_GIT_USER` / `TRAINS_AGENT_GIT_PASS` are not provided, 
-   the Trains-Agent Services will not be able to access any private repositories for running service tasks.
+1. Configure the ClearML-Agent Services (not supported on Windows installation). 
+   If `CLEARML_HOST_IP` is not provided, ClearML-Agent Services will use the external 
+   public address of the **ClearML Server**. If `CLEARML_AGENT_GIT_USER` / `CLEARML_AGENT_GIT_PASS` are not provided, 
+   the ClearML-Agent Services will not be able to access any private repositories for running service tasks.
    
    ```bash
-   export TRAINS_HOST_IP=server_host_ip_here
-   export TRAINS_AGENT_GIT_USER=git_username_here
-   export TRAINS_AGENT_GIT_PASS=git_password_here
+   export CLEARML_HOST_IP=server_host_ip_here
+   export CLEARML_AGENT_GIT_USER=git_username_here
+   export CLEARML_AGENT_GIT_PASS=git_password_here
    ```
 
-1. Spin up the docker containers, it will automatically pull the latest **trains-server** build    
+1. Spin up the docker containers, it will automatically pull the latest **ClearML Server** build    
    ```bash
    docker-compose -f docker-compose.yml pull
    docker-compose -f docker-compose.yml up
    ```
 
-**\* If something went wrong along the way, check our FAQ: [Common Docker Upgrade Errors](https://allegro.ai/docs/faq/faq/#common-docker-upgrade-errors).**
+**\* If something went wrong along the way, check our FAQ: [Common Docker Upgrade Errors](https://clear.ml/docs/latest/docs/faq/).**
 
 
 ## Community & Support
 
-If you have any questions, look to the Trains [FAQ](https://allegro.ai/docs/faq/faq/), or
-tag your questions on [stackoverflow](https://stackoverflow.com/questions/tagged/trains) with '**trains**' tag.
+If you have any questions, look to the ClearML [FAQ](https://clear.ml/docs/latest/docs/faq), or
+tag your questions on [stackoverflow](https://stackoverflow.com/questions/tagged/clearml) with '**clearml**' tag.
 
-For feature requests or bug reports, please use [GitHub issues](https://github.com/allegroai/trains-server/issues).
+For feature requests or bug reports, please use [GitHub issues](https://github.com/clearml/clearml-server/issues).
 
-Additionally, you can always find us at *trains@allegro.ai*
+Additionally, you can always find us at *clearml@allegro.ai*
 
 ## License
 
 [Server Side Public License v1.0](https://github.com/mongodb/mongo/blob/master/LICENSE-Community.txt)
 
-**trains-server** relies on both [MongoDB](https://github.com/mongodb/mongo) and [ElasticSearch](https://github.com/elastic/elasticsearch).
+The **ClearML Server** relies on both [MongoDB](https://github.com/mongodb/mongo) and [ElasticSearch](https://github.com/elastic/elasticsearch).
 With the recent changes in both MongoDB's and ElasticSearch's OSS license, we feel it is our responsibility as a
 member of the community to support the projects we love and cherish.
 We believe the cause for the license change in both cases is more than just,

apiserver/LICENSE

@@ -1,7 +1,7 @@
                      Server Side Public License
                      VERSION 1, OCTOBER 16, 2018
 
-                    Copyright © 2019 allegro.ai, Inc.
+                    Copyright © 2025 ClearML Inc.
 
   Everyone is permitted to copy and distribute verbatim copies of this
   license document, but changing it is not allowed.

apiserver/apierrors/__init__.py

@@ -0,0 +1,6 @@
+from .apierror import APIError
+from .base import BaseError
+
+from apiserver.apierrors_generator import ErrorsGenerator
+
+ErrorsGenerator.generate_python_files()

apiserver/apierrors/apierror.py

@@ -1,9 +1,10 @@
 class APIError(Exception):
-    def __init__(self, msg, code=500, subcode=0, **_):
+    def __init__(self, msg, code=500, subcode=0, error_data=None, **_):
         super(APIError, self).__init__()
         self._msg = msg
         self._code = code
         self._subcode = subcode
+        self._error_data = error_data or {}
 
     @property
     def msg(self):
@@ -17,5 +18,9 @@ class APIError(Exception):
     def subcode(self):
         return self._subcode
 
+    @property
+    def error_data(self):
+        return self._error_data
+
     def __str__(self):
         return self.msg

apiserver/apierrors/base.py

@@ -1,9 +1,13 @@
-import six
-from boltons.typeutils import classproperty
 from typing import Tuple
 
+import six
+from boltons.iterutils import is_collection, remap
+from boltons.typeutils import classproperty
+
 from .apierror import APIError
 
+jsonable_types = (dict, list, tuple, str, int, float, bool, type(None))
+
 
 class BaseError(APIError):
     _default_code = 500
@@ -19,15 +23,26 @@ class BaseError(APIError):
                 f"{k}={self._format_kwarg(v)}" for k, v in kwargs.items()
             )
             message += f": {kwargs_msg}"
-        params = kwargs.copy()
-        params.update(
-            code=self._default_code, subcode=self._default_subcode, msg=message
+
+        super(BaseError, self).__init__(
+            code=self._default_code,
+            subcode=self._default_subcode,
+            msg=message,
+            error_data=self._to_safe_json_types(kwargs),
         )
-        super(BaseError, self).__init__(**params)
+
+    @staticmethod
+    def _to_safe_json_types(data):
+        def visit(_, k, v):
+            if not isinstance(v, jsonable_types):
+                v = str(v)
+            return k, v
+
+        return remap(data, visit=visit)
 
     @staticmethod
     def _format_kwarg(value):
-        if isinstance(value, (tuple, list)):
+        if is_collection(value):
             return f'({", ".join(str(v) for v in value)})'
         elif isinstance(value, six.string_types):
             return value

apiserver/apierrors/errors.conf

@@ -0,0 +1,162 @@
+301 {
+    _: "moved_permanently"
+    1: ["not_supported", "this endpoint is no longer supported for the requested API version"]
+}
+
+400 {
+    _: "bad_request"
+    1: ["not_supported", "endpoint is not supported"]
+    2: ["request_path_has_invalid_version", "request path has invalid version"]
+    5: ["invalid_headers", "invalid headers"]
+    6: ["impersonation_error", "impersonation error"]
+
+    10: ["invalid_id", "invalid object id"]
+    11: ["missing_required_fields", "missing required fields"]
+    12: ["validation_error", "validation error"]
+    13: ["fields_not_allowed_for_role", "fields not allowed for role"]
+    14: ["invalid fields", "fields not defined for object"]
+    15: ["fields_conflict", "conflicting fields"]
+    16: ["fields_value_error", "invalid value for fields"]
+    17: ["batch_contains_no_items", "batch request contains no items"]
+    18: ["batch_validation_error", "batch request validation error"]
+    19: ["invalid_lucene_syntax", "malformed lucene query"]
+    20: ["fields_type_error", "invalid type for fields"]
+    21: ["invalid_regex_error", "malformed regular expression"]
+    22: ["invalid_email_address", "malformed email address"]
+    23: ["invalid_domain_name", "malformed domain name"]
+    24: ["not_public_object", "object is not public"]
+
+    # Auth / Login
+    75: ["invalid_access_key", "access key not found"]
+
+    # Tasks
+    100: ["task_error", "general task error"]
+    101: ["invalid_task_id", "invalid task id"]
+    102: ["task_validation_error", "task validation error"]
+    110: ["invalid_task_status", "invalid task status"]
+    111: ["task_not_started", "task not started (invalid task status)"]
+    112: ["task_in_progress", "task in progress (invalid task status)"]
+    113: ["task_published", "task published (invalid task status)"]
+    114: ["task_status_unknown", "task unknown (invalid task status)"]
+    120: ["invalid_task_execution_progress", "invalid task execution progress"]
+    121: ["failed_changing_task_status", "failed changing task status. probably someone changed it before you"]
+    122: ["missing_task_fields", "task is missing expected fields"]
+    123: ["task_cannot_be_deleted", "task cannot be deleted"]
+    125: ["task_has_jobs_running", "task has jobs that haven't completed yet"]
+    126: ["invalid_task_type", "invalid task type for this operations"]
+    127: ["invalid_task_input", "invalid task output"]
+    128: ["invalid_task_output", "invalid task output"]
+    129: ["task_publish_in_progress", "Task publish in progress"]
+    130: ["task_not_found", "task not found"]
+    131: ["events_not_added", "events not added"]
+
+    # Reports
+    150: ["operation_supported_on_reports_only", "passed task is not report"]
+
+    # Pipelines
+    160: ["cannot_remove_all_runs", "at least one pipeline run should be left"]
+
+    # Models
+    200: ["model_error", "general task error"]
+    201: ["invalid_model_id", "invalid model id"]
+    202: ["model_not_ready", "model is not ready"]
+    203: ["model_is_ready", "model is ready"]
+    204: ["invalid_model_uri", "invalid model URI"]
+    205: ["model_in_use", "model is used by tasks"]
+    206: ["model_creating_task_exists", "task that created this model exists"]
+
+    # Users
+    300: ["invalid_user", "invalid user"]
+    301: ["invalid_user_id", "invalid user id"]
+    302: ["user_id_exists", "user id already exists"]
+    305: ["invalid_preferences_update", "Malformed key and/or value"]
+
+    # Projects
+    401: ["invalid_project_id", "invalid project id"]
+    402: ["project_has_tasks", "project has associated tasks"]
+    403: ["project_not_found", "project not found"]
+    405: ["project_has_models", "project has associated models"]
+    406: ["project_has_datasets", "project has associated non-empty datasets"]
+    407: ["invalid_project_name", "invalid project name"]
+    408: ["cannot_update_project_location", "Cannot update project location. Use projects.move instead"]
+    409: ["project_path_exceeds_max", "Project path exceed the maximum allowed depth"]
+    410: ["project_source_and_destination_are_the_same", "Project has the same source and destination paths"]
+    411: ["project_cannot_be_moved_under_itself", "Project can not be moved under itself in the projects hierarchy"]
+    412: ["project_cannot_be_merged_into_its_child", "Project can not be merged into its own child"]
+    413: ["project_has_pipelines", "project has associated pipelines with active controllers"]
+    414: ["public_project_exists", "Cannot create project. Public project with the same name already exists"]
+
+    # Queues
+    701: ["invalid_queue_id", "invalid queue id"]
+    702: ["queue_not_empty", "queue is not empty"]
+    703: ["invalid_queue_or_task_not_queued", "invalid queue id or task not in queue"]
+    704: ["removed_during_reposition", "task was removed by another party during reposition"]
+    705: ["failed_adding_during_reposition", "failed adding task back to queue during reposition"]
+    706: ["task_already_queued", "failed adding task to queue since task is already queued"]
+    707: ["no_default_queue", "no queue is tagged as the default queue for this company"]
+    708: ["multiple_default_queues", "more than one queue is tagged as the default queue for this company"]
+
+    # Database
+    800: ["data_validation_error", "data validation error"]
+    801: ["expected_unique_data", "value combination already exists (unique field already contains this value)"]
+
+    # Workers
+    1001: ["invalid_worker_id", "invalid worker id"]
+    1002: ["worker_registration_failed", "worker registration failed"]
+    1003: ["worker_registered", "worker is already registered"]
+    1004: ["worker_not_registered", "worker is not registered"]
+    1005: ["worker_stats_not_found", "worker stats not found"]
+
+    # Serving
+    1050: ["invalid_container_id", "invalid container id"]
+    1051: ["container_not_registered", "container is not registered"]
+    1052: ["no_containers_for_url", "no container instances found for service url"]
+
+    1104: ["invalid_scroll_id", "Invalid scroll id"]
+}
+
+401 {
+    _:  "unauthorized"
+    1:  ["not_authorized", "unauthorized (not authorized for endpoint)"]
+    2:  ["entity_not_allowed", "unauthorized (entity not allowed)"]
+    10: ["bad_auth_type", "unauthorized (bad authentication header type)"]
+    20: ["no_credentials", "unauthorized (missing credentials)"]
+    21: ["bad_credentials", "unauthorized (malformed credentials)"]
+    22: ["invalid_credentials", "unauthorized (invalid credentials)"]
+    30: ["invalid_token", "invalid token"]
+    31: ["blocked_token", "token is blocked"]
+    40: ["invalid_fixed_user", "fixed user ID was not found"]
+}
+
+403: {
+    _:  "forbidden"
+    10: ["routing_error", "forbidden (routing error)"]
+    12: ["blocked_internal_endpoint", "forbidden (blocked internal endpoint)"]
+    20: ["role_not_allowed", "forbidden (not allowed for role)"]
+    21: ["no_write_permission", "forbidden (modification not allowed)"]
+}
+
+410: {
+    _:  "gone"
+    1: ["not_supported", "thus endpoint is not supported any more"]
+}
+
+500 {
+    _:   "server_error"
+    0:   ["general_error", "general server error"]
+    1:   ["internal_error", "internal server error"]
+    2:   ["config_error", "configuration error"]
+    3:   ["build_info_error", "build info unavailable or corrupted"]
+    4:   ["low_disk_space", "Critical server error! Server reports low or insufficient disk space. Please resolve immediately by allocating additional disk space or freeing up storage space."]
+    10:  ["transaction_error", "a transaction call has returned with an error"]
+    # Database-related issues
+    100: ["data_error", "general data error"]
+    101: ["inconsistent_data", "inconsistent data encountered in document"]
+    102: ["database_unavailable", "database is temporarily unavailable"]
+    110: ["update_failed", "update failed"]
+
+    # Index-related issues
+    201: ["missing_index", "missing internal index"]
+
+    9999: ["not_implemented", "action is not yet implemented"]
+}

apiserver/apierrors_generator/__init__.py

@@ -0,0 +1 @@
+from .errors_generator import ErrorsGenerator

apiserver/apierrors_generator/__main__.py

@@ -0,0 +1,4 @@
+from .errors_generator import ErrorsGenerator
+
+if __name__ == '__main__':
+    ErrorsGenerator.generate_python_files()

apiserver/apierrors_generator/errors_generator.py

@@ -0,0 +1,31 @@
+from functools import reduce
+from pathlib import Path
+from typing import Union
+
+from pyhocon import ConfigFactory, ConfigTree
+
+from .generator import Generator
+
+
+class ErrorsGenerator:
+    _apierrors_path = Path(__file__).parents[1] / "apierrors"
+    _files = [_apierrors_path / "errors.conf"]
+
+    @classmethod
+    def _get_codes(cls):
+        return {
+            (k, v.pop("_")): v
+            for k, v in reduce(
+                ConfigTree.merge_configs, map(ConfigFactory.parse_file, cls._files),
+            ).items()
+        }
+
+    @classmethod
+    def add_errors_file(cls, path: Union[Path, str]):
+        cls._files.append(path)
+
+    @classmethod
+    def generate_python_files(cls):
+        Generator(cls._apierrors_path / "errors", format_pep8=False).make_errors(
+            cls._get_codes()
+        )

apiserver/apierrors_generator/generator.py

@@ -8,9 +8,12 @@ from pathlib import Path
 
 env = jinja2.Environment(
     loader=jinja2.FileSystemLoader(str(Path(__file__).parent)),
-    autoescape=jinja2.select_autoescape(disabled_extensions=('py',), default_for_string=False),
+    autoescape=jinja2.select_autoescape(
+        disabled_extensions=("py",), default_for_string=False
+    ),
     trim_blocks=True,
-    lstrip_blocks=True)
+    lstrip_blocks=True,
+)
 
 
 def env_filter(name=None):
@@ -19,14 +22,14 @@ def env_filter(name=None):
 
 @env_filter()
 def cls_name(name):
-    delims = list(map(re.escape, (' ', '_')))
-    parts = re.split('|'.join(delims), name)
-    return ''.join(x.capitalize() for x in parts)
+    delims = list(map(re.escape, (" ", "_")))
+    parts = re.split("|".join(delims), name)
+    return "".join(x.capitalize() for x in parts)
 
 
 class Generator(object):
-    _base_class_name = 'BaseError'
-    _base_class_module = 'apierrors.base'
+    _base_class_name = "BaseError"
+    _base_class_module = "apiserver.apierrors.base"
 
     def __init__(self, path, format_pep8=True, use_md5=True):
         self._use_md5 = use_md5
@@ -35,29 +38,37 @@ class Generator(object):
         self._path.mkdir(parents=True, exist_ok=True)
 
     def _make_init_file(self, path):
-        (self._path / path / '__init__.py').write_bytes('')
+        (self._path / path / "__init__.py").write_bytes(b"")
 
     def _do_render(self, file, template, context):
-        with file.open('w') as f:
+        with file.open("w") as f:
             result = template.render(
                 base_class_name=self._base_class_name,
                 base_class_module=self._base_class_module,
-                **context)
+                **context
+            )
             if self._format_pep8:
-                result = autopep8.fix_code(result, options={'aggressive': 1, 'verbose': 0, 'max_line_length': 120})
+                import autopep8
+
+                result = autopep8.fix_code(
+                    result,
+                    options={"aggressive": 1, "verbose": 0, "max_line_length": 120},
+                )
             f.write(result)
 
     def _make_section(self, name, code, subcodes):
         self._do_render(
-            file=(self._path / name).with_suffix('.py'),
-            template=env.get_template('templates/section.jinja2'),
-            context=dict(code=code, subcodes=list(subcodes.items()),))
+            file=(self._path / name).with_suffix(".py"),
+            template=env.get_template("templates/section.jinja2"),
+            context=dict(code=code, subcodes=list(subcodes.items()),),
+        )
 
     def _make_init(self, sections):
         self._do_render(
-            file=(self._path / '__init__.py'),
-            template=env.get_template('templates/init.jinja2'),
-            context=dict(sections=sections,))
+            file=(self._path / "__init__.py"),
+            template=env.get_template("templates/init.jinja2"),
+            context=dict(sections=sections,),
+        )
 
     def _key_to_str(self, data):
         if isinstance(data, dict):
@@ -66,11 +77,11 @@ class Generator(object):
 
     def _calc_digest(self, data):
         data = json.dumps(self._key_to_str(data), sort_keys=True)
-        return hashlib.md5(data.encode('utf8')).hexdigest()
+        return hashlib.md5(data.encode("utf8")).hexdigest()
 
     def make_errors(self, errors):
         digest = None
-        digest_file = self._path / 'digest.md5'
+        digest_file = self._path / "digest.md5"
         if self._use_md5:
             digest = self._calc_digest(errors)
             if digest_file.is_file():
@@ -79,7 +90,7 @@ class Generator(object):
 
         self._make_init(errors)
         for (code, section_name), subcodes in errors.items():
-            self._make_section(section_name, code, subcodes)
+            self._make_section(section_name, int(code), subcodes)
 
         if self._use_md5:
             digest_file.write_text(digest)

apiserver/apierrors_generator/templates/section.jinja2

@@ -5,5 +5,5 @@ from {{ base_class_module }} import {{ base_class_name }}
 {% for subcode, (name, msg) in subcodes %}
 
 
-{{ error_class(name|cls_name, msg, code, subcode) -}}
+{{ error_class(name|cls_name, msg, code, subcode|int) -}}
 {% endfor %}

apiserver/apimodels/__init__.py

@@ -1,19 +1,36 @@
-from __future__ import absolute_import
-
 from enum import Enum
-from typing import Union, Type, Iterable
+from numbers import Number
+from typing import Union, Type, Iterable, Mapping
 
 import jsonmodels.errors
 import six
 from jsonmodels import fields
-from jsonmodels.fields import _LazyType, NotSet
+from jsonmodels.fields import _LazyType, NotSet, EmbeddedField
 from jsonmodels.models import Base as ModelBase
 from jsonmodels.validators import Enum as EnumValidator
-from luqum.parser import parser, ParseError
+from mongoengine.base import BaseDocument
 from validators import email as email_validator, domain as domain_validator
 
-from apierrors import errors
-from utilities.json import loads, dumps
+from apiserver.apierrors import errors
+from apiserver.utilities.json import loads, dumps
+
+
+class EmailField(fields.StringField):
+    def validate(self, value):
+        super().validate(value)
+        if value is None:
+            return
+        if email_validator(value) is not True:
+            raise errors.bad_request.InvalidEmailAddress()
+
+
+class DomainField(fields.StringField):
+    def validate(self, value):
+        super().validate(value)
+        if value is None:
+            return
+        if domain_validator(value) is not True:
+            raise errors.bad_request.InvalidDomainName()
 
 
 def make_default(field_cls, default_value):
@@ -24,6 +41,34 @@ def make_default(field_cls, default_value):
     return _FieldWithDefault
 
 
+class OneOfEmbeddedField(EmbeddedField):
+    def __init__(
+        self,
+        *args,
+        discriminator_property: str,
+        discriminator_mapping: Mapping[str, type],
+        **kwargs,
+    ):
+        self.discriminator_property = discriminator_property
+        self.discriminator_mapping = discriminator_mapping
+        model_types = tuple(set(self.discriminator_mapping.values()))
+
+        super().__init__(model_types, *args, **kwargs)
+
+    def parse_value(self, value):
+        """Parse value to proper model type."""
+        if not isinstance(value, dict) or self.discriminator_property not in value:
+            return super().parse_value(value)
+
+        property_value = value.get(self.discriminator_property)
+        embed_type = self.discriminator_mapping.get(property_value)
+        if not embed_type:
+            raise jsonmodels.errors.ValidationError(
+                f"Could not find type matching discriminator property value: {property_value}"
+            )
+        return embed_type(**value)
+
+
 class ListField(fields.ListField):
     def __init__(self, items_types=None, *args, default=NotSet, **kwargs):
         if default is not NotSet and callable(default):
@@ -35,6 +80,8 @@ class ListField(fields.ListField):
         try:
             return super(ListField, self)._cast_value(value)
         except TypeError:
+            if len(self.items_types) == 1 and issubclass(self.items_types[0], Enum):
+                return self.items_types[0](value)
             return value
 
     def validate_single_value(self, item):
@@ -43,6 +90,22 @@ class ListField(fields.ListField):
             item.validate()
 
 
+class ScalarField(fields.BaseField):
+
+    """String field."""
+
+    types = (str, int, float, bool)
+
+
+class SafeStringField(fields.StringField):
+    """String field that can also accept numbers as input"""
+    def parse_value(self, value):
+        if isinstance(value, Number):
+            value = str(value)
+
+        return super().parse_value(value)
+
+
 class DictField(fields.BaseField):
     types = (dict,)
 
@@ -71,6 +134,29 @@ class DictField(fields.BaseField):
             for type_ in value_types
         )
 
+    def parse_value(self, values):
+        """Cast value to proper collection."""
+        result = self.get_default_value()
+
+        if values is None:
+            return result
+
+        if not self.value_types or not isinstance(values, dict):
+            return values
+
+        return {key: self._cast_value(value) for key, value in values.items()}
+
+    def _cast_value(self, value):
+        if isinstance(value, self.value_types):
+            return value
+        else:
+            if len(self.value_types) != 1:
+                tpl = 'Cannot decide which type to choose from "{types}".'
+                raise jsonmodels.errors.ValidationError(
+                    tpl.format(types=", ".join([t.__name__ for t in self.value_types]))
+                )
+            return self.value_types[0](**value)
+
     def validate(self, value):
         super(DictField, self).validate(value)
 
@@ -96,6 +182,15 @@ class DictField(fields.BaseField):
                 )
             )
 
+    def _elem_to_struct(self, value):
+        try:
+            return value.to_struct()
+        except AttributeError:
+            return value
+
+    def to_struct(self, values):
+        return {k: self._elem_to_struct(v) for k, v in values.items()}
+
 
 class IntField(fields.IntField):
     def parse_value(self, value):
@@ -105,23 +200,6 @@ class IntField(fields.IntField):
             return value
 
 
-def validate_lucene_query(value):
-    if value == "":
-        return
-    try:
-        parser.parse(value)
-    except ParseError as e:
-        raise errors.bad_request.InvalidLuceneSyntax(error=e)
-
-
-class LuceneQueryField(fields.StringField):
-    def validate(self, value):
-        super(LuceneQueryField, self).validate(value)
-        if value is None:
-            return
-        validate_lucene_query(value)
-
-
 class NullableEnumValidator(EnumValidator):
     """Validator for enums that allows a None value."""
 
@@ -137,7 +215,7 @@ class EnumField(fields.StringField):
         *args,
         required=False,
         default=None,
-        **kwargs
+        **kwargs,
     ):
         choices = list(map(self.parse_value, values_or_type))
         validator_cls = EnumValidator if required else NullableEnumValidator
@@ -160,7 +238,7 @@ class ActualEnumField(fields.StringField):
         validators=None,
         required=False,
         default=None,
-        **kwargs
+        **kwargs,
     ):
         self.__enum = enum_class
         self.types = (enum_class,)
@@ -173,11 +251,11 @@ class ActualEnumField(fields.StringField):
             *args,
             required=required,
             validators=validators,
-            **kwargs
+            **kwargs,
         )
 
     def parse_value(self, value):
-        if value is None and not self.required:
+        if value is NotSet and not self.required:
             return self.get_default_value()
         try:
             # noinspection PyArgumentList
@@ -189,24 +267,6 @@ class ActualEnumField(fields.StringField):
         return super().to_struct(value.value)
 
 
-class EmailField(fields.StringField):
-    def validate(self, value):
-        super().validate(value)
-        if value is None:
-            return
-        if email_validator(value) is not True:
-            raise errors.bad_request.InvalidEmailAddress()
-
-
-class DomainField(fields.StringField):
-    def validate(self, value):
-        super().validate(value)
-        if value is None:
-            return
-        if domain_validator(value) is not True:
-            raise errors.bad_request.InvalidDomainName()
-
-
 class JsonSerializableMixin:
     def to_json(self: ModelBase):
         return dumps(self.to_struct())
@@ -214,3 +274,67 @@ class JsonSerializableMixin:
     @classmethod
     def from_json(cls: Type[ModelBase], s):
         return cls(**loads(s))
+
+
+def callable_default(cls: Type[fields.BaseField]) -> Type[fields.BaseField]:
+    class _Wrapped(cls):
+        _callable_default = None
+
+        def get_default_value(self):
+            if self._callable_default:
+                return self._callable_default()
+            return super(_Wrapped, self).get_default_value()
+
+        def __init__(self, *args, default=None, **kwargs):
+            if default and callable(default):
+                self._callable_default = default
+                default = default()
+            super(_Wrapped, self).__init__(*args, default=default, **kwargs)
+
+    return _Wrapped
+
+
+class MongoengineFieldsDict(DictField):
+    """
+    DictField representing mongoengine field names/value mapping.
+    Used to convert mongoengine-style field/subfield notation to user-presentable syntax, including handling update
+        operators.
+    """
+
+    mongoengine_update_operators = (
+        "inc",
+        "dec",
+        "push",
+        "push_all",
+        "pop",
+        "pull",
+        "pull_all",
+        "add_to_set",
+    )
+
+    @staticmethod
+    def _normalize_mongo_value(value):
+        if isinstance(value, BaseDocument):
+            return value.to_mongo()
+        return value
+
+    @classmethod
+    def _normalize_mongo_field_path(cls, path, value):
+        parts = path.split("__")
+        if len(parts) > 1:
+            if parts[0] == "set":
+                parts = parts[1:]
+            elif parts[0] == "unset":
+                parts = parts[1:]
+                value = None
+            elif parts[0] in cls.mongoengine_update_operators:
+                return None, None
+        return ".".join(parts), cls._normalize_mongo_value(value)
+
+    def parse_value(self, value):
+        value = super(MongoengineFieldsDict, self).parse_value(value)
+        return {
+            k: v
+            for k, v in (self._normalize_mongo_field_path(*p) for p in value.items())
+            if k is not None
+        }

apiserver/apimodels/auth.py

@@ -2,10 +2,10 @@ from jsonmodels.fields import IntField, StringField, BoolField, EmbeddedField, D
 from jsonmodels.models import Base
 from jsonmodels.validators import Max, Enum
 
-from apimodels import ListField, EnumField
-from config import config
-from database.model.auth import Role
-from database.utils import get_options
+from apiserver.apimodels import ListField, EnumField
+from apiserver.config_repo import config
+from apiserver.database.model.auth import Role
+from apiserver.database.utils import get_options
 
 
 class GetTokenRequest(Base):
@@ -75,11 +75,17 @@ class CreateUserResponse(Base):
 class Credentials(Base):
     access_key = StringField(required=True)
     secret_key = StringField(required=True)
+    label = StringField()
 
 
 class CredentialsResponse(Credentials):
     secret_key = StringField()
     last_used = DateTimeField(default=None)
+    last_used_from = StringField()
+
+
+class CreateCredentialsRequest(Base):
+    label = StringField()
 
 
 class CreateCredentialsResponse(Base):
@@ -90,6 +96,11 @@ class GetCredentialsResponse(Base):
     credentials = ListField(CredentialsResponse)
 
 
+class EditCredentialsRequest(Base):
+    access_key = StringField(required=True)
+    label = StringField()
+
+
 class RevokeCredentialsRequest(Base):
     access_key = StringField(required=True)
 

apiserver/apimodels/base.py

@@ -0,0 +1,28 @@
+from jsonmodels import models, fields
+from jsonmodels.validators import Length
+
+from apiserver.apimodels import MongoengineFieldsDict, ListField
+
+
+class UpdateResponse(models.Base):
+    updated = fields.IntField(required=True)
+    fields = MongoengineFieldsDict()
+
+
+class PagedRequest(models.Base):
+    page = fields.IntField()
+    page_size = fields.IntField()
+
+
+class IdResponse(models.Base):
+    id = fields.StringField(required=True)
+
+
+class MakePublicRequest(models.Base):
+    ids = ListField(items_types=str, validators=[Length(minimum_value=1)])
+
+
+class MoveRequest(models.Base):
+    ids = ListField([str], validators=Length(minimum_value=1))
+    project = fields.StringField()
+    project_name = fields.StringField()

apiserver/apimodels/batch.py

@@ -0,0 +1,25 @@
+from typing import Sequence
+
+from jsonmodels.fields import StringField
+from jsonmodels.models import Base
+from jsonmodels.validators import Length
+
+from apiserver.apimodels import ListField
+from apiserver.apimodels.base import UpdateResponse
+
+
+class BatchRequest(Base):
+    ids: Sequence[str] = ListField([str], validators=Length(minimum_value=1))
+
+
+class BatchResponse(Base):
+    succeeded: Sequence[dict] = ListField([dict])
+    failed: Sequence[dict] = ListField([dict])
+
+
+class UpdateBatchItem(UpdateResponse):
+    id: str = StringField()
+
+
+class UpdateBatchResponse(BatchResponse):
+    succeeded: Sequence[UpdateBatchItem] = ListField(UpdateBatchItem)

apiserver/apimodels/custom_validators/__init__.py

@@ -0,0 +1,34 @@
+import validators
+from jsonmodels.errors import ValidationError
+
+
+class ForEach(object):
+    def __init__(self, validator):
+        self.validator = validator
+
+    def validate(self, values):
+        for value in values:
+            self.validator.validate(value)
+
+    def modify_schema(self, field_schema):
+        return self.validator.modify_schema(field_schema)
+
+
+class Hostname(object):
+
+    def validate(self, value):
+        if validators.domain(value) is not True:
+            raise ValidationError(f"Value '{value}' is not a valid hostname")
+
+    def modify_schema(self, field_schema):
+        field_schema["format"] = "hostname"
+
+
+class Email(object):
+
+    def validate(self, value):
+        if validators.email(value) is not True:
+            raise ValidationError(f"Value '{value}' is not a valid email address")
+
+    def modify_schema(self, field_schema):
+        field_schema["format"] = "email"

apiserver/apimodels/events.py

@@ -0,0 +1,234 @@
+from enum import auto
+from typing import Sequence, Optional
+
+from jsonmodels import validators
+from jsonmodels.fields import StringField, BoolField, EmbeddedField
+from jsonmodels.models import Base
+from jsonmodels.validators import Length, Min, Max
+
+from apiserver.apimodels import ListField, IntField, ActualEnumField
+from apiserver.bll.event.event_common import EventType
+from apiserver.bll.event.scalar_key import ScalarKeyEnum
+from apiserver.config_repo import config
+from apiserver.utilities.stringenum import StringEnum
+
+
+class TaskRequest(Base):
+    task: str = StringField(required=True)
+
+
+class ModelRequest(Base):
+    model: str = StringField(required=True)
+
+
+class HistogramRequestBase(Base):
+    samples: int = IntField(default=2000, validators=[Min(1), Max(6000)])
+    key: ScalarKeyEnum = ActualEnumField(ScalarKeyEnum, default=ScalarKeyEnum.iter)
+
+
+class MetricVariants(Base):
+    metric: str = StringField(required=True)
+    variants: Sequence[str] = ListField(items_types=str)
+
+
+class ScalarMetricsIterHistogramRequest(HistogramRequestBase):
+    task: str = StringField(required=True)
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+    model_events: bool = BoolField(default=False)
+
+
+class GetMetricsAndVariantsRequest(Base):
+    task: str = StringField(required=True)
+    model_events: bool = BoolField(default=False)
+
+
+class MultiTaskScalarMetricsIterHistogramRequest(HistogramRequestBase):
+    tasks: Sequence[str] = ListField(
+        items_types=str,
+        validators=[
+            Length(
+                minimum_value=1,
+                maximum_value=config.get(
+                    "services.tasks.multi_task_histogram_limit", 100
+                ),
+            )
+        ],
+    )
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+    model_events: bool = BoolField(default=False)
+
+
+class TaskMetric(Base):
+    task: str = StringField(required=True)
+    metric: str = StringField(default=None)
+    variants: Sequence[str] = ListField(items_types=str)
+
+
+class LegacyMetricEventsRequest(TaskRequest):
+    iters: int = IntField(default=1, validators=validators.Min(1))
+    scroll_id: str = StringField()
+    model_events: bool = BoolField(default=False)
+
+
+class MetricEventsRequest(Base):
+    metrics: Sequence[TaskMetric] = ListField(
+        items_types=TaskMetric, validators=[Length(minimum_value=1)]
+    )
+    iters: int = IntField(default=1, validators=validators.Min(1))
+    navigate_earlier: bool = BoolField(default=True)
+    refresh: bool = BoolField(default=False)
+    scroll_id: str = StringField()
+    model_events: bool = BoolField(default=False)
+
+
+class VectorMetricsIterHistogramRequest(Base):
+    task: str = StringField(required=True)
+    metric: str = StringField(required=True)
+    variant: str = StringField(required=True)
+    model_events: bool = BoolField(default=False)
+
+
+class GetVariantSampleRequest(Base):
+    task: str = StringField(required=True)
+    metric: str = StringField(required=True)
+    variant: str = StringField(required=True)
+    iteration: Optional[int] = IntField()
+    refresh: bool = BoolField(default=False)
+    scroll_id: Optional[str] = StringField()
+    navigate_current_metric: bool = BoolField(default=True)
+    model_events: bool = BoolField(default=False)
+
+
+class GetMetricSamplesRequest(Base):
+    task: str = StringField(required=True)
+    metric: str = StringField(required=True)
+    iteration: Optional[int] = IntField()
+    refresh: bool = BoolField(default=False)
+    scroll_id: Optional[str] = StringField()
+    navigate_current_metric: bool = BoolField(default=True)
+    model_events: bool = BoolField(default=False)
+
+
+class NextHistorySampleRequest(Base):
+    task: str = StringField(required=True)
+    scroll_id: Optional[str] = StringField()
+    navigate_earlier: bool = BoolField(default=True)
+    next_iteration: bool = BoolField(default=False)
+    model_events: bool = BoolField(default=False)
+
+
+class LogOrderEnum(StringEnum):
+    asc = auto()
+    desc = auto()
+
+
+class TaskEventsRequestBase(Base):
+    task: str = StringField(required=True)
+    batch_size: int = IntField(default=500)
+
+
+class TaskEventsRequest(TaskEventsRequestBase):
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+    event_type: EventType = ActualEnumField(EventType, default=EventType.all)
+    order: Optional[str] = ActualEnumField(LogOrderEnum, default=LogOrderEnum.asc)
+    scroll_id: str = StringField()
+    count_total: bool = BoolField(default=True)
+    model_events: bool = BoolField(default=False)
+
+
+class LegacyLogEventsRequest(TaskEventsRequestBase):
+    order: Optional[str] = ActualEnumField(LogOrderEnum, default=LogOrderEnum.desc)
+    scroll_id: str = StringField()
+
+
+class LogEventsRequest(TaskEventsRequestBase):
+    batch_size: int = IntField(default=5000)
+    navigate_earlier: bool = BoolField(default=True)
+    from_timestamp: Optional[int] = IntField()
+    order: Optional[str] = ActualEnumField(LogOrderEnum)
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+
+
+class ScalarMetricsIterRawRequest(TaskEventsRequestBase):
+    batch_size: int = IntField()
+    key: ScalarKeyEnum = ActualEnumField(ScalarKeyEnum, default=ScalarKeyEnum.iter)
+    metric: MetricVariants = EmbeddedField(MetricVariants, required=True)
+    count_total: bool = BoolField(default=False)
+    scroll_id: str = StringField()
+    model_events: bool = BoolField(default=False)
+
+
+class IterationEvents(Base):
+    iter: int = IntField()
+    events: Sequence[dict] = ListField(items_types=dict)
+
+
+class MetricEvents(Base):
+    task: str = StringField()
+    iterations: Sequence[IterationEvents] = ListField(items_types=IterationEvents)
+
+
+class MetricEventsResponse(Base):
+    metrics: Sequence[MetricEvents] = ListField(items_types=MetricEvents)
+    scroll_id: str = StringField()
+
+
+class MultiTasksRequestBase(Base):
+    tasks: Sequence[str] = ListField(
+        items_types=str, validators=[Length(minimum_value=1)]
+    )
+    model_events: bool = BoolField(default=False)
+
+
+class SingleValueMetricsRequest(MultiTasksRequestBase):
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+
+
+class TaskMetricsRequest(MultiTasksRequestBase):
+    event_type: EventType = ActualEnumField(EventType, required=True)
+
+
+class MultiTaskMetricsRequest(MultiTasksRequestBase):
+    event_type: EventType = ActualEnumField(EventType, default=EventType.all)
+
+
+class LegacyMultiTaskEventsRequest(MultiTasksRequestBase):
+    iters: int = IntField(default=1, validators=validators.Min(1))
+    scroll_id: str = StringField()
+
+
+class MultiTaskPlotsRequest(MultiTasksRequestBase):
+    iters: int = IntField(default=1)
+    scroll_id: str = StringField()
+    no_scroll: bool = BoolField(default=False)
+    last_iters_per_task_metric: bool = BoolField(default=True)
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+
+
+class TaskPlotsRequest(Base):
+    task: str = StringField(required=True)
+    iters: int = IntField(default=1)
+    scroll_id: str = StringField()
+    no_scroll: bool = BoolField(default=False)
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+    model_events: bool = BoolField(default=False)
+
+
+class GetScalarMetricDataRequest(Base):
+    task: str = StringField(required=True)
+    metric: str = StringField(required=True)
+    scroll_id: str = StringField()
+    no_scroll: bool = BoolField(default=False)
+    model_events: bool = BoolField(default=False)
+
+
+class ClearScrollRequest(Base):
+    scroll_id: str = StringField()
+
+
+class ClearTaskLogRequest(Base):
+    task: str = StringField(required=True)
+    threshold_sec = IntField()
+    allow_locked = BoolField(default=False)
+    exclude_metrics = ListField(items_types=[str])
+    include_metrics = ListField(items_types=[str])

apiserver/apimodels/login.py

@@ -0,0 +1,35 @@
+from jsonmodels.fields import StringField, BoolField, EmbeddedField, ListField
+from jsonmodels.models import Base
+
+from apiserver.apimodels import DictField, callable_default
+
+
+class GetSupportedModesRequest(Base):
+    pass
+    # state = StringField(help_text="ASCII base64 encoded application state")
+    # callback_url_prefix = StringField()
+
+
+class BasicGuestMode(Base):
+    enabled = BoolField(default=False)
+    name = StringField()
+    username = StringField()
+    password = StringField()
+
+
+class BasicMode(Base):
+    enabled = BoolField(default=False)
+    guest = callable_default(EmbeddedField)(BasicGuestMode, default=BasicGuestMode)
+
+
+class ServerErrors(Base):
+    missed_es_upgrade = BoolField(default=False)
+    es_connection_error = BoolField(default=False)
+
+
+class GetSupportedModesResponse(Base):
+    basic = EmbeddedField(BasicMode)
+    server_errors = EmbeddedField(ServerErrors)
+    sso = DictField([str, type(None)])
+    sso_providers = ListField([dict])
+    authenticated = BoolField(default=False)

apiserver/apimodels/metadata.py

@@ -0,0 +1,24 @@
+from typing import Sequence
+
+from jsonmodels import validators
+from jsonmodels.fields import StringField, BoolField
+from jsonmodels.models import Base
+
+from apiserver.apimodels import ListField
+
+
+class MetadataItem(Base):
+    key = StringField(required=True)
+    type = StringField(required=True)
+    value = StringField(required=True)
+
+
+class DeleteMetadata(Base):
+    keys: Sequence[str] = ListField(str, validators=validators.Length(minimum_value=1))
+
+
+class AddOrUpdateMetadata(Base):
+    metadata: Sequence[MetadataItem] = ListField(
+        [MetadataItem], validators=validators.Length(minimum_value=1)
+    )
+    replace_metadata = BoolField(default=False)

apiserver/apimodels/models.py

@@ -0,0 +1,99 @@
+from jsonmodels import models, fields
+from six import string_types
+
+from apiserver.apimodels import ListField, DictField
+from apiserver.apimodels.base import UpdateResponse
+from apiserver.apimodels.batch import BatchRequest
+from apiserver.apimodels.metadata import (
+    MetadataItem,
+    DeleteMetadata,
+    AddOrUpdateMetadata,
+)
+
+
+class GetFrameworksRequest(models.Base):
+    projects = fields.ListField(items_types=[str])
+
+
+class CreateModelRequest(models.Base):
+    name = fields.StringField(required=True)
+    uri = fields.StringField(required=True)
+    labels = DictField(value_types=string_types + (int,))
+    tags = ListField(items_types=string_types)
+    system_tags = ListField(items_types=string_types)
+    comment = fields.StringField()
+    public = fields.BoolField(default=False)
+    project = fields.StringField()
+    parent = fields.StringField()
+    framework = fields.StringField()
+    design = DictField()
+    ready = fields.BoolField(default=True)
+    ui_cache = DictField()
+    task = fields.StringField()
+    metadata = DictField(value_types=[MetadataItem])
+
+
+class CreateModelResponse(models.Base):
+    id = fields.StringField(required=True)
+    created = fields.BoolField(required=True)
+
+
+class ModelRequest(models.Base):
+    model = fields.StringField(required=True)
+
+
+class TaskRequest(models.Base):
+    task = fields.StringField(required=True)
+
+
+class UpdateForTaskRequest(TaskRequest):
+    uri = fields.StringField()
+    iteration = fields.IntField()
+    override_model_id = fields.StringField()
+
+
+class UpdateModelRequest(ModelRequest):
+    task = fields.StringField()
+    iteration = fields.IntField()
+
+
+class DeleteModelRequest(ModelRequest):
+    force = fields.BoolField(default=False)
+    delete_external_artifacts = fields.BoolField(default=True)
+
+
+class ModelsDeleteManyRequest(BatchRequest):
+    force = fields.BoolField(default=False)
+    delete_external_artifacts = fields.BoolField(default=True)
+
+
+class PublishModelRequest(ModelRequest):
+    force_publish_task = fields.BoolField(default=False)
+    publish_task = fields.BoolField(default=True)
+
+
+class ModelTaskPublishResponse(models.Base):
+    id = fields.StringField(required=True)
+    data = fields.EmbeddedField(UpdateResponse)
+
+
+class PublishModelResponse(UpdateResponse):
+    published_task = fields.EmbeddedField(ModelTaskPublishResponse)
+
+
+class ModelsPublishManyRequest(BatchRequest):
+    force_publish_task = fields.BoolField(default=False)
+    publish_task = fields.BoolField(default=True)
+
+
+class DeleteMetadataRequest(DeleteMetadata):
+    model = fields.StringField(required=True)
+
+
+class AddOrUpdateMetadataRequest(AddOrUpdateMetadata):
+    model = fields.StringField(required=True)
+
+
+class ModelsGetRequest(models.Base):
+    include_stats = fields.BoolField(default=False)
+    allow_public = fields.BoolField(default=True)

apiserver/apimodels/organization.py

@@ -0,0 +1,62 @@
+from enum import auto
+from typing import Sequence
+
+from jsonmodels import fields, models
+from jsonmodels.validators import Length
+
+from apiserver.apimodels import DictField, ActualEnumField, ScalarField
+from apiserver.utilities.stringenum import StringEnum
+
+
+class Filter(models.Base):
+    tags = fields.ListField([str])
+    system_tags = fields.ListField([str])
+
+
+class TagsRequest(models.Base):
+    include_system = fields.BoolField(default=False)
+    filter = fields.EmbeddedField(Filter)
+
+
+class EntitiesCountRequest(models.Base):
+    projects = DictField()
+    tasks = DictField()
+    models = DictField()
+    pipelines = DictField()
+    datasets = DictField()
+    reports = DictField()
+    active_users = fields.ListField(str)
+    search_hidden = fields.BoolField(default=False)
+    allow_public = fields.BoolField(default=True)
+
+
+class EntityType(StringEnum):
+    task = auto()
+    model = auto()
+
+
+class ValueMapping(models.Base):
+    key = ScalarField(nullable=True)
+    value = ScalarField(nullable=True)
+
+
+class FieldMapping(models.Base):
+    field = fields.StringField(required=True)
+    name = fields.StringField()
+    values: Sequence[ValueMapping] = fields.ListField(items_types=[ValueMapping])
+
+
+class PrepareDownloadForGetAllRequest(models.Base):
+    entity_type = ActualEnumField(EntityType)
+    allow_public = fields.BoolField(default=True)
+    search_hidden = fields.BoolField(default=False)
+    only_fields = fields.ListField(
+        items_types=[str], validators=[Length(1)], required=True
+    )
+    field_mappings: Sequence[FieldMapping] = fields.ListField(
+        items_types=[FieldMapping], validators=[Length(1)], required=True
+    )
+
+
+class DownloadForGetAllRequest(models.Base):
+    prepare_id = fields.StringField(required=True)

apiserver/apimodels/pipelines.py

@@ -0,0 +1,21 @@
+from jsonmodels import models, fields
+from jsonmodels.validators import Length
+
+from apiserver.apimodels import ListField
+
+
+class Arg(models.Base):
+    name = fields.StringField(required=True)
+    value = fields.StringField(required=True)
+
+
+class DeleteRunsRequest(models.Base):
+    project = fields.StringField(required=True)
+    ids = ListField([str], required=True, validators=[Length(1)])
+
+
+class StartPipelineRequest(models.Base):
+    task = fields.StringField(required=True)
+    queue = fields.StringField(required=True)
+    args = ListField(Arg)
+    verify_watched_queue = fields.BoolField(default=False)

apiserver/apimodels/projects.py

@@ -0,0 +1,103 @@
+from enum import Enum, auto
+
+from jsonmodels import models, fields
+
+from apiserver.apimodels import ListField, ActualEnumField, DictField
+from apiserver.apimodels.organization import TagsRequest
+from apiserver.database.model import EntityVisibility
+from apiserver.utilities.stringenum import StringEnum
+
+
+class ProjectRequest(models.Base):
+    project = fields.StringField(required=True)
+
+
+class MergeRequest(ProjectRequest):
+    destination_project = fields.StringField()
+
+
+class MoveRequest(ProjectRequest):
+    new_location = fields.StringField()
+
+
+class DeleteRequest(ProjectRequest):
+    force = fields.BoolField(default=False)
+    delete_contents = fields.BoolField(default=False)
+    delete_external_artifacts = fields.BoolField(default=True)
+
+
+class ProjectOrNoneRequest(models.Base):
+    project = fields.StringField()
+    include_subprojects = fields.BoolField(default=True)
+
+
+class GetUniqueMetricsRequest(ProjectOrNoneRequest):
+    model_metrics = fields.BoolField(default=False)
+    ids = fields.ListField(str)
+
+
+class GetParamsRequest(ProjectOrNoneRequest):
+    page = fields.IntField(default=0)
+    page_size = fields.IntField(default=500)
+
+
+class ProjectTagsRequest(TagsRequest):
+    projects = ListField(str)
+
+
+class MultiProjectRequest(models.Base):
+    projects = fields.ListField(items_types=[str, type(None)])
+    include_subprojects = fields.BoolField(default=True)
+
+
+class ProjectTaskParentsRequest(MultiProjectRequest):
+    tasks_state = ActualEnumField(EntityVisibility)
+    task_name = fields.StringField()
+
+
+class EntityTypeEnum(StringEnum):
+    task = auto()
+    model = auto()
+
+
+class ProjectUserNamesRequest(MultiProjectRequest):
+    entity = ActualEnumField(EntityTypeEnum, default=EntityTypeEnum.task)
+
+
+class MultiProjectPagedRequest(MultiProjectRequest):
+    allow_public = fields.BoolField(default=True)
+    page = fields.IntField(default=0)
+    page_size = fields.IntField(default=500)
+
+
+class ProjectHyperparamValuesRequest(MultiProjectPagedRequest):
+    section = fields.StringField(required=True)
+    name = fields.StringField(required=True)
+    pattern = fields.StringField()
+
+
+class ProjectModelMetadataValuesRequest(MultiProjectPagedRequest):
+    key = fields.StringField(required=True)
+
+
+class ProjectChildrenType(Enum):
+    pipeline = "pipeline"
+    report = "report"
+    dataset = "dataset"
+
+
+class ProjectsGetRequest(models.Base):
+    include_dataset_stats = fields.BoolField(default=False)
+    include_stats = fields.BoolField(default=False)
+    include_stats_filter = DictField()
+    stats_with_children = fields.BoolField(default=True)
+    stats_for_state = ActualEnumField(EntityVisibility, default=EntityVisibility.active)
+    non_public = fields.BoolField(default=False)  # legacy, use allow_public instead
+    active_users = fields.ListField(str)
+    check_own_contents = fields.BoolField(default=False)
+    shallow_search = fields.BoolField(default=False)
+    search_hidden = fields.BoolField(default=False)
+    allow_public = fields.BoolField(default=True)
+    children_type = ActualEnumField(ProjectChildrenType)
+    children_tags = fields.ListField(str)
+    children_tags_filter = DictField()

apiserver/apimodels/queues.py

@@ -2,7 +2,12 @@ from jsonmodels import validators
 from jsonmodels.fields import StringField, IntField, BoolField, FloatField
 from jsonmodels.models import Base
 
-from apimodels import ListField
+from apiserver.apimodels import ListField, DictField
+from apiserver.apimodels.metadata import (
+    MetadataItem,
+    DeleteMetadata,
+    AddOrUpdateMetadata,
+)
 
 
 class GetDefaultResp(Base):
@@ -12,28 +17,55 @@ class GetDefaultResp(Base):
 
 class CreateRequest(Base):
     name = StringField(required=True)
+    display_name = StringField()
     tags = ListField(items_types=[str])
     system_tags = ListField(items_types=[str])
+    metadata = DictField(value_types=[MetadataItem])
 
 
 class QueueRequest(Base):
     queue = StringField(required=True)
 
 
+class GetByIdRequest(QueueRequest):
+    max_task_entries = IntField()
+
+
+class GetAllRequest(Base):
+    max_task_entries = IntField()
+    search_hidden = BoolField(default=False)
+
+
+class GetNextTaskRequest(QueueRequest):
+    queue = StringField(required=True)
+    get_task_info = BoolField(default=False)
+    task = StringField()
+
+
 class DeleteRequest(QueueRequest):
     force = BoolField(default=False)
 
 
 class UpdateRequest(QueueRequest):
     name = StringField()
+    display_name = StringField()
     tags = ListField(items_types=[str])
     system_tags = ListField(items_types=[str])
+    metadata = DictField(value_types=[MetadataItem])
 
 
 class TaskRequest(QueueRequest):
     task = StringField(required=True)
 
 
+class RemoveTaskRequest(TaskRequest):
+    update_task_status = BoolField(default=False)
+
+
+class AddTaskRequest(TaskRequest):
+    update_execution_queue = BoolField(default=True)
+
+
 class MoveTaskRequest(TaskRequest):
     count = IntField(default=1)
 
@@ -47,6 +79,7 @@ class GetMetricsRequest(Base):
     from_date = FloatField(required=True, validators=validators.Min(0))
     to_date = FloatField(required=True, validators=validators.Min(0))
     interval = IntField(required=True, validators=validators.Min(1))
+    refresh = BoolField(default=False)
 
 
 class QueueMetrics(Base):
@@ -58,3 +91,11 @@ class QueueMetrics(Base):
 
 class GetMetricsResponse(Base):
     queues = ListField(QueueMetrics)
+
+
+class DeleteMetadataRequest(DeleteMetadata):
+    queue = StringField(required=True)
+
+
+class AddOrUpdateMetadataRequest(AddOrUpdateMetadata):
+    queue = StringField(required=True)

apiserver/apimodels/reports.py

@@ -0,0 +1,84 @@
+from typing import Sequence
+
+from jsonmodels import validators
+from jsonmodels.fields import StringField, ListField, BoolField, EmbeddedField, IntField
+from jsonmodels.models import Base
+from jsonmodels.validators import Length
+
+from apiserver.apimodels.events import MetricVariants, HistogramRequestBase
+
+
+class UpdateReportRequest(Base):
+    task = StringField(required=True)
+    name = StringField(nullable=True, validators=Length(minimum_value=3))
+    tags = ListField(items_types=[str])
+    comment = StringField()
+    report = StringField()
+    report_assets = ListField(items_types=[str])
+
+
+class CreateReportRequest(Base):
+    name = StringField(required=True, validators=Length(minimum_value=3))
+    tags = ListField(items_types=[str])
+    comment = StringField()
+    report = StringField()
+    project = StringField()
+    report_assets = ListField(items_types=[str])
+
+
+class PublishReportRequest(Base):
+    task = StringField(required=True)
+    message = StringField(default="")
+
+
+class ArchiveReportRequest(Base):
+    task = StringField(required=True)
+    message = StringField(default="")
+
+
+class ShareReportRequest(Base):
+    task = StringField(required=True)
+    share = BoolField(default=True)
+
+
+class DeleteReportRequest(Base):
+    task = StringField(required=True)
+    force = BoolField(default=False)
+
+
+class MoveReportRequest(Base):
+    task = StringField(required=True)
+    project = StringField()
+    project_name = StringField()
+
+
+class EventsRequest(Base):
+    iters = IntField(default=1, validators=validators.Min(1))
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+
+
+class PlotEventsRequest(EventsRequest):
+    last_iters_per_task_metric: bool = BoolField(default=True)
+
+
+class ScalarMetricsIterHistogram(HistogramRequestBase):
+    metrics: Sequence[MetricVariants] = ListField(items_types=MetricVariants)
+
+
+class SingleValueMetrics(Base):
+    pass
+
+
+class GetTasksDataRequest(Base):
+    debug_images: EventsRequest = EmbeddedField(EventsRequest)
+    plots: PlotEventsRequest = EmbeddedField(PlotEventsRequest)
+    scalar_metrics_iter_histogram: ScalarMetricsIterHistogram = EmbeddedField(
+        ScalarMetricsIterHistogram
+    )
+    single_value_metrics: SingleValueMetrics = EmbeddedField(SingleValueMetrics)
+    allow_public = BoolField(default=True)
+    model_events: bool = BoolField(default=False)
+
+
+class GetAllRequest(Base):
+    allow_public = BoolField(default=True)

apiserver/apimodels/server.py

@@ -6,6 +6,10 @@ class ReportStatsOptionRequest(Base):
     enabled = BoolField(default=None, nullable=True)
 
 
+class GetConfigRequest(Base):
+    path = StringField()
+
+
 class ReportStatsOptionResponse(Base):
     supported = BoolField(default=True)
     enabled = BoolField()

apiserver/apimodels/serving.py

@@ -0,0 +1,104 @@
+from enum import Enum
+from typing import Sequence
+
+from jsonmodels.models import Base
+from jsonmodels.fields import (
+    StringField,
+    EmbeddedField,
+    DateTimeField,
+    IntField,
+    FloatField,
+    BoolField,
+)
+from jsonmodels import validators
+from jsonmodels.validators import Min
+
+from apiserver.apimodels import ListField, JsonSerializableMixin, SafeStringField
+from apiserver.apimodels import ActualEnumField
+from apiserver.config_repo import config
+from .workers import MachineStats
+
+
+class ReferenceItem(Base):
+    type = StringField(
+        required=True,
+        validators=validators.Enum("app_id", "app_instance", "model", "task", "url"),
+    )
+    value = StringField(required=True)
+
+
+class ServingModel(Base):
+    container_id = StringField(required=True)
+    endpoint_name = StringField(required=True)
+    endpoint_url = StringField()  # can be not existing yet at registration time
+    model_name = StringField(required=True)
+    model_source = StringField()
+    model_version = StringField()
+    preprocess_artifact = StringField()
+    input_type = StringField()
+    input_size = SafeStringField()
+    tags = ListField(str)
+    system_tags = ListField(str)
+    reference: Sequence[ReferenceItem] = ListField(ReferenceItem)
+
+
+class RegisterRequest(ServingModel):
+    timeout = IntField(
+        default=int(
+            config.get("services.serving.default_container_timeout_sec", 10 * 60)
+        ),
+        validators=[Min(1)],
+    )
+    """ registration timeout in seconds (default is 10min) """
+
+
+class UnregisterRequest(Base):
+    container_id = StringField(required=True)
+
+
+class StatusReportRequest(ServingModel):
+    uptime_sec = IntField()
+    requests_num = IntField()
+    requests_min = FloatField()
+    latency_ms = IntField()
+    machine_stats: MachineStats = EmbeddedField(MachineStats)
+
+
+class ServingContainerEntry(StatusReportRequest, JsonSerializableMixin):
+    key = StringField(required=True)
+    company_id = StringField(required=True)
+    ip = StringField()
+    register_time = DateTimeField(required=True)
+    register_timeout = IntField(required=True)
+    last_activity_time = DateTimeField(required=True)
+
+
+class GetEndpointDetailsRequest(Base):
+    endpoint_url = StringField(required=True)
+
+
+class MetricType(Enum):
+    requests = "requests"
+    requests_min = "requests_min"
+    latency_ms = "latency_ms"
+    cpu_count = "cpu_count"
+    gpu_count = "gpu_count"
+    cpu_util = "cpu_util"
+    gpu_util = "gpu_util"
+    ram_total = "ram_total"
+    ram_used = "ram_used"
+    ram_free = "ram_free"
+    gpu_ram_total = "gpu_ram_total"
+    gpu_ram_used = "gpu_ram_used"
+    gpu_ram_free = "gpu_ram_free"
+    network_rx = "network_rx"
+    network_tx = "network_tx"
+
+
+class GetEndpointMetricsHistoryRequest(Base):
+    from_date = FloatField(required=True, validators=Min(0))
+    to_date = FloatField(required=True, validators=Min(0))
+    interval = IntField(required=True, validators=Min(1))
+    endpoint_url = StringField(required=True)
+    metric_type = ActualEnumField(MetricType, default=MetricType.requests)
+    instance_charts = BoolField(default=True)

apiserver/apimodels/storage.py

@@ -0,0 +1,60 @@
+from jsonmodels.fields import StringField, BoolField, ListField, EmbeddedField
+from jsonmodels.models import Base
+from jsonmodels.validators import Enum
+
+
+class AWSBucketSettings(Base):
+    bucket = StringField()
+    subdir = StringField()
+    host = StringField()
+    key = StringField()
+    secret = StringField()
+    token = StringField()
+    multipart = BoolField(default=True)
+    acl = StringField()
+    secure = BoolField(default=True)
+    region = StringField()
+    verify = BoolField(default=True)
+    use_credentials_chain = BoolField(default=False)
+
+
+class AWSSettings(Base):
+    key = StringField()
+    secret = StringField()
+    region = StringField()
+    token = StringField()
+    use_credentials_chain = BoolField(default=False)
+    buckets = ListField(items_types=[AWSBucketSettings])
+
+
+class GoogleBucketSettings(Base):
+    bucket = StringField()
+    subdir = StringField()
+    project = StringField()
+    credentials_json = StringField()
+
+
+class GoogleSettings(Base):
+    project = StringField()
+    credentials_json = StringField()
+    buckets = ListField(items_types=[GoogleBucketSettings])
+
+
+class AzureContainerSettings(Base):
+    account_name = StringField()
+    account_key = StringField()
+    container_name = StringField()
+
+
+class AzureSettings(Base):
+    containers = ListField(items_types=[AzureContainerSettings])
+
+
+class SetSettingsRequest(Base):
+    aws = EmbeddedField(AWSSettings)
+    google = EmbeddedField(GoogleSettings)
+    azure = EmbeddedField(AzureSettings)
+
+
+class ResetSettingsRequest(Base):
+    keys = ListField([str], item_validators=[Enum("aws", "google", "azure")])

apiserver/apimodels/tasks.py

@@ -0,0 +1,357 @@
+from typing import Sequence
+
+from jsonmodels import models
+from jsonmodels.fields import StringField, BoolField, IntField, EmbeddedField
+from jsonmodels.validators import Enum, Length
+
+from apiserver.apimodels import DictField, ListField
+from apiserver.apimodels.base import UpdateResponse
+from apiserver.apimodels.batch import BatchRequest, UpdateBatchItem, BatchResponse
+from apiserver.database.model.task.task import (
+    TaskType,
+    ArtifactModes,
+    DEFAULT_ARTIFACT_MODE,
+    TaskModelTypes,
+)
+from apiserver.database.utils import get_options
+
+
+class ArtifactTypeData(models.Base):
+    preview = StringField()
+    content_type = StringField()
+    data_hash = StringField()
+
+
+class Artifact(models.Base):
+    key = StringField(required=True)
+    type = StringField(required=True)
+    mode = StringField(
+        validators=Enum(*get_options(ArtifactModes)), default=DEFAULT_ARTIFACT_MODE
+    )
+    uri = StringField()
+    hash = StringField()
+    content_size = IntField()
+    timestamp = IntField()
+    type_data = EmbeddedField(ArtifactTypeData)
+    display_data = ListField([list])
+
+
+class StartedResponse(UpdateResponse):
+    started = IntField()
+
+
+class EnqueueResponse(UpdateResponse):
+    queued = IntField()
+    queue_watched = BoolField()
+
+
+class EnqueueBatchItem(UpdateBatchItem):
+    queued: bool = BoolField()
+
+
+class EnqueueManyResponse(BatchResponse):
+    succeeded: Sequence[EnqueueBatchItem] = ListField(EnqueueBatchItem)
+    queue_watched = BoolField()
+
+
+class DequeueResponse(UpdateResponse):
+    dequeued = IntField()
+
+
+class DequeueBatchItem(UpdateBatchItem):
+    dequeued: bool = BoolField()
+
+
+class DequeueManyResponse(BatchResponse):
+    succeeded: Sequence[DequeueBatchItem] = ListField(DequeueBatchItem)
+
+
+class ResetResponse(UpdateResponse):
+    dequeued = DictField()
+    events = DictField()
+    deleted_models = IntField()
+    urls = DictField()
+
+
+class ResetBatchItem(UpdateBatchItem):
+    dequeued: bool = BoolField()
+    deleted_models = IntField()
+    urls = DictField()
+
+
+class ResetManyResponse(BatchResponse):
+    succeeded: Sequence[ResetBatchItem] = ListField(ResetBatchItem)
+
+
+class TaskRequest(models.Base):
+    task = StringField(required=True)
+
+
+class TaskUpdateRequest(TaskRequest):
+    force = BoolField(default=False)
+
+
+class UpdateRequest(TaskUpdateRequest):
+    status_reason = StringField(default="")
+    status_message = StringField(default="")
+
+
+class DequeueRequest(UpdateRequest):
+    remove_from_all_queues = BoolField(default=False)
+    new_status = StringField()
+
+
+class StopRequest(UpdateRequest):
+    include_pipeline_steps = BoolField(default=False)
+
+
+class EnqueueRequest(UpdateRequest):
+    queue = StringField()
+    queue_name = StringField()
+    verify_watched_queue = BoolField(default=False)
+    update_execution_queue = BoolField(default=True)
+
+
+class DeleteRequest(UpdateRequest):
+    move_to_trash = BoolField(default=True)
+    return_file_urls = BoolField(default=False)
+    delete_output_models = BoolField(default=True)
+    delete_external_artifacts = BoolField(default=True)
+    include_pipeline_steps = BoolField(default=False)
+
+
+class SetRequirementsRequest(TaskRequest):
+    requirements = DictField(required=True)
+
+
+class CompletedRequest(UpdateRequest):
+    publish = BoolField(default=False)
+
+
+class CompletedResponse(UpdateResponse):
+    published = IntField(default=0)
+
+
+class PublishRequest(UpdateRequest):
+    publish_model = BoolField(default=True)
+
+
+class TaskData(models.Base):
+    """
+    This is a partial description of task can be updated incrementally
+    """
+
+
+class CreateRequest(TaskData):
+    name = StringField(required=True)
+    type = StringField(required=True, validators=Enum(*get_options(TaskType)))
+
+
+class PingRequest(TaskRequest):
+    pass
+
+
+class GetTypesRequest(models.Base):
+    projects = ListField(items_types=[str])
+
+
+class TaskInputModel(models.Base):
+    name = StringField()
+    model = StringField()
+
+
+class CloneRequest(TaskRequest):
+    new_task_name = StringField()
+    new_task_comment = StringField()
+    new_task_tags = ListField([str])
+    new_task_system_tags = ListField([str])
+    new_task_parent = StringField()
+    new_task_project = StringField()
+    new_task_hyperparams = DictField()
+    new_task_configuration = DictField()
+    new_task_container = DictField()
+    new_task_input_models = ListField([TaskInputModel])
+    execution_overrides = DictField()
+    validate_references = BoolField(default=False)
+    new_project_name = StringField()
+
+
+class AddOrUpdateArtifactsRequest(TaskUpdateRequest):
+    artifacts = ListField([Artifact], validators=Length(minimum_value=1))
+
+
+class ArtifactId(models.Base):
+    key = StringField(required=True)
+    mode = StringField(
+        validators=Enum(*get_options(ArtifactModes)), default=DEFAULT_ARTIFACT_MODE
+    )
+
+
+class DeleteArtifactsRequest(TaskUpdateRequest):
+    artifacts = ListField([ArtifactId], validators=Length(minimum_value=1))
+
+
+class ResetRequest(UpdateRequest):
+    clear_all = BoolField(default=False)
+    return_file_urls = BoolField(default=False)
+    delete_output_models = BoolField(default=True)
+    delete_external_artifacts = BoolField(default=True)
+
+
+class MultiTaskRequest(models.Base):
+    tasks = ListField([str], validators=Length(minimum_value=1))
+
+
+class GetHyperParamsRequest(MultiTaskRequest):
+    pass
+
+
+class HyperParamItem(models.Base):
+    section = StringField(required=True, validators=Length(minimum_value=1))
+    name = StringField(required=True, validators=Length(minimum_value=1))
+    value = StringField(required=True)
+    type = StringField()
+    description = StringField()
+
+
+class ReplaceHyperparams(object):
+    none = "none"
+    section = "section"
+    all = "all"
+
+
+class EditHyperParamsRequest(TaskUpdateRequest):
+    hyperparams: Sequence[HyperParamItem] = ListField(
+        [HyperParamItem], validators=Length(minimum_value=1)
+    )
+    replace_hyperparams = StringField(
+        validators=Enum(*get_options(ReplaceHyperparams)),
+        default=ReplaceHyperparams.none,
+    )
+
+
+class HyperParamKey(models.Base):
+    section = StringField(required=True, validators=Length(minimum_value=1))
+    name = StringField(nullable=True)
+
+
+class DeleteHyperParamsRequest(TaskUpdateRequest):
+    hyperparams: Sequence[HyperParamKey] = ListField(
+        [HyperParamKey], validators=Length(minimum_value=1)
+    )
+
+
+class GetConfigurationsRequest(MultiTaskRequest):
+    names = ListField([str])
+
+
+class GetConfigurationNamesRequest(MultiTaskRequest):
+    skip_empty = BoolField(default=True)
+
+
+class Configuration(models.Base):
+    name = StringField(required=True, validators=Length(minimum_value=1))
+    value = StringField(required=True)
+    type = StringField()
+    description = StringField()
+
+
+class EditConfigurationRequest(TaskUpdateRequest):
+    configuration: Sequence[Configuration] = ListField(
+        [Configuration], validators=Length(minimum_value=1)
+    )
+    replace_configuration = BoolField(default=False)
+
+
+class DeleteConfigurationRequest(TaskUpdateRequest):
+    configuration: Sequence[str] = ListField([str], validators=Length(minimum_value=1))
+
+
+class ArchiveRequest(MultiTaskRequest):
+    status_reason = StringField(default="")
+    status_message = StringField(default="")
+    include_pipeline_steps = BoolField(default=False)
+
+
+class ArchiveResponse(models.Base):
+    archived = IntField()
+
+
+class TaskBatchRequest(BatchRequest):
+    status_reason = StringField(default="")
+    status_message = StringField(default="")
+
+
+class ArchiveManyRequest(TaskBatchRequest):
+    include_pipeline_steps = BoolField(default=False)
+
+
+class UnarchiveManyRequest(TaskBatchRequest):
+    include_pipeline_steps = BoolField(default=False)
+
+
+class StopManyRequest(TaskBatchRequest):
+    force = BoolField(default=False)
+    include_pipeline_steps = BoolField(default=False)
+
+
+class DequeueManyRequest(TaskBatchRequest):
+    remove_from_all_queues = BoolField(default=False)
+    new_status = StringField()
+
+
+class EnqueueManyRequest(TaskBatchRequest):
+    queue = StringField()
+    queue_name = StringField()
+    validate_tasks = BoolField(default=False)
+    verify_watched_queue = BoolField(default=False)
+
+
+class DeleteManyRequest(TaskBatchRequest):
+    move_to_trash = BoolField(default=True)
+    return_file_urls = BoolField(default=False)
+    delete_output_models = BoolField(default=True)
+    force = BoolField(default=False)
+    delete_external_artifacts = BoolField(default=True)
+    include_pipeline_steps = BoolField(default=False)
+
+
+class ResetManyRequest(TaskBatchRequest):
+    clear_all = BoolField(default=False)
+    return_file_urls = BoolField(default=False)
+    delete_output_models = BoolField(default=True)
+    force = BoolField(default=False)
+    delete_external_artifacts = BoolField(default=True)
+
+
+class PublishManyRequest(TaskBatchRequest):
+    publish_model = BoolField(default=True)
+    force = BoolField(default=False)
+
+
+class AddUpdateModelRequest(TaskRequest):
+    name = StringField(required=True)
+    model = StringField(required=True)
+    type = StringField(required=True, validators=Enum(*get_options(TaskModelTypes)))
+    iteration = IntField()
+
+
+class ModelItemKey(models.Base):
+    name = StringField(required=True)
+    type = StringField(required=True, validators=Enum(*get_options(TaskModelTypes)))
+
+
+class DeleteModelsRequest(TaskRequest):
+    models: Sequence[ModelItemKey] = ListField(
+        [ModelItemKey], validators=Length(minimum_value=1)
+    )
+
+
+class GetAllReq(models.Base):
+    allow_public = BoolField(default=True)
+    search_hidden = BoolField(default=False)
+
+
+class UpdateTagsRequest(BatchRequest):
+    add_tags = ListField([str])
+    remove_tags = ListField([str])

apiserver/apimodels/users.py

@@ -1,7 +1,11 @@
 from jsonmodels.fields import StringField
 from jsonmodels.models import Base
 
-from apimodels import DictField
+from apiserver.apimodels import DictField
+
+
+class UserRequest(Base):
+    user = StringField(required=True)
 
 
 class CreateRequest(Base):

apiserver/apimodels/workers.py

@@ -12,20 +12,21 @@ from jsonmodels.fields import (
 )
 from jsonmodels.models import Base
 
-from apimodels import make_default, ListField, EnumField, JsonSerializableMixin
-
-DEFAULT_TIMEOUT = 10 * 60
+from apiserver.apimodels import ListField, EnumField, JsonSerializableMixin, ActualEnumField
+from apiserver.config_repo import config
 
 
 class WorkerRequest(Base):
     worker = StringField(required=True)
     tags = ListField(str)
+    system_tags = ListField(str)
 
 
 class RegisterRequest(WorkerRequest):
-    timeout = make_default(
-        IntField, DEFAULT_TIMEOUT
-    )()  # registration timeout in seconds (default is 10min)
+    timeout = IntField(
+        default=int(config.get("services.workers.default_worker_timeout_sec", 10 * 60))
+    )
+    """ registration timeout in seconds (default is 10min) """
     queues = ListField(six.string_types)  # list of queues this worker listens to
 
 
@@ -76,6 +77,7 @@ class WorkerEntry(Base, JsonSerializableMixin):
     last_activity_time = DateTimeField(required=True)
     last_report_time = DateTimeField()
     tags = ListField(str)
+    system_tags = ListField(str)
 
 
 class CurrentTaskEntry(IdNameEntry):
@@ -84,6 +86,7 @@ class CurrentTaskEntry(IdNameEntry):
 
 
 class QueueEntry(IdNameEntry):
+    display_name = StringField()
     next_task = EmbeddedField(IdNameEntry)
     num_tasks = IntField()
 
@@ -96,12 +99,19 @@ class WorkerResponseEntry(WorkerEntry):
 
 class GetAllRequest(Base):
     last_seen = IntField(default=3600)
+    tags = ListField(str)
+    system_tags = ListField(str)
+    worker_pattern = StringField()
 
 
 class GetAllResponse(Base):
     workers = ListField(WorkerResponseEntry)
 
 
+class GetCountRequest(GetAllRequest):
+    last_seen = IntField(default=0)
+
+
 class StatsBase(Base):
     worker_ids = ListField(str)
 
@@ -120,7 +130,7 @@ class AggregationType(Enum):
 
 class StatItem(Base):
     key = StringField(required=True)
-    aggregation = EnumField(AggregationType, default=AggregationType.avg)
+    aggregation = ActualEnumField(AggregationType, default=AggregationType.avg)
 
 
 class GetStatsRequest(StatsReportBase):
@@ -128,17 +138,24 @@ class GetStatsRequest(StatsReportBase):
         StatItem, required=True, validators=validators.Length(minimum_value=1)
     )
     split_by_variant = BoolField(default=False)
+    split_by_resource = BoolField(default=False)
+
+
+class MetricResourceSeries(Base):
+    name = StringField()
+    values = ListField(float)
 
 
 class AggregationStats(Base):
     aggregation = EnumField(AggregationType)
+    dates = ListField(int)
     values = ListField(float)
+    resource_series = ListField(MetricResourceSeries)
 
 
 class MetricStats(Base):
     metric = StringField()
     variant = StringField()
-    dates = ListField(int)
     stats = ListField(AggregationStats)
 
 

apiserver/bll/auth/__init__.py

@@ -1,17 +1,21 @@
 from datetime import datetime
 
-import database
-from apierrors import errors
-from apimodels.auth import GetTokenResponse, CreateUserRequest, Credentials as CredModel
-from apimodels.users import CreateRequest as Users_CreateRequest
-from bll.user import UserBLL
-from config import config
-from config.info import get_version, get_build_number
-from database.errors import translate_errors_context
-from database.model.auth import User, Role, Credentials
-from database.model.company import Company
-from service_repo import APICall, ServiceRepo
-from service_repo.auth import Identity, Token, get_client_id, get_secret_key
+from apiserver import database
+from apiserver.apierrors import errors
+from apiserver.apimodels.auth import (
+    GetTokenResponse,
+    CreateUserRequest,
+    Credentials as CredModel,
+)
+from apiserver.apimodels.users import CreateRequest as Users_CreateRequest
+from apiserver.bll.user import UserBLL
+from apiserver.config_repo import config
+from apiserver.config.info import get_version, get_build_number
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.auth import User, Role, Credentials
+from apiserver.database.model.company import Company
+from apiserver.service_repo import APICall, ServiceRepo
+from apiserver.service_repo.auth import Identity, Token, get_client_id, get_secret_key
 
 log = config.logger("AuthBLL")
 
@@ -57,9 +61,10 @@ class AuthBLL:
                 api_version=str(ServiceRepo.max_endpoint_version()),
                 server_version=str(get_version()),
                 server_build=str(get_build_number()),
+                feature_set="basic",
             )
 
-            return GetTokenResponse(token=token.decode("ascii"))
+            return GetTokenResponse(token=token)
 
     @staticmethod
     def create_user(request: CreateUserRequest, call: APICall = None) -> str:
@@ -144,7 +149,7 @@ class AuthBLL:
 
     @classmethod
     def create_credentials(
-        cls, user_id: str, company_id: str, role: str = None
+        cls, user_id: str, company_id: str, role: str = None, label: str = None,
     ) -> CredModel:
 
         with translate_errors_context():
@@ -153,9 +158,11 @@ class AuthBLL:
             if not user:
                 raise errors.bad_request.InvalidUserId(**query)
 
-            cred = CredModel(access_key=get_client_id(), secret_key=get_secret_key())
+            cred = CredModel(
+                access_key=get_client_id(), secret_key=get_secret_key(), label=label
+            )
             user.credentials.append(
-                Credentials(key=cred.access_key, secret=cred.secret_key)
+                Credentials(key=cred.access_key, secret=cred.secret_key, label=label)
             )
             user.save()
 

apiserver/bll/event/event_common.py

@@ -0,0 +1,166 @@
+import base64
+import zlib
+from enum import Enum
+from typing import Union, Sequence, Mapping, Tuple
+
+from boltons.typeutils import classproperty
+from elasticsearch import Elasticsearch
+
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.task.task import Task
+from apiserver.utilities.dicts import nested_get
+
+
+class EventType(Enum):
+    metrics_scalar = "training_stats_scalar"
+    metrics_vector = "training_stats_vector"
+    metrics_image = "training_debug_image"
+    metrics_plot = "plot"
+    task_log = "log"
+    all = "*"
+
+
+SINGLE_SCALAR_ITERATION = -(2 ** 31)
+MetricVariants = Mapping[str, Sequence[str]]
+TaskCompanies = Mapping[str, Sequence[Task]]
+
+
+class EventSettings:
+    _max_es_allowed_aggregation_buckets = 10000
+
+    @classproperty
+    def max_workers(self):
+        return config.get("services.events.events_retrieval.max_metrics_concurrency", 4)
+
+    @classproperty
+    def state_expiration_sec(self):
+        return config.get(
+            f"services.events.events_retrieval.state_expiration_sec", 3600
+        )
+
+    @classproperty
+    def max_es_buckets(self):
+        percentage = (
+            min(
+                100,
+                config.get(
+                    "services.events.events_retrieval.dynamic_metrics_count_threshold",
+                    80,
+                ),
+            )
+            / 100
+        )
+        return int(self._max_es_allowed_aggregation_buckets * percentage)
+
+
+def get_index_name(company_id: Union[str, Sequence[str]], event_type: str):
+    event_type = event_type.lower().replace(" ", "_")
+    if isinstance(company_id, str):
+        company_id = [company_id]
+
+    return ",".join(f"events-{event_type}-{(c_id or '').lower()}" for c_id in company_id)
+
+
+def check_empty_data(es: Elasticsearch, company_id: str, event_type: EventType) -> bool:
+    es_index = get_index_name(company_id, event_type.value)
+    if not es.indices.exists(index=es_index):
+        return True
+    return False
+
+
+def search_company_events(
+    es: Elasticsearch,
+    company_id: Union[str, Sequence[str]],
+    event_type: EventType,
+    body: dict,
+    **kwargs,
+) -> dict:
+    es_index = get_index_name(company_id, event_type.value)
+    return es.search(index=es_index, body=body, **kwargs)
+
+
+def delete_company_events(
+    es: Elasticsearch, company_id: str, event_type: EventType, body: dict, **kwargs
+) -> dict:
+    es_index = get_index_name(company_id, event_type.value)
+    return es.delete_by_query(index=es_index, body=body, conflicts="proceed", **kwargs)
+
+
+def count_company_events(
+    es: Elasticsearch, company_id: str, event_type: EventType, body: dict, **kwargs
+) -> dict:
+    es_index = get_index_name(company_id, event_type.value)
+    return es.count(index=es_index, body=body, **kwargs)
+
+
+def get_max_metric_and_variant_counts(
+    es: Elasticsearch,
+    company_id: Union[str, Sequence[str]],
+    event_type: EventType,
+    query: dict,
+    **kwargs,
+) -> Tuple[int, int]:
+    dynamic = config.get(
+        "services.events.events_retrieval.dynamic_metrics_count", False
+    )
+    max_metrics_count = config.get(
+        "services.events.events_retrieval.max_metrics_count", 100
+    )
+    max_variants_count = config.get(
+        "services.events.events_retrieval.max_variants_count", 100
+    )
+    if not dynamic:
+        return max_metrics_count, max_variants_count
+
+    es_req: dict = {
+        "size": 0,
+        "query": query,
+        "aggs": {"metrics_count": {"cardinality": {"field": "metric"}}},
+    }
+    with translate_errors_context():
+        es_res = search_company_events(
+            es, company_id=company_id, event_type=event_type, body=es_req, **kwargs,
+        )
+
+    metrics_count = nested_get(
+        es_res, ("aggregations", "metrics_count", "value"), max_metrics_count
+    )
+    if not metrics_count:
+        return max_metrics_count, max_variants_count
+
+    return metrics_count, int(EventSettings.max_es_buckets / metrics_count)
+
+
+def get_metric_variants_condition(metric_variants: MetricVariants,) -> Sequence:
+    conditions = [
+        {
+            "bool": {
+                "must": [
+                    {"term": {"metric": metric}},
+                    {"terms": {"variant": variants}},
+                ]
+            }
+        }
+        if variants
+        else {"term": {"metric": metric}}
+        for metric, variants in metric_variants.items()
+    ]
+
+    return {"bool": {"should": conditions}}
+
+
+class PlotFields:
+    valid_plot = "valid_plot"
+    plot_len = "plot_len"
+    plot_str = "plot_str"
+    plot_data = "plot_data"
+    source_urls = "source_urls"
+
+
+def uncompress_plot(event: dict):
+    plot_data = event.pop(PlotFields.plot_data, None)
+    if plot_data and event.get(PlotFields.plot_str) is None:
+        event[PlotFields.plot_str] = zlib.decompress(
+            base64.b64decode(plot_data)
+        ).decode()

apiserver/bll/event/event_metrics.py

@@ -0,0 +1,618 @@
+import itertools
+import math
+from collections import defaultdict
+from concurrent.futures.thread import ThreadPoolExecutor
+from functools import partial
+from operator import itemgetter
+from typing import Sequence, Tuple, Mapping
+
+from boltons.iterutils import bucketize
+from elasticsearch import Elasticsearch
+
+from apiserver.bll.event.event_common import (
+    EventType,
+    EventSettings,
+    search_company_events,
+    check_empty_data,
+    MetricVariants,
+    get_metric_variants_condition,
+    get_max_metric_and_variant_counts,
+    SINGLE_SCALAR_ITERATION,
+    TaskCompanies,
+)
+from apiserver.bll.event.scalar_key import ScalarKey, ScalarKeyEnum
+from apiserver.bll.query import Builder as QueryBuilder
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task
+from apiserver.utilities.dicts import nested_get
+
+log = config.logger(__file__)
+
+
+class EventMetrics:
+    MAX_AGGS_ELEMENTS_COUNT = 50
+    MAX_SAMPLE_BUCKETS = 6000
+
+    def __init__(self, es: Elasticsearch):
+        self.es = es
+
+    def get_scalar_metrics_average_per_iter(
+        self,
+        company_id: str,
+        task_id: str,
+        samples: int,
+        key: ScalarKeyEnum,
+        metric_variants: MetricVariants = None,
+        model_events: bool = False,
+    ) -> dict:
+        """
+        Get scalar metric histogram per metric and variant
+        The amount of points in each histogram should not exceed
+        the requested samples
+        """
+        event_type = EventType.metrics_scalar
+        if check_empty_data(self.es, company_id=company_id, event_type=event_type):
+            return {}
+
+        return self._get_scalar_average_per_iter_core(
+            task_id=task_id,
+            company_id=company_id,
+            event_type=event_type,
+            samples=samples,
+            key=ScalarKey.resolve(key),
+            metric_variants=metric_variants,
+            model_events=model_events,
+        )
+
+    def _get_scalar_average_per_iter_core(
+        self,
+        task_id: str,
+        company_id: str,
+        event_type: EventType,
+        samples: int,
+        key: ScalarKey,
+        run_parallel: bool = True,
+        metric_variants: MetricVariants = None,
+        model_events: bool = False,
+    ) -> dict:
+        intervals = self._get_task_metric_intervals(
+            company_id=company_id,
+            event_type=event_type,
+            task_id=task_id,
+            samples=samples,
+            field=key.field,
+            metric_variants=metric_variants,
+        )
+        if not intervals:
+            return {}
+        interval_groups = self._group_task_metric_intervals(intervals)
+
+        get_scalar_average = partial(
+            self._get_scalar_average,
+            task_id=task_id,
+            company_id=company_id,
+            event_type=event_type,
+            key=key,
+        )
+        if run_parallel:
+            with ThreadPoolExecutor(max_workers=EventSettings.max_workers) as pool:
+                metrics = itertools.chain.from_iterable(
+                    pool.map(get_scalar_average, interval_groups)
+                )
+        else:
+            metrics = itertools.chain.from_iterable(
+                get_scalar_average(group) for group in interval_groups
+            )
+
+        ret = defaultdict(dict)
+        if not metrics:
+            return ret
+
+        last_metrics = {}
+        cls_ = Model if model_events else Task
+        task = cls_.objects(id=task_id).only("last_metrics").first()
+        if task and task.last_metrics:
+            for m_data in task.last_metrics.values():
+                for v_data in m_data.values():
+                    last_metrics[(v_data.metric, v_data.variant)] = v_data
+
+        for metric_key, metric_values in metrics:
+            for variant_key, data in metric_values.items():
+                last_metrics_data = last_metrics.get((metric_key, variant_key))
+                if last_metrics_data and last_metrics_data.x_axis_label is not None:
+                    data["x_axis_label"] = last_metrics_data.x_axis_label
+            ret[metric_key].update(metric_values)
+
+        return ret
+
+    def compare_scalar_metrics_average_per_iter(
+        self,
+        companies: TaskCompanies,
+        samples,
+        key: ScalarKeyEnum,
+        metric_variants: MetricVariants = None,
+        model_events: bool = False,
+    ):
+        """
+        Compare scalar metrics for different tasks per metric and variant
+        The amount of points in each histogram should not exceed the requested samples
+        """
+        event_type = EventType.metrics_scalar
+        companies = {
+            company_id: tasks
+            for company_id, tasks in companies.items()
+            if not check_empty_data(
+                self.es, company_id=company_id, event_type=event_type
+            )
+        }
+        if not companies:
+            return {}
+
+        get_scalar_average_per_iter = partial(
+            self._get_scalar_average_per_iter_core,
+            event_type=event_type,
+            samples=samples,
+            key=ScalarKey.resolve(key),
+            metric_variants=metric_variants,
+            run_parallel=False,
+            model_events=model_events,
+        )
+        task_ids, company_ids = zip(
+            *(
+                (t.id, t.company)
+                for t in itertools.chain.from_iterable(companies.values())
+            )
+        )
+        with ThreadPoolExecutor(max_workers=EventSettings.max_workers) as pool:
+            task_metrics = zip(
+                task_ids, pool.map(get_scalar_average_per_iter, task_ids, company_ids)
+            )
+
+        task_names = {
+            t.id: t.name for t in itertools.chain.from_iterable(companies.values())
+        }
+        res = defaultdict(lambda: defaultdict(dict))
+        for task_id, task_data in task_metrics:
+            task_name = task_names[task_id]
+            for metric_key, metric_data in task_data.items():
+                for variant_key, variant_data in metric_data.items():
+                    variant_data["name"] = task_name
+                    res[metric_key][variant_key][task_id] = variant_data
+
+        return res
+
+    def get_task_single_value_metrics(
+        self,
+        companies: TaskCompanies,
+        metric_variants: MetricVariants = None,
+    ) -> Mapping[str, Sequence[dict]]:
+        """
+        For the requested tasks return all the events delivered for the single iteration (-2**31)
+        """
+        companies = {
+            company_id: [t.id for t in tasks]
+            for company_id, tasks in companies.items()
+            if not check_empty_data(
+                self.es, company_id=company_id, event_type=EventType.metrics_scalar
+            )
+        }
+        if not companies:
+            return {}
+
+        with ThreadPoolExecutor(max_workers=EventSettings.max_workers) as pool:
+            task_events = list(
+                itertools.chain.from_iterable(
+                    pool.map(
+                        partial(
+                            self._get_task_single_value_metrics,
+                            metric_variants=metric_variants,
+                        ),
+                        companies.items(),
+                    )
+                ),
+            )
+
+        def _get_value(event: dict):
+            return {
+                field: event.get(field)
+                for field in ("metric", "variant", "value", "timestamp")
+            }
+
+        return {
+            task: [_get_value(e) for e in events]
+            for task, events in bucketize(task_events, itemgetter("task")).items()
+        }
+
+    def _get_task_single_value_metrics(
+        self, tasks: Tuple[str, Sequence[str]], metric_variants: MetricVariants = None
+    ) -> Sequence[dict]:
+        company_id, task_ids = tasks
+        must = [
+            {"terms": {"task": task_ids}},
+            {"term": {"iter": SINGLE_SCALAR_ITERATION}},
+        ]
+        if metric_variants:
+            must.append(get_metric_variants_condition(metric_variants))
+
+        es_req = {
+            "size": 10000,
+            "query": {"bool": {"must": must}},
+        }
+        with translate_errors_context():
+            es_res = search_company_events(
+                body=es_req,
+                es=self.es,
+                company_id=company_id,
+                event_type=EventType.metrics_scalar,
+            )
+            if not es_res["hits"]["total"]["value"]:
+                return []
+
+        return [hit["_source"] for hit in es_res["hits"]["hits"]]
+
+    MetricInterval = Tuple[str, str, int, int]
+    MetricIntervalGroup = Tuple[int, Sequence[Tuple[str, str]]]
+
+    @classmethod
+    def _group_task_metric_intervals(
+        cls, intervals: Sequence[MetricInterval]
+    ) -> Sequence[MetricIntervalGroup]:
+        """
+        Group task metric intervals so that the following conditions are meat:
+            - All the metrics in the same group have the same interval (with 10% rounding)
+            - The amount of metrics in the group does not exceed MAX_AGGS_ELEMENTS_COUNT
+            - The total count of samples in the group does not exceed MAX_SAMPLE_BUCKETS
+        """
+        metric_interval_groups = []
+        interval_group = []
+        group_interval_upper_bound = 0
+        group_max_interval = 0
+        group_samples = 0
+        for metric, variant, interval, size in sorted(intervals, key=itemgetter(2)):
+            if (
+                interval > group_interval_upper_bound
+                or (group_samples + size) > cls.MAX_SAMPLE_BUCKETS
+                or len(interval_group) >= cls.MAX_AGGS_ELEMENTS_COUNT
+            ):
+                if interval_group:
+                    metric_interval_groups.append((group_max_interval, interval_group))
+                    interval_group = []
+                group_max_interval = interval
+                group_interval_upper_bound = interval + int(interval * 0.1)
+                group_samples = 0
+            interval_group.append((metric, variant))
+            group_samples += size
+            group_max_interval = max(group_max_interval, interval)
+        if interval_group:
+            metric_interval_groups.append((group_max_interval, interval_group))
+
+        return metric_interval_groups
+
+    def _get_task_metric_intervals(
+        self,
+        company_id: str,
+        event_type: EventType,
+        task_id: str,
+        samples: int,
+        field: str = "iter",
+        metric_variants: MetricVariants = None,
+    ) -> Sequence[MetricInterval]:
+        """
+        Calculate interval per task metric variant so that the resulting
+        amount of points does not exceed sample.
+        Return the list og metric variant intervals as the following tuple:
+        (metric, variant, interval, samples)
+        """
+        must = self._task_conditions(task_id)
+        if metric_variants:
+            must.append(get_metric_variants_condition(metric_variants))
+        query = {"bool": {"must": must}}
+        search_args = dict(es=self.es, company_id=company_id, event_type=event_type)
+        max_metrics, max_variants = get_max_metric_and_variant_counts(
+            query=query,
+            **search_args,
+        )
+        max_variants = int(max_variants // 2)
+        es_req = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": max_metrics,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "variants": {
+                            "terms": {
+                                "field": "variant",
+                                "size": max_variants,
+                                "order": {"_key": "asc"},
+                            },
+                            "aggs": {
+                                "count": {"value_count": {"field": field}},
+                                "min_index": {"min": {"field": field}},
+                                "max_index": {"max": {"field": field}},
+                            },
+                        }
+                    },
+                }
+            },
+        }
+
+        es_res = search_company_events(body=es_req, **search_args)
+
+        aggs_result = es_res.get("aggregations")
+        if not aggs_result:
+            return []
+
+        return [
+            self._build_metric_interval(metric["key"], variant["key"], variant, samples)
+            for metric in aggs_result["metrics"]["buckets"]
+            for variant in metric["variants"]["buckets"]
+        ]
+
+    @staticmethod
+    def _build_metric_interval(
+        metric: str, variant: str, data: dict, samples: int
+    ) -> Tuple[str, str, int, int]:
+        """
+        Calculate index interval per metric_variant variant so that the
+        total amount of intervals does not exceeds the samples
+        Return the interval and resulting amount of intervals
+        """
+        count = nested_get(data, ("count", "value"), default=0)
+        if count < samples:
+            return metric, variant, 1, count
+
+        min_index = nested_get(data, ("min_index", "value"), default=0)
+        max_index = nested_get(data, ("max_index", "value"), default=min_index)
+        index_range = max_index - min_index + 1
+        interval = max(1, math.ceil(float(index_range) / samples))
+        max_samples = math.ceil(float(index_range) / interval)
+        return (
+            metric,
+            variant,
+            interval,
+            max_samples,
+        )
+
+    MetricData = Tuple[str, dict]
+
+    def _get_scalar_average(
+        self,
+        metrics_interval: MetricIntervalGroup,
+        task_id: str,
+        company_id: str,
+        event_type: EventType,
+        key: ScalarKey,
+    ) -> Sequence[MetricData]:
+        """
+        Retrieve scalar histograms per several metric variants that share the same interval
+        """
+        interval, metrics = metrics_interval
+        aggregation = self._add_aggregation_average(key.get_aggregation(interval))
+        query = self._get_task_metrics_query(task_id=task_id, metrics=metrics)
+        search_args = dict(es=self.es, company_id=company_id, event_type=event_type)
+        max_metrics, max_variants = get_max_metric_and_variant_counts(
+            query=query,
+            **search_args,
+        )
+        max_variants = int(max_variants // 2)
+        es_req = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": max_metrics,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "variants": {
+                            "terms": {
+                                "field": "variant",
+                                "size": max_variants,
+                                "order": {"_key": "asc"},
+                            },
+                            "aggs": aggregation,
+                        }
+                    },
+                }
+            },
+        }
+
+        with translate_errors_context():
+            es_res = search_company_events(body=es_req, **search_args)
+
+        aggs_result = es_res.get("aggregations")
+        if not aggs_result:
+            return {}
+
+        metrics = [
+            (
+                metric["key"],
+                {
+                    variant["key"]: {
+                        "name": variant["key"],
+                        **key.get_iterations_data(variant),
+                    }
+                    for variant in metric["variants"]["buckets"]
+                },
+            )
+            for metric in aggs_result["metrics"]["buckets"]
+        ]
+        return metrics
+
+    @staticmethod
+    def _add_aggregation_average(aggregation):
+        average_agg = {"avg_val": {"avg": {"field": "value"}}}
+        return {
+            key: {**value, "aggs": {**value.get("aggs", {}), **average_agg}}
+            for key, value in aggregation.items()
+        }
+
+    @staticmethod
+    def _task_conditions(task_id: str) -> list:
+        return [
+            {"term": {"task": task_id}},
+            {"range": {"iter": {"gt": SINGLE_SCALAR_ITERATION}}},
+        ]
+
+    @classmethod
+    def _get_task_metrics_query(
+        cls,
+        task_id: str,
+        metrics: Sequence[Tuple[str, str]],
+    ):
+        must = cls._task_conditions(task_id)
+        if metrics:
+            should = [
+                {
+                    "bool": {
+                        "must": [
+                            {"term": {"metric": metric}},
+                            {"term": {"variant": variant}},
+                        ]
+                    }
+                }
+                for metric, variant in metrics
+            ]
+            must.append({"bool": {"should": should}})
+
+        return {"bool": {"must": must}}
+
+    def get_multi_task_metrics(self, companies: TaskCompanies, event_type: EventType) -> Mapping[str, list]:
+        """
+        For the requested tasks return reported metrics and variants
+        """
+        tasks_ids = {
+            company: [t.id for t in tasks]
+            for company, tasks in companies.items()
+        }
+        with ThreadPoolExecutor(EventSettings.max_workers) as pool:
+            companies_res: Sequence = list(
+                pool.map(
+                    partial(
+                        self._get_multi_task_metrics,
+                        event_type=event_type,
+                    ),
+                    tasks_ids.items(),
+                )
+            )
+
+        if len(companies_res) == 1:
+            return companies_res[0]
+
+        res = defaultdict(set)
+        for c_res in companies_res:
+            for m, vars_ in c_res.items():
+                res[m].update(vars_)
+
+        return {
+            k: list(v)
+            for k, v in res.items()
+        }
+
+    def _get_multi_task_metrics(
+        self, company_tasks: Tuple[str, Sequence[str]], event_type: EventType
+    ) -> Mapping[str, list]:
+        company_id, task_ids = company_tasks
+        if check_empty_data(self.es, company_id, event_type):
+            return {}
+
+        search_args = dict(
+            es=self.es,
+            company_id=company_id,
+            event_type=event_type,
+        )
+        query = QueryBuilder.terms("task", task_ids)
+        max_metrics, max_variants = get_max_metric_and_variant_counts(
+            query=query,
+            **search_args,
+        )
+        es_req = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": max_metrics,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "variants": {
+                            "terms": {
+                                "field": "variant",
+                                "size": max_variants,
+                                "order": {"_key": "asc"},
+                            },
+                        }
+                    }
+                }
+            },
+        }
+
+        es_res = search_company_events(
+            body=es_req,
+            **search_args,
+        )
+        aggs_result = es_res.get("aggregations")
+        if not aggs_result:
+            return {}
+
+        return {
+            mb["key"]: [vb["key"] for vb in mb["variants"]["buckets"]]
+            for mb in aggs_result["metrics"]["buckets"]
+        }
+
+    def get_task_metrics(
+        self, company_id, task_ids: Sequence, event_type: EventType
+    ) -> Sequence:
+        """
+        For the requested tasks return reported metrics per task
+        """
+        if check_empty_data(self.es, company_id, event_type):
+            return {}
+
+        with ThreadPoolExecutor(EventSettings.max_workers) as pool:
+            res = pool.map(
+                partial(
+                    self._get_task_metrics,
+                    company_id=company_id,
+                    event_type=event_type,
+                ),
+                task_ids,
+            )
+        return list(zip(task_ids, res))
+
+    def _get_task_metrics(
+        self, task_id: str, company_id: str, event_type: EventType
+    ) -> Sequence:
+        es_req = {
+            "size": 0,
+            "query": {"bool": {"must": self._task_conditions(task_id)}},
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": EventSettings.max_es_buckets,
+                        "order": {"_key": "asc"},
+                    }
+                }
+            },
+        }
+
+        es_res = search_company_events(
+            self.es, company_id=company_id, event_type=event_type, body=es_req
+        )
+
+        return [
+            metric["key"]
+            for metric in nested_get(es_res, ("aggregations", "metrics", "buckets"), default=[])
+        ]

apiserver/bll/event/events_iterator.py

@@ -0,0 +1,197 @@
+from typing import Optional, Tuple, Sequence, Any
+
+import attr
+import jsonmodels.models
+import jwt
+from elasticsearch import Elasticsearch
+from jwt.algorithms import get_default_algorithms
+
+from apiserver.bll.event.event_common import (
+    check_empty_data,
+    search_company_events,
+    EventType,
+    MetricVariants,
+    get_metric_variants_condition,
+    count_company_events,
+)
+from apiserver.bll.event.scalar_key import ScalarKeyEnum, ScalarKey
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+
+
+@attr.s(auto_attribs=True)
+class TaskEventsResult:
+    total_events: int = 0
+    next_scroll_id: str = None
+    events: list = attr.Factory(list)
+
+
+class EventsIterator:
+    def __init__(self, es: Elasticsearch):
+        self.es = es
+
+    def get_task_events(
+        self,
+        event_type: EventType,
+        company_id: str,
+        task_id: str,
+        batch_size: int,
+        navigate_earlier: bool = True,
+        from_key_value: Optional[Any] = None,
+        metric_variants: MetricVariants = None,
+        key: ScalarKeyEnum = ScalarKeyEnum.timestamp,
+        **kwargs,
+    ) -> TaskEventsResult:
+        if check_empty_data(self.es, company_id, event_type):
+            return TaskEventsResult()
+
+        from_key_value = kwargs.pop("from_timestamp", from_key_value)
+
+        res = TaskEventsResult()
+        res.events, res.total_events = self._get_events(
+            event_type=event_type,
+            company_id=company_id,
+            task_id=task_id,
+            batch_size=batch_size,
+            navigate_earlier=navigate_earlier,
+            from_key_value=from_key_value,
+            metric_variants=metric_variants,
+            key=ScalarKey.resolve(key),
+        )
+        return res
+
+    def count_task_events(
+        self,
+        event_type: EventType,
+        company_id: str,
+        task_ids: Sequence[str],
+        metric_variants: MetricVariants = None,
+    ) -> int:
+        if check_empty_data(self.es, company_id, event_type):
+            return 0
+
+        query, _ = self._get_initial_query_and_must(task_ids, metric_variants)
+        es_req = {
+            "query": query,
+        }
+
+        with translate_errors_context():
+            es_result = count_company_events(
+                self.es, company_id=company_id, event_type=event_type, body=es_req,
+            )
+
+            return es_result["count"]
+
+    def _get_events(
+        self,
+        event_type: EventType,
+        company_id: str,
+        task_id: str,
+        batch_size: int,
+        navigate_earlier: bool,
+        key: ScalarKey,
+        from_key_value: Optional[Any],
+        metric_variants: MetricVariants = None,
+    ) -> Tuple[Sequence[dict], int]:
+        """
+        Return up to 'batch size' events starting from the previous key-field value (timestamp or iter) either in the
+        direction of earlier events (navigate_earlier=True) or in the direction of later events.
+        If from_key_field is not set then start either from latest or earliest.
+        For the last key-field value all the events are brought (even if the resulting size exceeds batch_size)
+        so that events with this value will not be lost between the calls.
+        """
+        query, must = self._get_initial_query_and_must([task_id], metric_variants)
+
+        # retrieve the next batch of events
+        es_req = {
+            "size": batch_size,
+            "query": query,
+            "sort": {key.field: "desc" if navigate_earlier else "asc"},
+        }
+
+        if from_key_value:
+            es_req["search_after"] = [from_key_value]
+
+        with translate_errors_context():
+            es_result = search_company_events(
+                self.es, company_id=company_id, event_type=event_type, body=es_req,
+            )
+            hits = es_result["hits"]["hits"]
+            hits_total = es_result["hits"]["total"]["value"]
+            if not hits:
+                return [], hits_total
+
+            events = [hit["_source"] for hit in hits]
+
+            # retrieve the events that match the last event timestamp
+            # but did not make it into the previous call due to batch_size limitation
+            es_req = {
+                "size": 10000,
+                "query": {
+                    "bool": {
+                        "must": must + [{"term": {key.field: events[-1][key.field]}}]
+                    }
+                },
+            }
+            es_result = search_company_events(
+                self.es, company_id=company_id, event_type=event_type, body=es_req,
+            )
+            last_second_hits = es_result["hits"]["hits"]
+            if not last_second_hits or len(last_second_hits) < 2:
+                # if only one element is returned for the last timestamp
+                # then it is already present in the events
+                return events, hits_total
+
+            already_present_ids = set(hit["_id"] for hit in hits)
+            last_second_events = [
+                hit["_source"]
+                for hit in last_second_hits
+                if hit["_id"] not in already_present_ids
+            ]
+
+            # return the list merged from original query results +
+            # leftovers from the last timestamp
+            return (
+                [*events, *last_second_events],
+                hits_total,
+            )
+
+    @staticmethod
+    def _get_initial_query_and_must(
+        task_ids: Sequence[str], metric_variants: MetricVariants = None
+    ) -> Tuple[dict, list]:
+        if not metric_variants:
+            query = {"terms": {"task": task_ids}}
+            must = [query]
+        else:
+            must = [
+                {"terms": {"task": task_ids}},
+                get_metric_variants_condition(metric_variants),
+            ]
+            query = {"bool": {"must": must}}
+        return query, must
+
+
+class Scroll(jsonmodels.models.Base):
+    def get_scroll_id(self) -> str:
+        return jwt.encode(
+            self.to_struct(),
+            key=config.get(
+                "services.events.events_retrieval.scroll_id_key", "1234567890"
+            ),
+        )
+
+    @classmethod
+    def from_scroll_id(cls, scroll_id: str):
+        try:
+            return cls(
+                **jwt.decode(
+                    scroll_id,
+                    key=config.get(
+                        "services.events.events_retrieval.scroll_id_key", "1234567890"
+                    ),
+                    algorithms=get_default_algorithms(),
+                )
+            )
+        except jwt.PyJWTError:
+            raise ValueError("Invalid Scroll ID")

apiserver/bll/event/history_debug_image_iterator.py

@@ -0,0 +1,455 @@
+import operator
+from operator import attrgetter
+from typing import Sequence, Tuple, Optional, Mapping
+
+import attr
+from boltons.iterutils import first, bucketize
+from elasticsearch import Elasticsearch
+from jsonmodels.fields import StringField, IntField, BoolField, ListField
+from jsonmodels.models import Base
+from redis.client import StrictRedis
+
+from apiserver.utilities.dicts import nested_get
+from .event_common import (
+    EventType,
+    EventSettings,
+    check_empty_data,
+    search_company_events,
+    get_max_metric_and_variant_counts,
+)
+from apiserver.apimodels import JsonSerializableMixin
+from apiserver.bll.redis_cache_manager import RedisCacheManager
+from apiserver.apierrors import errors
+
+
+class VariantState(Base):
+    name: str = StringField(required=True)
+    metric: str = StringField(default=None)
+    min_iteration: int = IntField()
+    max_iteration: int = IntField()
+
+
+class DebugImageSampleState(Base, JsonSerializableMixin):
+    id: str = StringField(required=True)
+    iteration: int = IntField()
+    variant: str = StringField()
+    task: str = StringField()
+    metric: str = StringField()
+    variant_states: Sequence[VariantState] = ListField([VariantState])
+    warning: str = StringField()
+    navigate_current_metric = BoolField(default=True)
+
+
+@attr.s(auto_attribs=True)
+class VariantSampleResult(object):
+    scroll_id: str = None
+    event: dict = None
+    min_iteration: int = None
+    max_iteration: int = None
+
+
+class HistoryDebugImageIterator:
+    event_type = EventType.metrics_image
+
+    def __init__(self, redis: StrictRedis, es: Elasticsearch):
+        self.es = es
+        self.cache_manager = RedisCacheManager(
+            state_class=DebugImageSampleState,
+            redis=redis,
+            expiration_interval=EventSettings.state_expiration_sec,
+        )
+
+    def get_next_sample(
+        self,
+        company_id: str,
+        task: str,
+        state_id: str,
+        navigate_earlier: bool,
+        next_iteration: bool,
+    ) -> VariantSampleResult:
+        """
+        Get the sample for next/prev variant on the current iteration
+        If does not exist then try getting sample for the first/last variant from next/prev iteration
+        """
+        res = VariantSampleResult(scroll_id=state_id)
+        state = self.cache_manager.get_state(state_id)
+        if not state or state.task != task:
+            raise errors.bad_request.InvalidScrollId(scroll_id=state_id)
+
+        if check_empty_data(self.es, company_id=company_id, event_type=self.event_type):
+            return res
+
+        if next_iteration:
+            event = self._get_next_for_another_iteration(
+                company_id=company_id, navigate_earlier=navigate_earlier, state=state
+            )
+        else:
+            # noinspection PyArgumentList
+            event = first(
+                f(company_id=company_id, navigate_earlier=navigate_earlier, state=state)
+                for f in (
+                    self._get_next_for_current_iteration,
+                    self._get_next_for_another_iteration,
+                )
+            )
+        if not event:
+            return res
+
+        self._fill_res_and_update_state(event=event, res=res, state=state)
+        self.cache_manager.set_state(state=state)
+        return res
+
+    @staticmethod
+    def _fill_res_and_update_state(
+        event: dict, res: VariantSampleResult, state: DebugImageSampleState
+    ):
+        state.variant = event["variant"]
+        state.metric = event["metric"]
+        state.iteration = event["iter"]
+        res.event = event
+        var_state = first(
+            vs
+            for vs in state.variant_states
+            if vs.name == state.variant and vs.metric == state.metric
+        )
+        if var_state:
+            res.min_iteration = var_state.min_iteration
+            res.max_iteration = var_state.max_iteration
+
+    @staticmethod
+    def _get_metric_conditions(variants: Sequence[VariantState]) -> dict:
+        metrics = bucketize(variants, key=attrgetter("metric"))
+
+        def _get_variants_conditions(metric_variants: Sequence[VariantState]) -> dict:
+            variants_conditions = [
+                {
+                    "bool": {
+                        "must": [
+                            {"term": {"variant": v.name}},
+                            {"range": {"iter": {"gte": v.min_iteration}}},
+                        ]
+                    }
+                }
+                for v in metric_variants
+            ]
+            return {"bool": {"should": variants_conditions}}
+
+        metrics_conditions = [
+            {
+                "bool": {
+                    "must": [
+                        {"term": {"metric": metric}},
+                        _get_variants_conditions(metric_variants),
+                    ]
+                }
+            }
+            for metric, metric_variants in metrics.items()
+        ]
+        return {"bool": {"should": metrics_conditions}}
+
+    def _get_next_for_current_iteration(
+        self, company_id: str, navigate_earlier: bool, state: DebugImageSampleState
+    ) -> Optional[dict]:
+        """
+        Get the sample for next (if navigate_earlier is False) or previous variant sorted by name for the same iteration
+        Only variants for which the iteration falls into their valid range are considered
+        Return None if no such variant or sample is found
+        """
+        if state.navigate_current_metric:
+            variants = [
+                var_state
+                for var_state in state.variant_states
+                if var_state.metric == state.metric
+            ]
+        else:
+            variants = state.variant_states
+
+        cmp = operator.lt if navigate_earlier else operator.gt
+        variants = [
+            var_state
+            for var_state in variants
+            if cmp((var_state.metric, var_state.name), (state.metric, state.variant))
+            and var_state.min_iteration <= state.iteration
+        ]
+        if not variants:
+            return
+
+        must_conditions = [
+            {"term": {"task": state.task}},
+            {"term": {"iter": state.iteration}},
+            self._get_metric_conditions(variants),
+            {"exists": {"field": "url"}},
+        ]
+        order = "desc" if navigate_earlier else "asc"
+        es_req = {
+            "size": 1,
+            "sort": [{"metric": order}, {"variant": order}, {"url": "desc"}],
+            "query": {"bool": {"must": must_conditions}},
+        }
+
+        es_res = search_company_events(
+            self.es,
+            company_id=company_id,
+            event_type=self.event_type,
+            body=es_req,
+        )
+
+        hits = nested_get(es_res, ("hits", "hits"))
+        if not hits:
+            return
+
+        return hits[0]["_source"]
+
+    def _get_next_for_another_iteration(
+        self, company_id: str, navigate_earlier: bool, state: DebugImageSampleState
+    ) -> Optional[dict]:
+        """
+        Get the sample for the first variant for the next iteration (if navigate_earlier is set to False)
+        or from the last variant for the previous iteration (otherwise)
+        The variants for which the sample falls in invalid range are discarded
+        If no suitable sample is found then None is returned
+        """
+        if state.navigate_current_metric:
+            variants = [
+                var_state
+                for var_state in state.variant_states
+                if var_state.metric == state.metric
+            ]
+        else:
+            variants = state.variant_states
+
+        if navigate_earlier:
+            range_operator = "lt"
+            order = "desc"
+            variants = [
+                var_state
+                for var_state in variants
+                if var_state.min_iteration < state.iteration
+            ]
+        else:
+            range_operator = "gt"
+            order = "asc"
+            variants = variants
+
+        if not variants:
+            return
+
+        must_conditions = [
+            {"term": {"task": state.task}},
+            self._get_metric_conditions(variants),
+            {"range": {"iter": {range_operator: state.iteration}}},
+            {"exists": {"field": "url"}},
+        ]
+        es_req = {
+            "size": 1,
+            "sort": [{"iter": order}, {"metric": order}, {"variant": order}, {"url": "desc"}],
+            "query": {"bool": {"must": must_conditions}},
+        }
+        es_res = search_company_events(
+            self.es,
+            company_id=company_id,
+            event_type=self.event_type,
+            body=es_req,
+        )
+
+        hits = nested_get(es_res, ("hits", "hits"))
+        if not hits:
+            return
+
+        return hits[0]["_source"]
+
+    def get_sample_for_variant(
+        self,
+        company_id: str,
+        task: str,
+        metric: str,
+        variant: str,
+        iteration: Optional[int] = None,
+        refresh: bool = False,
+        state_id: str = None,
+        navigate_current_metric: bool = True,
+    ) -> VariantSampleResult:
+        """
+        Get the sample for the requested iteration or the latest before it
+        If the iteration is not passed then get the latest event
+        """
+        res = VariantSampleResult()
+        if check_empty_data(self.es, company_id=company_id, event_type=self.event_type):
+            return res
+
+        def init_state(state_: DebugImageSampleState):
+            state_.task = task
+            state_.metric = metric
+            state_.navigate_current_metric = navigate_current_metric
+            self._reset_variant_states(company_id=company_id, state=state_)
+
+        def validate_state(state_: DebugImageSampleState):
+            if (
+                state_.task != task
+                or state_.navigate_current_metric != navigate_current_metric
+                or (state_.navigate_current_metric and state_.metric != metric)
+            ):
+                raise errors.bad_request.InvalidScrollId(
+                    "Task and metric stored in the state do not match the passed ones",
+                    scroll_id=state_.id,
+                )
+            # fix old variant states:
+            for vs in state_.variant_states:
+                if vs.metric is None:
+                    vs.metric = metric
+            if refresh:
+                self._reset_variant_states(company_id=company_id, state=state_)
+
+        state: DebugImageSampleState
+        with self.cache_manager.get_or_create_state(
+            state_id=state_id, init_state=init_state, validate_state=validate_state,
+        ) as state:
+            res.scroll_id = state.id
+
+            var_state = first(
+                vs
+                for vs in state.variant_states
+                if vs.name == variant and vs.metric == metric
+            )
+            if not var_state:
+                return res
+
+            res.min_iteration = var_state.min_iteration
+            res.max_iteration = var_state.max_iteration
+
+            must_conditions = [
+                {"term": {"task": task}},
+                {"term": {"metric": metric}},
+                {"term": {"variant": variant}},
+                {"exists": {"field": "url"}},
+            ]
+            if iteration is not None:
+                must_conditions.append(
+                    {
+                        "range": {
+                            "iter": {"lte": iteration, "gte": var_state.min_iteration}
+                        }
+                    }
+                )
+            else:
+                must_conditions.append(
+                    {"range": {"iter": {"gte": var_state.min_iteration}}}
+                )
+
+            es_req = {
+                "size": 1,
+                "sort": [{"iter": "desc"}, {"url": "desc"}],
+                "query": {"bool": {"must": must_conditions}},
+            }
+
+            es_res = search_company_events(
+                self.es,
+                company_id=company_id,
+                event_type=self.event_type,
+                body=es_req,
+            )
+
+            hits = nested_get(es_res, ("hits", "hits"))
+            if not hits:
+                return res
+
+            self._fill_res_and_update_state(
+                event=hits[0]["_source"], res=res, state=state
+            )
+            return res
+
+    def _reset_variant_states(self, company_id: str, state: DebugImageSampleState):
+        metrics = self._get_metric_variant_iterations(
+            company_id=company_id,
+            task=state.task,
+            metric=state.metric if state.navigate_current_metric else None,
+        )
+        state.variant_states = [
+            VariantState(
+                metric=metric,
+                name=var_name,
+                min_iteration=min_iter,
+                max_iteration=max_iter,
+            )
+            for metric, variants in metrics.items()
+            for var_name, min_iter, max_iter in variants
+        ]
+
+    def _get_metric_variant_iterations(
+        self, company_id: str, task: str, metric: str,
+    ) -> Mapping[str, Sequence[Tuple[str, int, int]]]:
+        """
+        Return valid min and max iterations that the task reported events of the required type
+        """
+        must = [
+            {"term": {"task": task}},
+            {"exists": {"field": "url"}},
+        ]
+        if metric is not None:
+            must.append({"term": {"metric": metric}})
+        query = {"bool": {"must": must}}
+
+        search_args = dict(
+            es=self.es, company_id=company_id, event_type=self.event_type,
+        )
+        max_metrics, max_variants = get_max_metric_and_variant_counts(
+            query=query, **search_args
+        )
+        max_variants = int(max_variants // 2)
+        es_req: dict = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": max_metrics,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "variants": {
+                            "terms": {
+                                "field": "variant",
+                                "size": max_variants,
+                                "order": {"_key": "asc"},
+                            },
+                            "aggs": {
+                                "last_iter": {"max": {"field": "iter"}},
+                                "urls": {
+                                    # group by urls and choose the minimal iteration
+                                    # from all the maximal iterations per url
+                                    "terms": {
+                                        "field": "url",
+                                        "order": {"max_iter": "asc"},
+                                        "size": 1,
+                                    },
+                                    "aggs": {
+                                        # find max iteration for each url
+                                        "max_iter": {"max": {"field": "iter"}}
+                                    },
+                                },
+                            },
+                        }
+                    },
+                }
+            },
+        }
+
+        es_res = search_company_events(body=es_req, **search_args)
+
+        def get_variant_data(variant_bucket: dict) -> Tuple[str, int, int]:
+            variant = variant_bucket["key"]
+            urls = nested_get(variant_bucket, ("urls", "buckets"))
+            min_iter = int(urls[0]["max_iter"]["value"])
+            max_iter = int(variant_bucket["last_iter"]["value"])
+            return variant, min_iter, max_iter
+
+        return {
+            metric_bucket["key"]: [
+                get_variant_data(variant_bucket)
+                for variant_bucket in nested_get(metric_bucket, ("variants", "buckets"))
+            ]
+            for metric_bucket in nested_get(
+                es_res, ("aggregations", "metrics", "buckets")
+            )
+        }

apiserver/bll/event/history_plots_iterator.py

@@ -0,0 +1,316 @@
+from typing import Sequence, Tuple, Optional, Mapping
+
+import attr
+from boltons.iterutils import first
+from elasticsearch import Elasticsearch
+from jsonmodels.fields import StringField, IntField, ListField, BoolField
+from jsonmodels.models import Base
+from redis.client import StrictRedis
+
+from .event_common import (
+    EventType,
+    uncompress_plot,
+    EventSettings,
+    check_empty_data,
+    search_company_events,
+)
+from apiserver.apimodels import JsonSerializableMixin
+from apiserver.utilities.dicts import nested_get
+from apiserver.bll.redis_cache_manager import RedisCacheManager
+from apiserver.apierrors import errors
+
+
+class MetricState(Base):
+    name: str = StringField(default=None)
+    min_iteration: int = IntField()
+    max_iteration: int = IntField()
+
+
+class PlotsSampleState(Base, JsonSerializableMixin):
+    id: str = StringField(required=True)
+    iteration: int = IntField()
+    task: str = StringField()
+    metric: str = StringField()
+    metric_states: Sequence[MetricState] = ListField([MetricState])
+    warning: str = StringField()
+    navigate_current_metric = BoolField(default=True)
+
+
+@attr.s(auto_attribs=True)
+class MetricSamplesResult(object):
+    scroll_id: str = None
+    events: list = []
+    min_iteration: int = None
+    max_iteration: int = None
+
+
+class HistoryPlotsIterator:
+    event_type = EventType.metrics_plot
+
+    def __init__(self, redis: StrictRedis, es: Elasticsearch):
+        self.es = es
+        self.cache_manager = RedisCacheManager(
+            state_class=PlotsSampleState,
+            redis=redis,
+            expiration_interval=EventSettings.state_expiration_sec,
+        )
+
+    def get_next_sample(
+        self,
+        company_id: str,
+        task: str,
+        state_id: str,
+        navigate_earlier: bool,
+        next_iteration: bool,
+    ) -> MetricSamplesResult:
+        """
+        Get the samples for next/prev metric on the current iteration
+        If does not exist then try getting sample for the first/last metric from next/prev iteration
+        """
+        res = MetricSamplesResult(scroll_id=state_id)
+        state = self.cache_manager.get_state(state_id)
+        if not state or state.task != task:
+            raise errors.bad_request.InvalidScrollId(scroll_id=state_id)
+
+        if check_empty_data(self.es, company_id=company_id, event_type=self.event_type):
+            return res
+
+        if navigate_earlier:
+            range_operator = "lt"
+            order = "desc"
+        else:
+            range_operator = "gt"
+            order = "asc"
+
+        must_conditions = [
+            {"term": {"task": state.task}},
+        ]
+        if state.navigate_current_metric:
+            must_conditions.append({"term": {"metric": state.metric}})
+
+        next_iteration_condition = {
+            "range": {"iter": {range_operator: state.iteration}}
+        }
+        if next_iteration or state.navigate_current_metric:
+            must_conditions.append(next_iteration_condition)
+        else:
+            next_metric_condition = {
+                "bool": {
+                    "must": [
+                        {"term": {"iter": state.iteration}},
+                        {"range": {"metric": {range_operator: state.metric}}},
+                    ]
+                }
+            }
+            must_conditions.append(
+                {"bool": {"should": [next_metric_condition, next_iteration_condition]}}
+            )
+
+        events = self._get_metric_events_for_condition(
+            company_id=company_id,
+            task=state.task,
+            order=order,
+            must_conditions=must_conditions,
+        )
+
+        if not events:
+            return res
+
+        self._fill_res_and_update_state(events=events, res=res, state=state)
+        self.cache_manager.set_state(state=state)
+        return res
+
+    def get_samples_for_metric(
+        self,
+        company_id: str,
+        task: str,
+        metric: str,
+        iteration: Optional[int] = None,
+        refresh: bool = False,
+        state_id: str = None,
+        navigate_current_metric: bool = True,
+    ) -> MetricSamplesResult:
+        """
+        Get the sample for the requested iteration or the latest before it
+        If the iteration is not passed then get the latest event
+        """
+        res = MetricSamplesResult()
+        if check_empty_data(self.es, company_id=company_id, event_type=self.event_type):
+            return res
+
+        def init_state(state_: PlotsSampleState):
+            state_.task = task
+            state_.metric = metric
+            state_.navigate_current_metric = navigate_current_metric
+            self._reset_metric_states(company_id=company_id, state=state_)
+
+        def validate_state(state_: PlotsSampleState):
+            if (
+                state_.task != task
+                or state_.navigate_current_metric != navigate_current_metric
+                or (state_.navigate_current_metric and state_.metric != metric)
+            ):
+                raise errors.bad_request.InvalidScrollId(
+                    "Task and metric stored in the state do not match the passed ones",
+                    scroll_id=state_.id,
+                )
+            if refresh:
+                self._reset_metric_states(company_id=company_id, state=state_)
+
+        state: PlotsSampleState
+        with self.cache_manager.get_or_create_state(
+            state_id=state_id, init_state=init_state, validate_state=validate_state,
+        ) as state:
+            res.scroll_id = state.id
+
+            metric_state = first(ms for ms in state.metric_states if ms.name == metric)
+            if not metric_state:
+                return res
+
+            res.min_iteration = metric_state.min_iteration
+            res.max_iteration = metric_state.max_iteration
+
+            must_conditions = [
+                {"term": {"task": task}},
+                {"term": {"metric": metric}},
+            ]
+            if iteration is not None:
+                must_conditions.append({"range": {"iter": {"lte": iteration}}})
+
+            events = self._get_metric_events_for_condition(
+                company_id=company_id,
+                task=state.task,
+                order="desc",
+                must_conditions=must_conditions,
+            )
+            if not events:
+                return res
+
+            self._fill_res_and_update_state(events=events, res=res, state=state)
+            return res
+
+    def _reset_metric_states(self, company_id: str, state: PlotsSampleState):
+        metrics = self._get_metric_iterations(
+            company_id=company_id,
+            task=state.task,
+            metric=state.metric if state.navigate_current_metric else None,
+        )
+        state.metric_states = [
+            MetricState(name=metric, min_iteration=min_iter, max_iteration=max_iter)
+            for metric, (min_iter, max_iter) in metrics.items()
+        ]
+
+    def _get_metric_iterations(
+        self, company_id: str, task: str, metric: str,
+    ) -> Mapping[str, Tuple[int, int]]:
+        """
+        Return valid min and max iterations that the task reported events of the required type
+        """
+        must = [
+            {"term": {"task": task}},
+        ]
+        if metric is not None:
+            must.append({"term": {"metric": metric}})
+        query = {"bool": {"must": must}}
+
+        es_req: dict = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": 5000,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "last_iter": {"max": {"field": "iter"}},
+                        "first_iter": {"min": {"field": "iter"}},
+                    },
+                }
+            },
+        }
+
+        es_res = search_company_events(
+            body=es_req,
+            es=self.es,
+            company_id=company_id,
+            event_type=self.event_type,
+        )
+
+        return {
+            metric_bucket["key"]: (
+                int(metric_bucket["first_iter"]["value"]),
+                int(metric_bucket["last_iter"]["value"]),
+            )
+            for metric_bucket in nested_get(
+                es_res, ("aggregations", "metrics", "buckets")
+            )
+        }
+
+    @staticmethod
+    def _fill_res_and_update_state(
+        events: Sequence[dict], res: MetricSamplesResult, state: PlotsSampleState
+    ):
+        for event in events:
+            uncompress_plot(event)
+        state.metric = events[0]["metric"]
+        state.iteration = events[0]["iter"]
+        res.events = events
+        metric_state = first(
+            ms for ms in state.metric_states if ms.name == state.metric
+        )
+        if metric_state:
+            res.min_iteration = metric_state.min_iteration
+            res.max_iteration = metric_state.max_iteration
+
+    def _get_metric_events_for_condition(
+        self, company_id: str, task: str, order: str, must_conditions: Sequence
+    ) -> Sequence:
+        es_req = {
+            "size": 0,
+            "query": {"bool": {"must": must_conditions}},
+            "aggs": {
+                "iters": {
+                    "terms": {"field": "iter", "size": 1, "order": {"_key": order}},
+                    "aggs": {
+                        "metrics": {
+                            "terms": {
+                                "field": "metric",
+                                "size": 1,
+                                "order": {"_key": order},
+                            },
+                            "aggs": {
+                                "events": {
+                                    "top_hits": {
+                                        "sort": {"variant": {"order": "asc"}},
+                                        "size": 100,
+                                    }
+                                }
+                            },
+                        },
+                    },
+                }
+            },
+        }
+        es_res = search_company_events(
+            self.es,
+            company_id=company_id,
+            event_type=self.event_type,
+            body=es_req,
+        )
+
+        aggs_result = es_res.get("aggregations")
+        if not aggs_result:
+            return []
+
+        for level in ("iters", "metrics"):
+            level_data = aggs_result[level]["buckets"]
+            if not level_data:
+                return []
+            aggs_result = level_data[0]
+
+        return [
+            hit["_source"]
+            for hit in nested_get(aggs_result, ("events", "hits", "hits"))
+        ]

apiserver/bll/event/metric_debug_images_iterator.py

@@ -0,0 +1,53 @@
+from typing import Sequence, Tuple, Callable
+
+from elasticsearch import Elasticsearch
+from redis.client import StrictRedis
+
+from apiserver.utilities.dicts import nested_get
+from .event_common import EventType
+from .metric_events_iterator import MetricEventsIterator, VariantState
+
+
+class MetricDebugImagesIterator(MetricEventsIterator):
+    def __init__(self, redis: StrictRedis, es: Elasticsearch):
+        super().__init__(redis, es, EventType.metrics_image)
+
+    def _get_extra_conditions(self) -> Sequence[dict]:
+        return [{"exists": {"field": "url"}}]
+
+    def _get_variant_state_aggs(self) -> Tuple[dict, Callable[[dict, VariantState], None]]:
+        aggs = {
+            "urls": {
+                "terms": {
+                    "field": "url",
+                    "order": {"max_iter": "desc"},
+                    "size": 1,  # we need only one url from the most recent iteration
+                },
+                "aggs": {
+                    "max_iter": {"max": {"field": "iter"}},
+                    "iters": {
+                        "top_hits": {
+                            "sort": {"iter": {"order": "desc"}},
+                            "size": 2,  # need two last iterations so that we can take
+                            # the second one as invalid
+                            "_source": "iter",
+                        }
+                    },
+                },
+            }
+        }
+
+        def fill_variant_state_data(variant_bucket: dict,  state: VariantState):
+            """If the image urls get recycled then fill the last_invalid_iteration field"""
+            top_iter_url = nested_get(variant_bucket, ("urls", "buckets"))[0]
+            iters = nested_get(top_iter_url, ("iters", "hits", "hits"))
+            if len(iters) > 1:
+                state.last_invalid_iteration = nested_get(iters[1], ("_source", "iter"))
+
+        return aggs, fill_variant_state_data
+
+    def _process_event(self, event: dict) -> dict:
+        return event
+
+    def _get_same_variant_events_order(self) -> dict:
+        return {"url": {"order": "desc"}}

apiserver/bll/event/metric_events_iterator.py

@@ -0,0 +1,452 @@
+import abc
+from concurrent.futures.thread import ThreadPoolExecutor
+from datetime import datetime
+from functools import partial
+from operator import itemgetter
+from typing import Sequence, Tuple, Optional, Mapping, Callable
+
+import attr
+from boltons.iterutils import first
+from elasticsearch import Elasticsearch
+from jsonmodels.fields import StringField, ListField, IntField
+from jsonmodels.models import Base
+from redis import StrictRedis
+
+from apiserver.apimodels import JsonSerializableMixin
+from apiserver.bll.event.event_common import (
+    EventSettings,
+    check_empty_data,
+    search_company_events,
+    EventType,
+    get_metric_variants_condition,
+    get_max_metric_and_variant_counts,
+)
+from apiserver.bll.redis_cache_manager import RedisCacheManager
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.task.metrics import MetricEventStats
+from apiserver.database.model.task.task import Task
+from apiserver.utilities.dicts import nested_get
+
+
+class VariantState(Base):
+    variant: str = StringField(required=True)
+    last_invalid_iteration: int = IntField()
+
+
+class MetricState(Base):
+    metric: str = StringField(required=True)
+    variants: Sequence[VariantState] = ListField([VariantState], required=True)
+    timestamp: int = IntField(default=0)
+
+
+class TaskScrollState(Base):
+    task: str = StringField(required=True)
+    metrics: Sequence[MetricState] = ListField([MetricState], required=True)
+    last_min_iter: Optional[int] = IntField()
+    last_max_iter: Optional[int] = IntField()
+
+    def reset(self):
+        """Reset the scrolling state for the metric"""
+        self.last_min_iter = self.last_max_iter = None
+
+
+class MetricEventsScrollState(Base, JsonSerializableMixin):
+    id: str = StringField(required=True)
+    tasks: Sequence[TaskScrollState] = ListField([TaskScrollState])
+    warning: str = StringField()
+
+
+@attr.s(auto_attribs=True)
+class MetricEventsResult(object):
+    metric_events: Sequence[tuple] = []
+    next_scroll_id: str = None
+
+
+class MetricEventsIterator:
+    def __init__(self, redis: StrictRedis, es: Elasticsearch, event_type: EventType):
+        self.es = es
+        self.event_type = event_type
+        self.cache_manager = RedisCacheManager(
+            state_class=MetricEventsScrollState,
+            redis=redis,
+            expiration_interval=EventSettings.state_expiration_sec,
+        )
+
+    def get_task_events(
+        self,
+        companies: Mapping[str, str],
+        task_metrics: Mapping[str, dict],
+        iter_count: int,
+        navigate_earlier: bool = True,
+        refresh: bool = False,
+        state_id: str = None,
+    ) -> MetricEventsResult:
+        companies = {
+            task_id: company_id
+            for task_id, company_id in companies.items()
+            if not check_empty_data(
+                self.es, company_id=company_id, event_type=self.event_type
+            )
+        }
+        if not companies:
+            return MetricEventsResult()
+
+        def init_state(state_: MetricEventsScrollState):
+            state_.tasks = self._init_task_states(companies, task_metrics)
+
+        def validate_state(state_: MetricEventsScrollState):
+            """
+            Validate that the metrics stored in the state are the same
+            as requested in the current call.
+            Refresh the state if requested
+            """
+            if refresh:
+                self._reinit_outdated_task_states(companies, state_, task_metrics)
+
+        with self.cache_manager.get_or_create_state(
+            state_id=state_id, init_state=init_state, validate_state=validate_state
+        ) as state:
+            res = MetricEventsResult(next_scroll_id=state.id)
+            specific_variants_requested = any(
+                variants
+                for t, metrics in task_metrics.items()
+                if metrics
+                for m, variants in metrics.items()
+            )
+            with ThreadPoolExecutor(EventSettings.max_workers) as pool:
+                res.metric_events = list(
+                    pool.map(
+                        partial(
+                            self._get_task_metric_events,
+                            companies=companies,
+                            iter_count=iter_count,
+                            navigate_earlier=navigate_earlier,
+                            specific_variants_requested=specific_variants_requested,
+                        ),
+                        state.tasks,
+                    )
+                )
+
+            return res
+
+    def _reinit_outdated_task_states(
+        self,
+        companies: Mapping[str, str],
+        state: MetricEventsScrollState,
+        task_metrics: Mapping[str, dict],
+    ):
+        """
+        Determine the metrics for which new event_type events were added
+        since their states were initialized and re-init these states
+        """
+        tasks = Task.objects(id__in=list(task_metrics)).only("id", "metric_stats")
+
+        def get_last_update_times_for_task_metrics(
+            task: Task,
+        ) -> Mapping[str, datetime]:
+            """For metrics that reported event_type events get mapping of the metric name to the last update times"""
+            metric_stats: Mapping[str, MetricEventStats] = task.metric_stats
+            if not metric_stats:
+                return {}
+
+            requested_metrics = task_metrics[task.id]
+            return {
+                stats.metric: stats.event_stats_by_type[
+                    self.event_type.value
+                ].last_update
+                for stats in metric_stats.values()
+                if self.event_type.value in stats.event_stats_by_type
+                and (not requested_metrics or stats.metric in requested_metrics)
+            }
+
+        update_times = {
+            task.id: get_last_update_times_for_task_metrics(task) for task in tasks
+        }
+        task_metric_states = {
+            task_state.task: {
+                metric_state.metric: metric_state for metric_state in task_state.metrics
+            }
+            for task_state in state.tasks
+        }
+        task_metrics_to_recalc = {}
+        for task, metrics_times in update_times.items():
+            old_metric_states = task_metric_states[task]
+            metrics_to_recalc = {
+                m: task_metrics[task].get(m)
+                for m, t in metrics_times.items()
+                if m not in old_metric_states or old_metric_states[m].timestamp < t
+            }
+            if metrics_to_recalc:
+                task_metrics_to_recalc[task] = metrics_to_recalc
+
+        updated_task_states = self._init_task_states(companies, task_metrics_to_recalc)
+
+        def merge_with_updated_task_states(
+            old_state: TaskScrollState, updates: Sequence[TaskScrollState]
+        ) -> TaskScrollState:
+            task = old_state.task
+            updated_state = first(uts for uts in updates if uts.task == task)
+            if not updated_state:
+                old_state.reset()
+                return old_state
+
+            updated_metrics = [m.metric for m in updated_state.metrics]
+            return TaskScrollState(
+                task=task,
+                metrics=[
+                    *updated_state.metrics,
+                    *(
+                        old_metric
+                        for old_metric in old_state.metrics
+                        if old_metric.metric not in updated_metrics
+                    ),
+                ],
+            )
+
+        state.tasks = [
+            merge_with_updated_task_states(task_state, updated_task_states)
+            for task_state in state.tasks
+        ]
+
+    def _init_task_states(
+        self, companies: Mapping[str, str], task_metrics: Mapping[str, dict]
+    ) -> Sequence[TaskScrollState]:
+        """
+        Returned initialized metric scroll stated for the requested task metrics
+        """
+        with ThreadPoolExecutor(EventSettings.max_workers) as pool:
+            task_metric_states = pool.map(
+                partial(self._init_metric_states_for_task, companies=companies),
+                task_metrics.items(),
+            )
+
+        return [
+            TaskScrollState(task=task, metrics=metric_states,)
+            for task, metric_states in zip(task_metrics, task_metric_states)
+        ]
+
+    @abc.abstractmethod
+    def _get_extra_conditions(self) -> Sequence[dict]:
+        pass
+
+    @abc.abstractmethod
+    def _get_variant_state_aggs(
+        self,
+    ) -> Tuple[dict, Callable[[dict, VariantState], None]]:
+        pass
+
+    def _init_metric_states_for_task(
+        self, task_metrics: Tuple[str, dict], companies: Mapping[str, str]
+    ) -> Sequence[MetricState]:
+        """
+        Return metric scroll states for the task filled with the variant states
+        for the variants that reported any event_type events
+        """
+        task, metrics = task_metrics
+        company_id = companies[task]
+        must = [{"term": {"task": task}}, *self._get_extra_conditions()]
+        if metrics:
+            must.append(get_metric_variants_condition(metrics))
+        query = {"bool": {"must": must}}
+
+        search_args = dict(
+            es=self.es, company_id=company_id, event_type=self.event_type
+        )
+        max_metrics, max_variants = get_max_metric_and_variant_counts(
+            query=query, **search_args
+        )
+        max_variants = int(max_variants // 2)
+        variant_state_aggs, fill_variant_state_data = self._get_variant_state_aggs()
+        es_req: dict = {
+            "size": 0,
+            "query": query,
+            "aggs": {
+                "metrics": {
+                    "terms": {
+                        "field": "metric",
+                        "size": max_metrics,
+                        "order": {"_key": "asc"},
+                    },
+                    "aggs": {
+                        "last_event_timestamp": {"max": {"field": "timestamp"}},
+                        "variants": {
+                            "terms": {
+                                "field": "variant",
+                                "size": max_variants,
+                                "order": {"_key": "asc"},
+                            },
+                            **(
+                                {"aggs": variant_state_aggs}
+                                if variant_state_aggs
+                                else {}
+                            ),
+                        },
+                    },
+                }
+            },
+        }
+
+        with translate_errors_context():
+            es_res = search_company_events(body=es_req, **search_args)
+        if "aggregations" not in es_res:
+            return []
+
+        def init_variant_state(variant: dict):
+            """
+            Return new variant state for the passed variant bucket
+            """
+            state = VariantState(variant=variant["key"])
+            if fill_variant_state_data:
+                fill_variant_state_data(variant, state)
+
+            return state
+
+        return [
+            MetricState(
+                metric=metric["key"],
+                timestamp=nested_get(metric, ("last_event_timestamp", "value")),
+                variants=[
+                    init_variant_state(variant)
+                    for variant in nested_get(metric, ("variants", "buckets"))
+                ],
+            )
+            for metric in nested_get(es_res, ("aggregations", "metrics", "buckets"))
+        ]
+
+    @abc.abstractmethod
+    def _process_event(self, event: dict) -> dict:
+        pass
+
+    @abc.abstractmethod
+    def _get_same_variant_events_order(self) -> dict:
+        pass
+
+    def _get_task_metric_events(
+        self,
+        task_state: TaskScrollState,
+        companies: Mapping[str, str],
+        iter_count: int,
+        navigate_earlier: bool,
+        specific_variants_requested: bool,
+    ) -> Tuple:
+        """
+        Return task metric events grouped by iterations
+        Update task scroll state
+        """
+        if not task_state.metrics:
+            return task_state.task, []
+
+        if task_state.last_max_iter is None:
+            # the first fetch is always from the latest iteration to the earlier ones
+            navigate_earlier = True
+
+        must_conditions = [
+            {"term": {"task": task_state.task}},
+            {"terms": {"metric": [m.metric for m in task_state.metrics]}},
+            *self._get_extra_conditions(),
+        ]
+
+        range_condition = None
+        if navigate_earlier and task_state.last_min_iter is not None:
+            range_condition = {"lt": task_state.last_min_iter}
+        elif not navigate_earlier and task_state.last_max_iter is not None:
+            range_condition = {"gt": task_state.last_max_iter}
+        if range_condition:
+            must_conditions.append({"range": {"iter": range_condition}})
+
+        metrics_count = len(task_state.metrics)
+        max_variants = int(EventSettings.max_es_buckets / (metrics_count * iter_count))
+        es_req = {
+            "size": 0,
+            "query": {"bool": {"must": must_conditions}},
+            "aggs": {
+                "iters": {
+                    "terms": {
+                        "field": "iter",
+                        "size": iter_count,
+                        "order": {"_key": "desc" if navigate_earlier else "asc"},
+                    },
+                    "aggs": {
+                        "metrics": {
+                            "terms": {
+                                "field": "metric",
+                                "size": metrics_count,
+                                "order": {"_key": "asc"},
+                            },
+                            "aggs": {
+                                "variants": {
+                                    "terms": {
+                                        "field": "variant",
+                                        "size": max_variants,
+                                        "order": {"_key": "asc"},
+                                    },
+                                    "aggs": {
+                                        "events": {
+                                            "top_hits": {
+                                                "sort": self._get_same_variant_events_order(),
+                                                "size": 1,
+                                            }
+                                        }
+                                    },
+                                }
+                            },
+                        }
+                    },
+                }
+            },
+        }
+        with translate_errors_context():
+            es_res = search_company_events(
+                self.es,
+                company_id=companies[task_state.task],
+                event_type=self.event_type,
+                body=es_req,
+            )
+        if "aggregations" not in es_res:
+            return task_state.task, []
+
+        invalid_iterations = {
+            (m.metric, v.variant): v.last_invalid_iteration
+            for m in task_state.metrics
+            for v in m.variants
+        }
+        allow_uninitialized = (
+            False
+            if specific_variants_requested
+            else config.get(
+                "services.events.events_retrieval.debug_images.allow_uninitialized_variants",
+                False,
+            )
+        )
+
+        def is_valid_event(event: dict) -> bool:
+            key = event.get("metric"), event.get("variant")
+            if key not in invalid_iterations:
+                return allow_uninitialized
+
+            max_invalid = invalid_iterations[key]
+            return max_invalid is None or event.get("iter") > max_invalid
+
+        def get_iteration_events(it_: dict) -> Sequence:
+            return [
+                self._process_event(ev["_source"])
+                for m in nested_get(it_, ("metrics", "buckets"))
+                for v in nested_get(m, ("variants", "buckets"))
+                for ev in nested_get(v, ("events", "hits", "hits"))
+                if is_valid_event(ev["_source"])
+            ]
+
+        iterations = []
+        for it in nested_get(es_res, ("aggregations", "iters", "buckets")):
+            events = get_iteration_events(it)
+            if events:
+                iterations.append({"iter": it["key"], "events": events})
+
+        if not navigate_earlier:
+            iterations.sort(key=itemgetter("iter"), reverse=True)
+        if iterations:
+            task_state.last_max_iter = iterations[0]["iter"]
+            task_state.last_min_iter = iterations[-1]["iter"]
+
+        return task_state.task, iterations

apiserver/bll/event/metric_plots_iterator.py

@@ -0,0 +1,25 @@
+from typing import Sequence
+
+from elasticsearch import Elasticsearch
+from redis.client import StrictRedis
+
+from .event_common import EventType, uncompress_plot
+from .metric_events_iterator import MetricEventsIterator
+
+
+class MetricPlotsIterator(MetricEventsIterator):
+    def __init__(self, redis: StrictRedis, es: Elasticsearch):
+        super().__init__(redis, es, EventType.metrics_plot)
+
+    def _get_extra_conditions(self) -> Sequence[dict]:
+        return []
+
+    def _get_variant_state_aggs(self):
+        return None, None
+
+    def _process_event(self, event: dict) -> dict:
+        uncompress_plot(event)
+        return event
+
+    def _get_same_variant_events_order(self) -> dict:
+        return {"timestamp": {"order": "desc"}}

apiserver/bll/event/scalar_key.py

@@ -4,9 +4,11 @@ Module for polymorphism over different types of X axes in scalar aggregations
 from abc import ABC, abstractmethod
 from enum import auto
 
-from utilities.stringenum import StringEnum
-from bll.util import extract_properties_to_lists
-from config import config
+from typing import Any
+
+from apiserver.utilities import extract_properties_to_lists
+from apiserver.utilities.stringenum import StringEnum
+from apiserver.config_repo import config
 
 log = config.logger(__file__)
 
@@ -96,6 +98,10 @@ class ScalarKey(ABC):
         """
         return int(iter_data[self.bucket_key_key]), iter_data["avg_val"]["value"]
 
+    def cast_value(self, value: Any) -> Any:
+        """Cast value to appropriate type"""
+        return value
+
 
 class TimestampKey(ScalarKey):
     """
@@ -117,6 +123,9 @@ class TimestampKey(ScalarKey):
             }
         }
 
+    def cast_value(self, value: Any) -> int:
+        return int(value)
+
 
 class IterKey(ScalarKey):
     """
@@ -134,6 +143,9 @@ class IterKey(ScalarKey):
             }
         }
 
+    def cast_value(self, value: Any) -> int:
+        return int(value)
+
 
 class ISOTimeKey(ScalarKey):
     """

apiserver/bll/model/__init__.py

@@ -0,0 +1,237 @@
+from datetime import datetime
+from typing import Callable, Tuple, Sequence, Dict, Optional
+
+from mongoengine import Q
+
+from apiserver.apierrors import errors
+from apiserver.apimodels.models import ModelTaskPublishResponse
+from apiserver.bll.task.utils import deleted_prefix, get_last_metric_updates
+from apiserver.database.model import EntityVisibility
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task, TaskStatus
+from apiserver.service_repo.auth import Identity
+from .metadata import Metadata
+
+
+class ModelBLL:
+    @classmethod
+    def get_company_model_by_id(
+        cls, company_id: str, model_id: str, only_fields=None
+    ) -> Model:
+        query = dict(company=company_id, id=model_id)
+        qs = Model.objects(**query)
+        if only_fields:
+            qs = qs.only(*only_fields)
+        model = qs.first()
+        if not model:
+            raise errors.bad_request.InvalidModelId(**query)
+        return model
+
+    @staticmethod
+    def assert_exists(
+        company_id, model_ids, only=None, allow_public=False, return_models=True,
+    ) -> Optional[Sequence[Model]]:
+        model_ids = [model_ids] if isinstance(model_ids, str) else model_ids
+        ids = set(model_ids)
+        query = Q(id__in=ids)
+
+        q = Model.get_many(
+            company=company_id,
+            query=query,
+            allow_public=allow_public,
+            return_dicts=False,
+        )
+        if only:
+            q = q.only(*only)
+
+        if q.count() != len(ids):
+            raise errors.bad_request.InvalidModelId(ids=model_ids)
+
+        if return_models:
+            return list(q)
+
+    @classmethod
+    def publish_model(
+        cls,
+        model_id: str,
+        company_id: str,
+        identity: Identity,
+        force_publish_task: bool = False,
+        publish_task_func: Callable[[str, str, Identity, bool], dict] = None,
+    ) -> Tuple[int, ModelTaskPublishResponse]:
+        model = cls.get_company_model_by_id(company_id=company_id, model_id=model_id)
+        if model.ready:
+            raise errors.bad_request.ModelIsReady(company=company_id, model=model_id)
+
+        user_id = identity.user
+        published_task = None
+        if model.task and publish_task_func:
+            task = (
+                Task.objects(id=model.task, company=company_id)
+                .only("id", "status")
+                .first()
+            )
+            if task and task.status != TaskStatus.published:
+                task_publish_res = publish_task_func(
+                    model.task, company_id, identity, force_publish_task
+                )
+                published_task = ModelTaskPublishResponse(
+                    id=model.task, data=task_publish_res
+                )
+
+        now = datetime.utcnow()
+        updated = model.update(
+            upsert=False,
+            ready=True,
+            last_update=now,
+            last_change=now,
+            last_changed_by=user_id,
+        )
+        return updated, published_task
+
+    @classmethod
+    def delete_model(
+        cls, model_id: str, company_id: str, user_id: str, force: bool
+    ) -> Tuple[int, Model]:
+        model = cls.get_company_model_by_id(
+            company_id=company_id,
+            model_id=model_id,
+            only_fields=("id", "task", "project", "uri"),
+        )
+        deleted_model_id = f"{deleted_prefix}{model_id}"
+
+        using_tasks = Task.objects(models__input__model=model_id).only("id")
+        if using_tasks:
+            if not force:
+                raise errors.bad_request.ModelInUse(
+                    "as execution model, use force=True to delete",
+                    num_tasks=len(using_tasks),
+                )
+            # update deleted model id in using tasks
+            Task._get_collection().update_many(
+                filter={"_id": {"$in": [t.id for t in using_tasks]}},
+                update={"$set": {"models.input.$[elem].model": deleted_model_id}},
+                array_filters=[{"elem.model": model_id}],
+                upsert=False,
+            )
+
+        if model.task:
+            task = Task.objects(id=model.task).first()
+            if task:
+                now = datetime.utcnow()
+                if task.status == TaskStatus.published:
+                    if not force:
+                        raise errors.bad_request.ModelCreatingTaskExists(
+                            "and published, use force=True to delete", task=model.task
+                        )
+                    Task._get_collection().update_one(
+                        filter={"_id": model.task, "models.output.model": model_id},
+                        update={
+                            "$set": {
+                                "models.output.$[elem].model": deleted_model_id,
+                                "output.error": f"model deleted on {now.isoformat()}",
+                                "last_change": now,
+                                "last_changed_by": user_id,
+                            },
+                        },
+                        array_filters=[{"elem.model": model_id}],
+                        upsert=False,
+                    )
+                else:
+                    task.update(
+                        pull__models__output__model=model_id,
+                        set__last_change=now,
+                        set__last_changed_by=user_id,
+                    )
+
+        del_count = Model.objects(id=model_id, company=company_id).delete()
+        return del_count, model
+
+    @classmethod
+    def archive_model(cls, model_id: str, company_id: str, user_id: str):
+        cls.get_company_model_by_id(
+            company_id=company_id, model_id=model_id, only_fields=("id",)
+        )
+        now = datetime.utcnow()
+        archived = Model.objects(company=company_id, id=model_id).update(
+            add_to_set__system_tags=EntityVisibility.archived.value,
+            last_change=now,
+            last_changed_by=user_id,
+        )
+
+        return archived
+
+    @classmethod
+    def unarchive_model(cls, model_id: str, company_id: str, user_id: str):
+        cls.get_company_model_by_id(
+            company_id=company_id, model_id=model_id, only_fields=("id",)
+        )
+        now = datetime.utcnow()
+        unarchived = Model.objects(company=company_id, id=model_id).update(
+            pull__system_tags=EntityVisibility.archived.value,
+            last_change=now,
+            last_changed_by=user_id,
+        )
+
+        return unarchived
+
+    @classmethod
+    def get_model_stats(
+        cls, company: str, model_ids: Sequence[str],
+    ) -> Dict[str, dict]:
+        if not model_ids:
+            return {}
+
+        result = Model.aggregate(
+            [
+                {
+                    "$match": {
+                        "company": {"$in": ["", company]},
+                        "_id": {"$in": model_ids},
+                    }
+                },
+                {
+                    "$addFields": {
+                        "labels_count": {"$size": {"$objectToArray": "$labels"}}
+                    }
+                },
+                {"$project": {"labels_count": 1}},
+            ]
+        )
+        return {r.pop("_id"): r for r in result}
+
+    @staticmethod
+    def update_statistics(
+        company_id: str,
+        user_id: str,
+        model_id: str,
+        last_update: datetime = None,
+        last_iteration_max: int = None,
+        last_scalar_events: Dict[str, Dict[str, dict]] = None,
+    ):
+        last_update = last_update or datetime.utcnow()
+        updates = {
+            "last_update": datetime.utcnow(),
+            "last_change": last_update,
+            "last_changed_by": user_id,
+        }
+        if last_iteration_max is not None:
+            updates.update(max__last_iteration=last_iteration_max)
+
+        raw_updates = {}
+        if last_scalar_events is not None:
+            raw_updates = {}
+            if last_scalar_events is not None:
+                get_last_metric_updates(
+                    task_id=model_id,
+                    last_scalar_events=last_scalar_events,
+                    raw_updates=raw_updates,
+                    extra_updates=updates,
+                    model_events=True,
+                )
+
+        ret = Model.objects(id=model_id).update_one(**updates)
+        if ret and raw_updates:
+            Model.objects(id=model_id).update_one(__raw__=[{"$set": raw_updates}])
+
+        return ret

apiserver/bll/model/metadata.py

@@ -0,0 +1,107 @@
+from typing import Sequence, Union, Mapping
+
+from mongoengine import Document
+
+from apiserver.apierrors import errors
+from apiserver.apimodels.metadata import MetadataItem
+from apiserver.database.model.base import GetMixin
+from apiserver.utilities.parameter_key_escaper import (
+    ParameterKeyEscaper,
+    mongoengine_safe,
+)
+from apiserver.config_repo import config
+
+log = config.logger(__file__)
+
+
+class Metadata:
+    @staticmethod
+    def metadata_from_api(
+        api_data: Union[Mapping[str, MetadataItem], Sequence[MetadataItem]]
+    ) -> dict:
+        if not api_data:
+            return {}
+
+        if isinstance(api_data, dict):
+            return {
+                ParameterKeyEscaper.escape(k): v.to_struct()
+                for k, v in api_data.items()
+            }
+
+        return {
+            ParameterKeyEscaper.escape(item.key): item.to_struct() for item in api_data
+        }
+
+    @classmethod
+    def edit_metadata(
+        cls,
+        obj: Document,
+        items: Sequence[MetadataItem],
+        replace_metadata: bool,
+        **more_updates,
+    ) -> int:
+        update_cmds = dict()
+        metadata = cls.metadata_from_api(items)
+        if replace_metadata:
+            update_cmds["set__metadata"] = metadata
+        else:
+            for key, value in metadata.items():
+                update_cmds[f"set__metadata__{mongoengine_safe(key)}"] = value
+
+        return obj.update(**update_cmds, **more_updates)
+
+    @classmethod
+    def delete_metadata(cls, obj: Document, keys: Sequence[str], **more_updates) -> int:
+        return obj.update(
+            **{
+                f"unset__metadata__{ParameterKeyEscaper.escape(key)}": 1
+                for key in set(keys)
+            },
+            **more_updates,
+        )
+
+    @staticmethod
+    def _process_path(path: str):
+        """
+        Frontend does a partial escaping on the path so the all '.' in key names are escaped
+        Need to unescape and apply a full mongo escaping
+        """
+        parts = path.split(".")
+        if len(parts) < 2 or len(parts) > 3:
+            raise errors.bad_request.ValidationError("invalid field", path=path)
+        return ".".join(
+            ParameterKeyEscaper.escape(ParameterKeyEscaper.unescape(p)) for p in parts
+        )
+
+    @classmethod
+    def escape_paths(cls, paths: Sequence[str]) -> Sequence[str]:
+        for prefix in (
+            "metadata.",
+            "-metadata.",
+        ):
+            paths = [
+                cls._process_path(path) if path.startswith(prefix) else path
+                for path in paths
+            ]
+        return paths
+
+    @classmethod
+    def escape_query_parameters(cls, call_data: dict) -> dict:
+        if not call_data:
+            return call_data
+
+        keys = list(call_data)
+        call_data = {
+            safe_key: call_data[key]
+            for key, safe_key in zip(keys, Metadata.escape_paths(keys))
+        }
+
+        projection = GetMixin.get_projection(call_data)
+        if projection:
+            GetMixin.set_projection(call_data, Metadata.escape_paths(projection))
+
+        ordering = GetMixin.get_ordering(call_data)
+        if ordering:
+            GetMixin.set_ordering(call_data, Metadata.escape_paths(ordering))
+
+        return call_data

apiserver/bll/organization/__init__.py

@@ -0,0 +1,117 @@
+from collections import defaultdict
+from datetime import datetime
+from enum import Enum
+from typing import Sequence, Dict, Type
+
+from apiserver.apierrors import errors
+from apiserver.bll.util import update_project_time
+from apiserver.config_repo import config
+from apiserver.database.model.model import AttributedDocument
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task
+from apiserver.redis_manager import redman
+from .tags_cache import _TagsCache
+
+log = config.logger(__file__)
+
+
+class Tags(Enum):
+    Task = "task"
+    Model = "model"
+
+
+class OrgBLL:
+    def __init__(self, redis=None):
+        self.redis = redis or redman.connection("apiserver")
+        self._task_tags = _TagsCache(Task, self.redis)
+        self._model_tags = _TagsCache(Model, self.redis)
+
+    def edit_entity_tags(
+        self,
+        company_id,
+        user_id: str,
+        entity_cls: Type[AttributedDocument],
+        entity_ids: Sequence[str],
+        add_tags: Sequence[str],
+        remove_tags: Sequence[str],
+    ) -> int:
+        if entity_cls not in (Task, Model):
+            raise errors.bad_request.ValidationError(
+                "Tags editing can be called on tasks or models only"
+            )
+        if not entity_ids:
+            raise errors.bad_request.ValidationError(
+                "No entity ids provided for editing tags"
+            )
+        if not (add_tags or remove_tags):
+            raise errors.bad_request.ValidationError(
+                "Either add tags or remove tags should be provided"
+            )
+
+        updated = 0
+        last_changed = {
+            "set__last_change": datetime.utcnow(),
+            "set__last_changed_by": user_id,
+        }
+        if add_tags:
+            updated += entity_cls.objects(company=company_id, id__in=entity_ids).update(
+                add_to_set__tags=add_tags, **last_changed,
+            )
+        if remove_tags:
+            updated += entity_cls.objects(company=company_id, id__in=entity_ids).update(
+                pull_all__tags=remove_tags, **last_changed,
+            )
+        if not updated:
+            return 0
+
+        projects = entity_cls.objects(company=company_id, id__in=entity_ids).distinct(
+            "project"
+        )
+        update_project_time(project_ids=projects)
+        self.update_tags(
+            company_id,
+            entity=Tags.Task if entity_cls is Task else Tags.Model,
+            projects=projects,
+            tags=add_tags or remove_tags
+        )
+        return updated
+
+    def get_tags(
+        self,
+        company_id: str,
+        entity: Tags,
+        include_system: bool = False,
+        filter_: Dict[str, Sequence[str]] = None,
+        projects: Sequence[str] = None,
+    ) -> dict:
+        tags_cache = self._get_tags_cache_for_entity(entity)
+        if not projects:
+            return tags_cache.get_tags(
+                company_id, include_system=include_system, filter_=filter_
+            )
+
+        ret = defaultdict(set)
+        for project in projects:
+            project_tags = tags_cache.get_tags(
+                company_id,
+                include_system=include_system,
+                filter_=filter_,
+                project=project,
+            )
+            for field, tags in project_tags.items():
+                ret[field] |= tags
+
+        return ret
+
+    def update_tags(
+        self, company_id: str, entity: Tags, projects: Sequence[str], tags=None, system_tags=None,
+    ):
+        tags_cache = self._get_tags_cache_for_entity(entity)
+        tags_cache.update_tags(company_id, projects, tags, system_tags)
+
+    def reset_tags(self, company_id: str, entity: Tags, projects: Sequence[str]):
+        tags_cache = self._get_tags_cache_for_entity(entity)
+        tags_cache.reset_tags(company_id, projects=projects)
+
+    def _get_tags_cache_for_entity(self, entity: Tags) -> _TagsCache:
+        return self._task_tags if entity == Tags.Task else self._model_tags

apiserver/bll/organization/tags_cache.py

@@ -1,17 +1,14 @@
-from collections import defaultdict
-from enum import Enum
 from itertools import chain
 from typing import Sequence, Union, Type, Dict
 
 from mongoengine import Q
 from redis import Redis
 
-from config import config
-from database.model.base import GetMixin
-from database.model.model import Model
-from database.model.task.task import Task
-from redis_manager import redman
-from utilities import json
+from apiserver.config_repo import config
+from apiserver.bll.project import project_ids_with_children
+from apiserver.database.model.base import GetMixin
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task
 
 log = config.logger(__file__)
 _settings_prefix = "services.organization"
@@ -20,6 +17,8 @@ _settings_prefix = "services.organization"
 class _TagsCache:
     _tags_field = "tags"
     _system_tags_field = "system_tags"
+    _dummy_tag = "__dummy__"
+    # prepend our list in redis with this tag since empty lists are auto deleted
 
     def __init__(self, db_cls: Union[Type[Model], Type[Task]], redis: Redis):
         self.db_cls = db_cls
@@ -31,24 +30,26 @@ class _TagsCache:
 
     def _get_tags_from_db(
         self,
-        company: str,
+        company_id: str,
         field: str,
         project: str = None,
         filter_: Dict[str, Sequence[str]] = None,
     ) -> set:
-        query = Q(company=company)
+        query = Q(company=company_id)
         if filter_:
             for name, vals in filter_.items():
                 if vals:
                     query &= GetMixin.get_list_field_query(name, vals)
         if project:
-            query &= Q(project=project)
+            query &= Q(project__in=project_ids_with_children([project]))
+        # else:
+        #     query &= Q(system_tags__nin=[EntityVisibility.hidden.value])
 
         return self.db_cls.objects(query).distinct(field)
 
     def _get_tags_cache_key(
         self,
-        company: str,
+        company_id: str,
         field: str,
         project: str = None,
         filter_: Dict[str, Sequence[str]] = None,
@@ -64,12 +65,12 @@ class _TagsCache:
             filter_str = "_".join(
                 ["filter", *chain.from_iterable([f, *v] for f, v in filter_.items())]
             )
-        key_parts = [company, project, self.db_cls.__name__, field, filter_str]
+        key_parts = [field, company_id, project, self.db_cls.__name__, filter_str]
         return "_".join(filter(None, key_parts))
 
     def get_tags(
         self,
-        company: str,
+        company_id: str,
         include_system: bool = False,
         filter_: Dict[str, Sequence[str]] = None,
         project: str = None,
@@ -83,27 +84,29 @@ class _TagsCache:
         fields = [self._tags_field]
         if include_system:
             fields.append(self._system_tags_field)
-        redis_keys = [
-            self._get_tags_cache_key(company, field=f, project=project, filter_=filter_)
-            for f in fields
-        ]
-        cached = self.redis.mget(redis_keys)
+
         ret = {}
-        for field, tag_data, key in zip(fields, cached, redis_keys):
-            if tag_data is not None:
-                tags = json.loads(tag_data)
+        for field in fields:
+            redis_key = self._get_tags_cache_key(
+                company_id, field=field, project=project, filter_=filter_
+            )
+            cached_tags = self.redis.lrange(redis_key, 0, -1)
+            if cached_tags:
+                tags = [c.decode() for c in cached_tags[1:]]
             else:
-                tags = list(self._get_tags_from_db(company, field, project, filter_))
-                self.redis.setex(
-                    key,
-                    time=self._tags_cache_expiration_seconds,
-                    value=json.dumps(tags),
+                tags = list(
+                    self._get_tags_from_db(
+                        company_id, field=field, project=project, filter_=filter_
+                    )
                 )
+                self.redis.rpush(redis_key, self._dummy_tag, *tags)
+                self.redis.expire(redis_key, self._tags_cache_expiration_seconds)
+
             ret[field] = set(tags)
 
         return ret
 
-    def update_tags(self, company: str, project: str, tags=None, system_tags=None):
+    def update_tags(self, company_id: str, projects: Sequence[str], tags=None, system_tags=None):
         """
         Updates tags. If reset is set then both tags and system_tags
         are recalculated. Otherwise only those that are not 'None'
@@ -119,22 +122,22 @@ class _TagsCache:
         if not fields:
             return
 
-        self._delete_redis_keys(company, projects=[project], fields=fields)
+        self._delete_redis_keys(company_id, projects=projects, fields=fields)
 
-    def reset_tags(self, company: str, projects: Sequence[str]):
+    def reset_tags(self, company_id: str, projects: Sequence[str]):
         self._delete_redis_keys(
-            company,
+            company_id,
             projects=projects,
             fields=(self._tags_field, self._system_tags_field),
         )
 
     def _delete_redis_keys(
-        self, company: str, projects: [Sequence[str]], fields: Sequence[str]
+        self, company_id: str, projects: [Sequence[str]], fields: Sequence[str]
     ):
         redis_keys = list(
             chain.from_iterable(
                 self.redis.keys(
-                    self._get_tags_cache_key(company, field=f, project=p) + "*"
+                    self._get_tags_cache_key(company_id, field=f, project=p) + "*"
                 )
                 for f in fields
                 for p in set(projects) | {None}
@@ -142,52 +145,3 @@ class _TagsCache:
         )
         if redis_keys:
             self.redis.delete(*redis_keys)
-
-
-class Tags(Enum):
-    Task = "task"
-    Model = "model"
-
-
-class OrgBLL:
-    def __init__(self, redis=None):
-        self.redis = redis or redman.connection("apiserver")
-        self._task_tags = _TagsCache(Task, self.redis)
-        self._model_tags = _TagsCache(Model, self.redis)
-
-    def get_tags(
-        self,
-        company: str,
-        entity: Tags,
-        include_system: bool = False,
-        filter_: Dict[str, Sequence[str]] = None,
-        projects: Sequence[str] = None,
-    ) -> dict:
-        tags_cache = self._get_tags_cache_for_entity(entity)
-        if not projects:
-            return tags_cache.get_tags(
-                company, include_system=include_system, filter_=filter_
-            )
-
-        ret = defaultdict(set)
-        for project in projects:
-            project_tags = tags_cache.get_tags(
-                company, include_system=include_system, filter_=filter_, project=project
-            )
-            for field, tags in project_tags.items():
-                ret[field] |= tags
-
-        return ret
-
-    def update_tags(
-        self, company: str, entity: Tags, project: str, tags=None, system_tags=None,
-    ):
-        tags_cache = self._get_tags_cache_for_entity(entity)
-        tags_cache.update_tags(company, project, tags, system_tags)
-
-    def reset_tags(self, company: str, entity: Tags, projects: Sequence[str]):
-        tags_cache = self._get_tags_cache_for_entity(entity)
-        tags_cache.reset_tags(company, projects=projects)
-
-    def _get_tags_cache_for_entity(self, entity: Tags) -> _TagsCache:
-        return self._task_tags if entity == Tags.Task else self._model_tags

apiserver/bll/project/__init__.py

@@ -0,0 +1,3 @@
+from .project_bll import ProjectBLL
+from .project_queries import ProjectQueries
+from .sub_projects import _ids_with_children as project_ids_with_children

apiserver/bll/project/project_cleanup.py

@@ -0,0 +1,324 @@
+from collections import defaultdict
+from datetime import datetime
+from typing import Tuple, Set, Sequence
+
+import attr
+from mongoengine import Q
+
+from apiserver.apierrors import errors
+from apiserver.bll.event import EventBLL
+from apiserver.bll.task.task_cleanup import (
+    TaskUrls,
+    schedule_for_delete,
+    delete_task_events_and_collect_urls,
+)
+from apiserver.config_repo import config
+from apiserver.database.model import EntityVisibility
+from apiserver.database.model.model import Model
+from apiserver.database.model.project import Project
+from apiserver.database.model.task.task import Task, ArtifactModes, TaskType, TaskStatus
+from .project_bll import (
+    ProjectBLL,
+    pipeline_tag,
+    pipelines_project_name,
+    dataset_tag,
+    datasets_project_name,
+    reports_tag,
+)
+from .sub_projects import _ids_with_children
+
+log = config.logger(__file__)
+event_bll = EventBLL()
+
+
+@attr.s(auto_attribs=True)
+class DeleteProjectResult:
+    deleted: int = 0
+    disassociated_tasks: int = 0
+    deleted_models: int = 0
+    deleted_tasks: int = 0
+    urls: TaskUrls = None
+
+
+def _get_child_project_ids(
+    project_id: str,
+) -> Tuple[Sequence[str], Sequence[str], Sequence[str]]:
+    project_ids = _ids_with_children([project_id])
+    pipeline_ids = list(
+        Project.objects(
+            id__in=project_ids,
+            system_tags__in=[pipeline_tag],
+            basename__ne=pipelines_project_name,
+        ).scalar("id")
+    )
+    dataset_ids = list(
+        Project.objects(
+            id__in=project_ids,
+            system_tags__in=[dataset_tag],
+            basename__ne=datasets_project_name,
+        ).scalar("id")
+    )
+    return project_ids, pipeline_ids, dataset_ids
+
+
+def validate_project_delete(company: str, project_id: str):
+    project = Project.get_for_writing(
+        company=company, id=project_id, _only=("id", "path", "system_tags")
+    )
+    if not project:
+        raise errors.bad_request.InvalidProjectId(id=project_id)
+
+    project_ids, pipeline_ids, dataset_ids = _get_child_project_ids(project_id)
+    ret = {}
+    if pipeline_ids:
+        pipelines_with_active_controllers = Task.objects(
+            project__in=pipeline_ids,
+            type=TaskType.controller,
+            system_tags__nin=[EntityVisibility.archived.value],
+        ).distinct("project")
+        ret["pipelines"] = len(pipelines_with_active_controllers)
+    else:
+        ret["pipelines"] = 0
+    if dataset_ids:
+        datasets_with_data = Task.objects(
+            project__in=dataset_ids,
+            system_tags__nin=[EntityVisibility.archived.value],
+        ).distinct("project")
+        ret["datasets"] = len(datasets_with_data)
+    else:
+        ret["datasets"] = 0
+
+    project_ids = list(set(project_ids) - set(pipeline_ids) - set(dataset_ids))
+    if project_ids:
+        in_project_query = Q(project__in=project_ids)
+        for cls in (Task, Model):
+            query = (
+                in_project_query & Q(system_tags__nin=[reports_tag])
+                if cls is Task
+                else in_project_query
+            )
+            ret[f"{cls.__name__.lower()}s"] = cls.objects(query).count()
+            ret[f"non_archived_{cls.__name__.lower()}s"] = cls.objects(
+                query & Q(system_tags__nin=[EntityVisibility.archived.value])
+            ).count()
+        ret["reports"] = Task.objects(
+            in_project_query & Q(system_tags__in=[reports_tag])
+        ).count()
+        ret["non_archived_reports"] = Task.objects(
+            in_project_query
+            & Q(
+                system_tags__in=[reports_tag],
+                system_tags__nin=[EntityVisibility.archived.value],
+            )
+        ).count()
+    else:
+        for cls in (Task, Model):
+            ret[f"{cls.__name__.lower()}s"] = 0
+            ret[f"non_archived_{cls.__name__.lower()}s"] = 0
+        ret["reports"] = 0
+        ret["non_archived_reports"] = 0
+
+    return ret
+
+
+def delete_project(
+    company: str,
+    user: str,
+    project_id: str,
+    force: bool,
+    delete_contents: bool,
+    delete_external_artifacts: bool,
+) -> Tuple[DeleteProjectResult, Set[str]]:
+    project = Project.get_for_writing(
+        company=company, id=project_id, _only=("id", "path", "system_tags")
+    )
+    if not project:
+        raise errors.bad_request.InvalidProjectId(id=project_id)
+
+    delete_external_artifacts = delete_external_artifacts and config.get(
+        "services.async_urls_delete.enabled", True
+    )
+    project_ids, pipeline_ids, dataset_ids = _get_child_project_ids(project_id)
+    if not force:
+        if pipeline_ids:
+            active_controllers = Task.objects(
+                project__in=pipeline_ids,
+                type=TaskType.controller,
+                system_tags__nin=[EntityVisibility.archived.value],
+            ).only("id")
+            if active_controllers:
+                raise errors.bad_request.ProjectHasPipelines(
+                    "please archive all the controllers or use force=true",
+                    id=project_id,
+                )
+        if dataset_ids:
+            datasets_with_data = Task.objects(
+                project__in=dataset_ids,
+                system_tags__nin=[EntityVisibility.archived.value],
+            ).only("id")
+            if datasets_with_data:
+                raise errors.bad_request.ProjectHasDatasets(
+                    "please delete all the dataset versions or use force=true",
+                    id=project_id,
+                )
+
+        regular_projects = list(set(project_ids) - set(pipeline_ids) - set(dataset_ids))
+        if regular_projects:
+            for cls, error in (
+                (Task, errors.bad_request.ProjectHasTasks),
+                (Model, errors.bad_request.ProjectHasModels),
+            ):
+                non_archived = cls.objects(
+                    project__in=regular_projects,
+                    system_tags__nin=[EntityVisibility.archived.value],
+                ).only("id")
+                if non_archived:
+                    raise error("use force=true", id=project_id)
+
+    if not delete_contents:
+        disassociated = defaultdict(int)
+        for cls in ProjectBLL.child_classes:
+            disassociated[cls] = cls.objects(project__in=project_ids).update(
+                project=None
+            )
+        res = DeleteProjectResult(disassociated_tasks=disassociated[Task])
+    else:
+        deleted_models, model_event_urls, model_urls = _delete_models(
+            company=company, user=user, projects=project_ids
+        )
+        deleted_tasks, task_event_urls, artifact_urls = _delete_tasks(
+            company=company, user=user, projects=project_ids
+        )
+        event_urls = task_event_urls | model_event_urls
+        if delete_external_artifacts:
+            scheduled = schedule_for_delete(
+                task_id=project_id,
+                company=company,
+                user=user,
+                urls=event_urls | model_urls | artifact_urls,
+                can_delete_folders=True,
+            )
+            for urls in (event_urls, model_urls, artifact_urls):
+                urls.difference_update(scheduled)
+        res = DeleteProjectResult(
+            deleted_tasks=deleted_tasks,
+            deleted_models=deleted_models,
+            urls=TaskUrls(
+                model_urls=list(model_urls),
+                artifact_urls=list(artifact_urls),
+            ),
+        )
+
+    affected = {*project_ids, *(project.path or [])}
+    res.deleted = Project.objects(id__in=project_ids).delete()
+
+    return res, affected
+
+
+def _delete_tasks(
+    company: str, user: str, projects: Sequence[str]
+) -> Tuple[int, Set, Set]:
+    """
+    Delete only the task themselves and their non published version.
+    Child models under the same project are deleted separately.
+    Children tasks should be deleted in the same api call.
+    If any child entities are left in another projects then updated their parent task to None
+    """
+    tasks = Task.objects(project__in=projects).only("id", "execution__artifacts")
+    if not tasks:
+        return 0, set(), set()
+
+    task_ids = list({t.id for t in tasks})
+    now = datetime.utcnow()
+    Task.objects(parent__in=task_ids, project__nin=projects).update(
+        parent=None,
+        last_change=now,
+        last_changed_by=user,
+    )
+    Model.objects(task__in=task_ids, project__nin=projects).update(
+        task=None,
+        last_change=now,
+        last_changed_by=user,
+    )
+
+    artifact_urls = set()
+    for task in tasks:
+        if task.execution and task.execution.artifacts:
+            artifact_urls.update(
+                {
+                    a.uri
+                    for a in task.execution.artifacts.values()
+                    if a.mode == ArtifactModes.output and a.uri
+                }
+            )
+
+    event_urls = delete_task_events_and_collect_urls(
+        company=company, task_ids=task_ids, wait_for_delete=False
+    )
+    deleted = tasks.delete()
+
+    return deleted, event_urls, artifact_urls
+
+
+def _delete_models(
+    company: str, user: str, projects: Sequence[str]
+) -> Tuple[int, Set[str], Set[str]]:
+    """
+    Delete project models and update the tasks from other projects
+    that reference them to reference None.
+    """
+    models = Model.objects(project__in=projects).only("task", "id", "uri")
+    if not models:
+        return 0, set(), set()
+
+    model_ids = list({m.id for m in models})
+    deleted = "__DELETED__"
+    Task._get_collection().update_many(
+        filter={
+            "project": {"$nin": projects},
+            "models.input.model": {"$in": model_ids},
+        },
+        update={"$set": {"models.input.$[elem].model": deleted}},
+        array_filters=[{"elem.model": {"$in": model_ids}}],
+        upsert=False,
+    )
+
+    model_tasks = list({m.task for m in models if m.task})
+    if model_tasks:
+        now = datetime.utcnow()
+        # update published tasks
+        Task._get_collection().update_many(
+            filter={
+                "_id": {"$in": model_tasks},
+                "project": {"$nin": projects},
+                "models.output.model": {"$in": model_ids},
+                "status": TaskStatus.published,
+            },
+            update={
+                "$set": {
+                    "models.output.$[elem].model": deleted,
+                    "last_change": now,
+                    "last_changed_by": user,
+                }
+            },
+            array_filters=[{"elem.model": {"$in": model_ids}}],
+            upsert=False,
+        )
+        # update unpublished tasks
+        Task.objects(
+            id__in=model_tasks,
+            project__nin=projects,
+            status__ne=TaskStatus.published,
+        ).update(
+            pull__models__output__model__in=model_ids,
+            set__last_change=now,
+            set__last_changed_by=user,
+        )
+
+    model_urls = {m.uri for m in models if m.uri}
+    event_urls = delete_task_events_and_collect_urls(
+        company=company, task_ids=model_ids, model=True, wait_for_delete=False
+    )
+    deleted = models.delete()
+
+    return deleted, event_urls, model_urls

apiserver/bll/project/project_queries.py

@@ -0,0 +1,407 @@
+import json
+from collections import OrderedDict
+from datetime import datetime
+from typing import (
+    Sequence,
+    Optional,
+    Tuple,
+)
+
+from redis import StrictRedis
+
+from apiserver.config_repo import config
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task
+from apiserver.redis_manager import redman
+from apiserver.utilities.dicts import nested_get
+from apiserver.utilities.parameter_key_escaper import ParameterKeyEscaper
+from .sub_projects import _ids_with_children
+
+log = config.logger(__file__)
+
+
+class ProjectQueries:
+    def __init__(self, redis=None):
+        self.redis: StrictRedis = redis or redman.connection("apiserver")
+
+    @staticmethod
+    def _get_project_constraint(
+        project_ids: Sequence[str], include_subprojects: bool
+    ) -> dict:
+        """
+        If passed projects is None means top level projects
+        If passed projects is empty means no project filtering
+        """
+        if include_subprojects:
+            if not project_ids:
+                return {}
+            project_ids = _ids_with_children(project_ids)
+
+        if project_ids is None:
+            project_ids = [None]
+        if not project_ids:
+            return {}
+
+        return {"project": {"$in": project_ids}}
+
+    @staticmethod
+    def _get_company_constraint(company_id: str, allow_public: bool = True) -> dict:
+        if allow_public:
+            return {"company": {"$in": ["", company_id]}}
+
+        return {"company": company_id}
+
+    @classmethod
+    def get_aggregated_project_parameters(
+        cls,
+        company_id,
+        project_ids: Sequence[str],
+        include_subprojects: bool,
+        page: int = 0,
+        page_size: int = 500,
+    ) -> Tuple[int, int, Sequence[dict]]:
+        page = max(0, page)
+        page_size = max(1, page_size)
+        pipeline = [
+            {
+                "$match": {
+                    **cls._get_company_constraint(company_id),
+                    **cls._get_project_constraint(project_ids, include_subprojects),
+                    "hyperparams": {"$exists": True, "$gt": {}},
+                }
+            },
+            {"$project": {"sections": {"$objectToArray": "$hyperparams"}}},
+            {"$unwind": "$sections"},
+            {
+                "$project": {
+                    "section": "$sections.k",
+                    "names": {"$objectToArray": "$sections.v"},
+                }
+            },
+            {"$unwind": "$names"},
+            {"$group": {"_id": {"section": "$section", "name": "$names.k"}}},
+            {"$sort": OrderedDict({"_id.section": 1, "_id.name": 1})},
+            {"$skip": page * page_size},
+            {"$limit": page_size},
+            {
+                "$group": {
+                    "_id": 1,
+                    "total": {"$sum": 1},
+                    "results": {"$push": "$$ROOT"},
+                }
+            },
+        ]
+
+        result = next(Task.aggregate(pipeline), None)
+
+        total = 0
+        remaining = 0
+        results = []
+
+        if result:
+            total = int(result.get("total", -1))
+            results = [
+                {
+                    "section": ParameterKeyEscaper.unescape(
+                        nested_get(r, ("_id", "section"))
+                    ),
+                    "name": ParameterKeyEscaper.unescape(
+                        nested_get(r, ("_id", "name"))
+                    ),
+                }
+                for r in result.get("results", [])
+            ]
+            remaining = max(0, total - (len(results) + page * page_size))
+
+        return total, remaining, results
+
+    ParamValues = Tuple[int, Sequence[str]]
+
+    def _get_cached_param_values(
+        self, key: str, last_update: datetime, allowed_delta_sec=0
+    ) -> Optional[ParamValues]:
+        try:
+            cached = self.redis.get(key)
+            if not cached:
+                return
+
+            data = json.loads(cached)
+            cached_last_update = datetime.fromtimestamp(data["last_update"])
+            if (last_update - cached_last_update).total_seconds() <= allowed_delta_sec:
+                return data["total"], data["values"]
+        except Exception as ex:
+            log.error(f"Error retrieving params cached values: {str(ex)}")
+
+    def get_task_hyperparam_distinct_values(
+        self,
+        company_id: str,
+        project_ids: Sequence[str],
+        section: str,
+        name: str,
+        include_subprojects: bool,
+        allow_public: bool = True,
+        pattern: str = None,
+        page: int = 0,
+        page_size: int = 500,
+    ) -> ParamValues:
+        page = max(0, page)
+        page_size = max(1, page_size)
+        company_constraint = self._get_company_constraint(company_id, allow_public)
+        project_constraint = self._get_project_constraint(
+            project_ids, include_subprojects
+        )
+        key_path = f"hyperparams.{ParameterKeyEscaper.escape(section)}.{ParameterKeyEscaper.escape(name)}"
+        last_updated_task = (
+            Task.objects(
+                **company_constraint,
+                **project_constraint,
+                **{f"{key_path.replace('.', '__')}__exists": True},
+            )
+            .only("last_update")
+            .order_by("-last_update")
+            .limit(1)
+            .first()
+        )
+        if not last_updated_task:
+            return 0, []
+
+        redis_key = "_".join(
+            str(part)
+            for part in (
+                "hyperparam_values",
+                company_id,
+                "_".join(project_ids),
+                section,
+                name,
+                allow_public,
+                pattern,
+                page,
+                page_size,
+            )
+        )
+        last_update = last_updated_task.last_update or datetime.utcnow()
+        cached_res = self._get_cached_param_values(
+            key=redis_key,
+            last_update=last_update,
+            allowed_delta_sec=config.get(
+                "services.tasks.hyperparam_values.cache_allowed_outdate_sec", 60
+            ),
+        )
+        if cached_res:
+            return cached_res
+
+        match_condition = {
+            **company_constraint,
+            **project_constraint,
+            key_path: {"$exists": True},
+        }
+        if pattern:
+            match_condition["$expr"] = {
+                "$regexMatch": {
+                    "input": f"${key_path}.value",
+                    "regex": pattern,
+                    "options": "i",
+                }
+            }
+
+        pipeline = [
+            {"$match": match_condition},
+            {"$project": {"value": f"${key_path}.value"}},
+            {"$group": {"_id": "$value"}},
+            {"$sort": {"_id": 1}},
+            {"$skip": page * page_size},
+            {"$limit": page_size},
+            {
+                "$group": {
+                    "_id": 1,
+                    "total": {"$sum": 1},
+                    "results": {"$push": "$$ROOT._id"},
+                }
+            },
+        ]
+
+        result = next(Task.aggregate(pipeline, collation=Task._numeric_locale), None)
+        if not result:
+            return 0, []
+
+        total = int(result.get("total", 0))
+        values = result.get("results", [])
+
+        ttl = config.get("services.tasks.hyperparam_values.cache_ttl_sec", 86400)
+        cached = dict(last_update=last_update.timestamp(), total=total, values=values)
+        self.redis.setex(redis_key, ttl, json.dumps(cached))
+
+        return total, values
+
+    @classmethod
+    def get_unique_metric_variants(
+        cls,
+        company_id,
+        project_ids: Sequence[str],
+        include_subprojects: bool,
+        ids: Sequence[str],
+        model_metrics: bool = False,
+    ):
+        pipeline = [
+            {
+                "$match": {
+                    **cls._get_company_constraint(company_id),
+                    **cls._get_project_constraint(project_ids, include_subprojects),
+                    **({"_id": {"$in": ids}} if ids else {}),
+                }
+            },
+            {"$project": {"metrics": {"$objectToArray": "$last_metrics"}}},
+            {"$unwind": "$metrics"},
+            {
+                "$project": {
+                    "metric": "$metrics.k",
+                    "variants": {"$objectToArray": "$metrics.v"},
+                }
+            },
+            {"$unwind": "$variants"},
+            {
+                "$group": {
+                    "_id": {
+                        "metric": "$variants.v.metric",
+                        "variant": "$variants.v.variant",
+                    },
+                    "metrics": {
+                        "$addToSet": {
+                            "metric": "$variants.v.metric",
+                            "metric_hash": "$metric",
+                            "variant": "$variants.v.variant",
+                            "variant_hash": "$variants.k",
+                        }
+                    },
+                }
+            },
+            {"$sort": OrderedDict({"_id.metric": 1, "_id.variant": 1})},
+        ]
+
+        entity_cls = Model if model_metrics else Task
+        result = entity_cls.aggregate(pipeline)
+        return [r["metrics"][0] for r in result]
+
+    @classmethod
+    def get_model_metadata_keys(
+        cls,
+        company_id,
+        project_ids: Sequence[str],
+        include_subprojects: bool,
+        page: int = 0,
+        page_size: int = 500,
+    ) -> Tuple[int, int, Sequence[dict]]:
+        page = max(0, page)
+        page_size = max(1, page_size)
+        pipeline = [
+            {
+                "$match": {
+                    **cls._get_company_constraint(company_id),
+                    **cls._get_project_constraint(project_ids, include_subprojects),
+                    "metadata": {"$exists": True, "$gt": {}},
+                }
+            },
+            {"$project": {"metadata": {"$objectToArray": "$metadata"}}},
+            {"$unwind": "$metadata"},
+            {"$group": {"_id": "$metadata.k"}},
+            {"$sort": {"_id": 1}},
+            {"$skip": page * page_size},
+            {"$limit": page_size},
+            {
+                "$group": {
+                    "_id": 1,
+                    "total": {"$sum": 1},
+                    "results": {"$push": "$$ROOT"},
+                }
+            },
+        ]
+
+        result = next(Model.aggregate(pipeline), None)
+
+        total = 0
+        remaining = 0
+        results = []
+
+        if result:
+            total = int(result.get("total", -1))
+            results = [
+                ParameterKeyEscaper.unescape(r.get("_id"))
+                for r in result.get("results", [])
+            ]
+            remaining = max(0, total - (len(results) + page * page_size))
+
+        return total, remaining, results
+
+    def get_model_metadata_distinct_values(
+        self,
+        company_id: str,
+        project_ids: Sequence[str],
+        key: str,
+        include_subprojects: bool,
+        allow_public: bool = True,
+        page: int = 0,
+        page_size: int = 500,
+    ) -> ParamValues:
+        page = max(0, page)
+        page_size = max(1, page_size)
+        company_constraint = self._get_company_constraint(company_id, allow_public)
+        project_constraint = self._get_project_constraint(
+            project_ids, include_subprojects
+        )
+        key_path = f"metadata.{ParameterKeyEscaper.escape(key)}"
+        last_updated_model = (
+            Model.objects(
+                **company_constraint,
+                **project_constraint,
+                **{f"{key_path.replace('.', '__')}__exists": True},
+            )
+            .only("last_update")
+            .order_by("-last_update")
+            .limit(1)
+            .first()
+        )
+        if not last_updated_model:
+            return 0, []
+
+        redis_key = f"modelmetadata_values_{company_id}_{'_'.join(project_ids)}_{key}_{allow_public}_{page}_{page_size}"
+        last_update = last_updated_model.last_update or datetime.utcnow()
+        cached_res = self._get_cached_param_values(
+            key=redis_key, last_update=last_update
+        )
+        if cached_res:
+            return cached_res
+
+        pipeline = [
+            {
+                "$match": {
+                    **company_constraint,
+                    **project_constraint,
+                    key_path: {"$exists": True},
+                }
+            },
+            {"$project": {"value": f"${key_path}.value"}},
+            {"$group": {"_id": "$value"}},
+            {"$sort": {"_id": 1}},
+            {"$skip": page * page_size},
+            {"$limit": page_size},
+            {
+                "$group": {
+                    "_id": 1,
+                    "total": {"$sum": 1},
+                    "results": {"$push": "$$ROOT._id"},
+                }
+            },
+        ]
+
+        result = next(Model.aggregate(pipeline, collation=Model._numeric_locale), None)
+        if not result:
+            return 0, []
+
+        total = int(result.get("total", 0))
+        values = result.get("results", [])
+
+        ttl = config.get("services.models.metadata_values.cache_ttl_sec", 86400)
+        cached = dict(last_update=last_update.timestamp(), total=total, values=values)
+        self.redis.setex(redis_key, ttl, json.dumps(cached))
+
+        return total, values

apiserver/bll/project/sub_projects.py

@@ -0,0 +1,211 @@
+import itertools
+from datetime import datetime
+from typing import Tuple, Optional, Sequence, Mapping
+
+from boltons.iterutils import first
+
+from apiserver import database
+from apiserver.apierrors import errors
+from apiserver.database.model import EntityVisibility
+from apiserver.database.model.project import Project
+
+name_separator = "/"
+
+
+def _get_project_depth(project_name: str) -> int:
+    return len(list(filter(None, project_name.split(name_separator))))
+
+
+def _validate_project_name(project_name: str, raise_if_empty=True) -> Tuple[str, str]:
+    """
+    Remove redundant '/' characters. Ensure that the project name is not empty
+    Return the cleaned up project name and location
+    """
+    name_parts = [p.strip() for p in project_name.split(name_separator) if p]
+    if not name_parts:
+        if raise_if_empty:
+            raise errors.bad_request.InvalidProjectName(name=project_name)
+        return "", ""
+
+    return name_separator.join(name_parts), name_separator.join(name_parts[:-1])
+
+
+def _ensure_project(
+    company: str, user: str, name: str, creation_params: dict = None
+) -> Optional[Project]:
+    """
+    Makes sure that the project with the given name exists
+    If needed auto-create the project and all the missing projects in the path to it
+    Return the project
+    """
+    name, location = _validate_project_name(name, raise_if_empty=False)
+    if not name:
+        return None
+
+    project = _get_writable_project_from_name(company, name)
+    if project:
+        return project
+
+    now = datetime.utcnow()
+    project = Project(
+        id=database.utils.id(),
+        user=user,
+        company=company,
+        created=now,
+        last_update=now,
+        name=name,
+        basename=name.split("/")[-1],
+        **(creation_params or dict(description="")),
+    )
+    parent = _ensure_project(company, user, location, creation_params=creation_params)
+    _save_under_parent(project=project, parent=parent)
+    if parent:
+        parent.update(last_update=now)
+
+    return project
+
+
+def _save_under_parent(project: Project, parent: Optional[Project]):
+    """
+    Save the project under the given parent project or top level (parent=None)
+    Check that the project location matches the parent name
+    """
+    location, _, _ = project.name.rpartition(name_separator)
+    if not parent:
+        if location:
+            raise ValueError(
+                f"Project location {location} does not match empty parent name"
+            )
+        project.parent = None
+        project.path = []
+        project.save()
+        return
+
+    if location != parent.name:
+        raise ValueError(
+            f"Project location {location} does not match parent name {parent.name}"
+        )
+    project.parent = parent.id
+    project.path = [*(parent.path or []), parent.id]
+    project.save()
+
+
+def _get_writable_project_from_name(
+    company,
+    name,
+    _only: Optional[Sequence[str]] = ("id", "name", "path", "company", "parent"),
+) -> Optional[Project]:
+    """
+    Return a project from name. If the project not found then return None
+    """
+    qs = Project.objects(company__in=[company, ""], name=name)
+    if _only:
+        if "company" not in _only:
+            _only = ["company", *_only]
+        qs = qs.only(*_only)
+    projects = list(qs)
+
+    if not projects:
+        return
+
+    project = first(p for p in projects if p.company == company)
+    if not project:
+        raise errors.bad_request.PublicProjectExists(name=name)
+
+    return project
+
+
+ProjectsChildren = Mapping[str, Sequence[Project]]
+
+
+def _get_sub_projects(
+    project_ids: Sequence[str],
+    _only: Sequence[str] = ("id", "path"),
+    search_hidden=True,
+    allowed_ids: Sequence[str] = None,
+) -> ProjectsChildren:
+    """
+    Return the list of child projects of all the levels for the parent project ids
+    """
+    query = dict(path__in=project_ids)
+    if not search_hidden:
+        query["system_tags__nin"] = [EntityVisibility.hidden.value]
+    if allowed_ids:
+        query["id__in"] = allowed_ids
+
+    qs = Project.objects(**query)
+    if _only:
+        _only = set(_only) | {"path"}
+        qs = qs.only(*_only)
+    subprojects = list(qs)
+
+    return {
+        pid: [s for s in subprojects if pid in (s.path or [])] for pid in project_ids
+    }
+
+
+def _ids_with_parents(project_ids: Sequence[str]) -> Sequence[str]:
+    """
+    Return project ids with all the parent projects
+    """
+    projects = Project.objects(id__in=project_ids).only("id", "path")
+    parent_ids = set(itertools.chain.from_iterable(p.path for p in projects if p.path))
+    return list({*(p.id for p in projects), *parent_ids})
+
+
+def _ids_with_children(project_ids: Sequence[str]) -> Sequence[str]:
+    """
+    Return project ids with the ids of all the subprojects
+    """
+    children_ids = Project.objects(path__in=project_ids).scalar("id")
+    return list({*project_ids, *children_ids})
+
+
+def _update_subproject_names(
+    project: Project,
+    children: Sequence[Project],
+    old_name: str,
+    update_path: bool = False,
+    old_path: Sequence[str] = None,
+) -> int:
+    """
+    Update sub project names when the base project name changes
+    Optionally update the paths
+    """
+    updated = 0
+    now = datetime.utcnow()
+    for child in children:
+        child_suffix = name_separator.join(
+            child.name.split(name_separator)[len(old_name.split(name_separator)):]
+        )
+        updates = {
+            "name": name_separator.join((project.name, child_suffix)),
+            "last_update": now,
+        }
+        if update_path:
+            updates["path"] = project.path + child.path[len(old_path):]
+        updated += child.update(upsert=False, **updates)
+
+    return updated
+
+
+def _reposition_project_with_children(
+    project: Project, children: Sequence[Project], parent: Project
+) -> int:
+    new_location = parent.name if parent else None
+    old_name = project.name
+    old_path = project.path
+    project.name = name_separator.join(
+        filter(None, (new_location, project.name.split(name_separator)[-1]))
+    )
+    project.last_update = datetime.utcnow()
+    _save_under_parent(project, parent=parent)
+
+    moved = 1 + _update_subproject_names(
+        project=project,
+        children=children,
+        old_name=old_name,
+        update_path=True,
+        old_path=old_path,
+    )
+    return moved

apiserver/bll/query/builder.py

@@ -0,0 +1,51 @@
+from typing import Optional, Sequence, Iterable, Union
+
+from apiserver.config_repo import config
+
+log = config.logger(__file__)
+
+RANGE_IGNORE_VALUE = -1
+
+
+class Builder:
+    @staticmethod
+    def dates_range(
+        from_date: Optional[Union[int, float]] = None,
+        to_date: Optional[Union[int, float]] = None,
+    ) -> dict:
+        assert (
+            from_date or to_date
+        ), "range condition requires that at least one of from_date or to_date specified"
+        conditions = {}
+        if from_date:
+            conditions["gte"] = int(from_date)
+        if to_date:
+            conditions["lte"] = int(to_date)
+        return {
+            "range": {
+                "timestamp": {
+                    **conditions,
+                    "format": "epoch_second",
+                }
+            }
+        }
+
+    @staticmethod
+    def terms(field: str, values: Iterable) -> dict:
+        if isinstance(values, str):
+            assert not isinstance(values, str), "apparently 'term' should be used here"
+        return {"terms": {field: list(values)}}
+    @staticmethod
+    def term(field: str, value) -> dict:
+        return {"term": {field: value}}
+
+    @staticmethod
+    def normalize_range(
+        range_: Sequence[Union[int, float]],
+        ignore_value: Union[int, float] = RANGE_IGNORE_VALUE,
+    ) -> Optional[Sequence[Union[int, float]]]:
+        if not range_ or set(range_) == {ignore_value}:
+            return None
+        if len(range_) < 2:
+            return [range_[0]] * 2
+        return range_

apiserver/bll/queue/queue_bll.py

@@ -0,0 +1,541 @@
+from collections import defaultdict
+from datetime import datetime
+from typing import Sequence, Optional, Tuple, Union, Iterable
+
+from elasticsearch import Elasticsearch
+from mongoengine import Q
+
+from apiserver import database
+from apiserver.database.model.task.task import Task, TaskStatus
+from apiserver.es_factory import es_factory
+from apiserver.apierrors import errors
+from apiserver.bll.queue.queue_metrics import QueueMetrics
+from apiserver.bll.workers import WorkerBLL
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.queue import Queue, Entry
+
+log = config.logger(__file__)
+MOVE_FIRST = "first"
+MOVE_LAST = "last"
+
+
+class QueueBLL(object):
+    def __init__(self, worker_bll: WorkerBLL = None, es: Elasticsearch = None):
+        self.worker_bll = worker_bll or WorkerBLL()
+        self.es = es or es_factory.connect("workers")
+        self._metrics = QueueMetrics(self.es)
+
+    @property
+    def metrics(self) -> QueueMetrics:
+        return self._metrics
+
+    @staticmethod
+    def create(
+        company_id: str,
+        name: str,
+        display_name: str = None,
+        tags: Optional[Sequence[str]] = None,
+        system_tags: Optional[Sequence[str]] = None,
+        metadata: Optional[dict] = None,
+    ) -> Queue:
+        """Creates a queue"""
+        with translate_errors_context():
+            now = datetime.utcnow()
+            queue = Queue(
+                id=database.utils.id(),
+                company=company_id,
+                created=now,
+                name=name,
+                display_name=display_name,
+                tags=tags or [],
+                system_tags=system_tags or [],
+                metadata=metadata,
+                last_update=now,
+            )
+            queue.save()
+            return queue
+
+    def get_by_name(
+        self, company_id: str, queue_name: str, only: Optional[Sequence[str]] = None,
+    ) -> Queue:
+        qs = Queue.objects(name=queue_name, company=company_id)
+        if only:
+            qs = qs.only(*only)
+
+        return qs.first()
+
+    @staticmethod
+    def _get_task_entries_projection(max_task_entries: int) -> dict:
+        return dict(slice__entries=max_task_entries)
+
+    def get_by_id(
+        self,
+        company_id: str,
+        queue_id: str,
+        only: Optional[Sequence[str]] = None,
+        max_task_entries: int = None,
+    ) -> Queue:
+        """
+        Get queue by id
+        :raise errors.bad_request.InvalidQueueId: if the queue is not found
+        """
+        with translate_errors_context():
+            query = dict(id=queue_id, company=company_id)
+            qs = Queue.objects(**query)
+            if only:
+                qs = qs.only(*only)
+            if max_task_entries:
+                qs = qs.fields(**self._get_task_entries_projection(max_task_entries))
+            queue = qs.first()
+            if not queue:
+                raise errors.bad_request.InvalidQueueId(**query)
+
+            return queue
+
+    @classmethod
+    def get_queue_with_task(cls, company_id: str, queue_id: str, task_id: str) -> Queue:
+        with translate_errors_context():
+            query = dict(id=queue_id, company=company_id)
+            queue = Queue.objects(entries__task=task_id, **query).first()
+            if not queue:
+                raise errors.bad_request.InvalidQueueOrTaskNotQueued(
+                    task=task_id, **query
+                )
+
+            return queue
+
+    def get_default(self, company_id: str) -> Queue:
+        """
+        Get the default queue
+        :raise errors.bad_request.NoDefaultQueue: if the default queue not found
+        :raise errors.bad_request.MultipleDefaultQueues: if more than one default queue is found
+        """
+        with translate_errors_context():
+            res = Queue.objects(company=company_id, system_tags="default").only(
+                "id", "name"
+            )
+            if not res:
+                raise errors.bad_request.NoDefaultQueue()
+            if len(res) > 1:
+                raise errors.bad_request.MultipleDefaultQueues(
+                    queues=tuple(r.id for r in res)
+                )
+
+            return res.first()
+
+    def update(
+        self, company_id: str, queue_id: str, **update_fields
+    ) -> Tuple[int, dict]:
+        """
+        Partial update of the queue from update_fields
+        :raise errors.bad_request.InvalidQueueId: if the queue is not found
+        :return: number of updated objects and updated fields dictionary
+        """
+        with translate_errors_context():
+            # validate the queue exists
+            self.get_by_id(company_id=company_id, queue_id=queue_id, only=("id",))
+            return Queue.safe_update(company_id, queue_id, update_fields)
+
+    def _update_task_status_on_removal_from_queue(
+        self,
+        company_id: str,
+        user_id: str,
+        task_ids: Iterable[str],
+        queue_id: str,
+        reason: str
+    ) -> Sequence[str]:
+        from apiserver.bll.task import ChangeStatusRequest
+        tasks = []
+        for task_id in task_ids:
+            try:
+                task = Task.get(
+                    company=company_id,
+                    id=task_id,
+                    execution__queue=queue_id,
+                    _only=[
+                        "id",
+                        "company",
+                        "status",
+                        "enqueue_status",
+                        "project",
+                    ],
+                )
+                if not task:
+                    continue
+
+                tasks.append(task.id)
+                ChangeStatusRequest(
+                    task=task,
+                    new_status=task.enqueue_status or TaskStatus.created,
+                    status_reason=reason,
+                    status_message="",
+                    user_id=user_id,
+                    force=True,
+                ).execute(
+                    enqueue_status=None,
+                    unset__execution__queue=1,
+                )
+            except Exception as ex:
+                log.error(
+                    f"Failed updating task {task_id} status on removal from queue: {queue_id}, {str(ex)}"
+                )
+
+        return tasks
+
+    def delete(self, company_id: str, user_id: str, queue_id: str, force: bool) -> Sequence[str]:
+        """
+        Delete the queue
+        :raise errors.bad_request.InvalidQueueId: if the queue is not found
+        :raise errors.bad_request.QueueNotEmpty: if the queue is not empty and 'force' not set
+        """
+        queue = self.get_by_id(company_id=company_id, queue_id=queue_id)
+        if not queue.entries:
+            queue.delete()
+            return []
+
+        if not force:
+            raise errors.bad_request.QueueNotEmpty(
+                "use force=true to delete", id=queue_id
+            )
+
+        tasks = self._update_task_status_on_removal_from_queue(
+            company_id=company_id,
+            user_id=user_id,
+            task_ids={item.task for item in queue.entries},
+            queue_id=queue_id,
+            reason=f"Queue {queue_id} was deleted",
+        )
+
+        queue.delete()
+        return tasks
+
+    def get_all(
+        self,
+        company_id: str,
+        query_dict: dict,
+        query: Q = None,
+        max_task_entries: int = None,
+        ret_params: dict = None,
+    ) -> Sequence[dict]:
+        """Get all the queues according to the query"""
+        with translate_errors_context():
+            return Queue.get_many(
+                company=company_id,
+                parameters=query_dict,
+                query_dict=query_dict,
+                query=query,
+                projection_fields=self._get_task_entries_projection(max_task_entries)
+                if max_task_entries
+                else None,
+                ret_params=ret_params,
+            )
+
+    def check_for_workers(self, company_id: str, queue_id: str) -> bool:
+        for worker in self.worker_bll.get_all(company_id):
+            if queue_id in worker.queues:
+                return True
+
+        return False
+
+    def get_queue_infos(
+        self,
+        company_id: str,
+        query_dict: dict,
+        query: Q = None,
+        max_task_entries: int = None,
+        ret_params: dict = None,
+    ) -> Sequence[dict]:
+        """
+        Get infos on all the company queues, including queue tasks and workers
+        """
+        projection = Queue.get_extra_projection("entries.task.name")
+        with translate_errors_context():
+            res = Queue.get_many_with_join(
+                company=company_id,
+                query_dict=query_dict,
+                query=query,
+                override_projection=projection,
+                projection_fields=self._get_task_entries_projection(max_task_entries)
+                if max_task_entries
+                else None,
+                ret_params=ret_params,
+            )
+
+            queue_workers = defaultdict(list)
+            for worker in self.worker_bll.get_all(company_id):
+                for queue in worker.queues:
+                    queue_workers[queue].append(worker)
+
+            for item in res:
+                item["workers"] = [
+                    {
+                        "name": w.id,
+                        "ip": w.ip,
+                        "key": w.key,
+                        "task": w.task.to_struct() if w.task else None,
+                    }
+                    for w in queue_workers.get(item["id"], [])
+                ]
+
+        return res
+
+    def add_task(self, company_id: str, queue_id: str, task_id: str) -> dict:
+        """
+        Add the task to the queue and return the queue update results
+        :raise errors.bad_request.TaskAlreadyQueued: if the task is already in the queue
+        :raise errors.bad_request.InvalidQueueOrTaskNotQueued: if the queue update operation failed
+        """
+        with translate_errors_context():
+            queue = self.get_by_id(company_id=company_id, queue_id=queue_id)
+            if any(e.task == task_id for e in queue.entries):
+                raise errors.bad_request.TaskAlreadyQueued(task=task_id)
+
+            entry = Entry(added=datetime.utcnow(), task=task_id)
+            query = dict(id=queue_id, company=company_id)
+            res = Queue.objects(entries__task__ne=task_id, **query).update_one(
+                push__entries=entry, last_update=datetime.utcnow(), upsert=False
+            )
+
+            queue.reload()
+            self.metrics.log_queue_metrics_to_es(company_id=company_id, queues=[queue])
+
+            if not res:
+                raise errors.bad_request.InvalidQueueOrTaskNotQueued(
+                    task=task_id, **query
+                )
+
+            return res
+
+    def get_next_task(
+        self, company_id: str, queue_id: str, task_id: str = None
+    ) -> Optional[Entry]:
+        """
+        Atomically pop and return the first task from the queue (or None)
+        :raise errors.bad_request.InvalidQueueId: if the queue does not exist
+        """
+        with translate_errors_context():
+            query = dict(id=queue_id, company=company_id)
+            queue = Queue.objects(
+                **query, **({"entries__0__task": task_id} if task_id else {})
+            ).modify(pop__entries=-1, upsert=False)
+            if not queue:
+                if not task_id or not Queue.objects(**query).first():
+                    raise errors.bad_request.InvalidQueueId(**query)
+                return
+
+            self.metrics.log_queue_metrics_to_es(company_id, queues=[queue])
+
+            if not queue.entries:
+                return
+
+            try:
+                Queue.objects(**query).update(last_update=datetime.utcnow())
+            except Exception:
+                log.exception("Error while updating Queue.last_update")
+
+            return queue.entries[0]
+
+    def clear_queue(
+        self,
+        company_id: str,
+        user_id: str,
+        queue_id: str,
+    ):
+        queue = Queue.objects(company=company_id, id=queue_id).first()
+        if not queue:
+            raise errors.bad_request.InvalidQueueId(
+                queue=queue_id
+            )
+
+        if not queue.entries:
+            return []
+
+        tasks = self._update_task_status_on_removal_from_queue(
+            company_id=company_id,
+            user_id=user_id,
+            task_ids={item.task for item in queue.entries},
+            queue_id=queue_id,
+            reason=f"Queue {queue_id} was cleared",
+        )
+
+        queue.update(entries=[])
+        queue.reload()
+        self.metrics.log_queue_metrics_to_es(company_id=company_id, queues=[queue])
+
+        return tasks
+
+    def remove_task(self, company_id: str, user_id: str, queue_id: str, task_id: str, update_task_status: bool = False) -> int:
+        """
+        Removes the task from the queue and returns the number of removed items
+        :raise errors.bad_request.InvalidQueueOrTaskNotQueued: if the task is not found in the queue
+        """
+        with translate_errors_context():
+            queue = self.get_queue_with_task(
+                company_id=company_id, queue_id=queue_id, task_id=task_id
+            )
+
+            entries_to_remove = [e for e in queue.entries if e.task == task_id]
+            query = dict(id=queue_id, company=company_id)
+            res = Queue.objects(entries__task=task_id, **query).update_one(
+                pull_all__entries=entries_to_remove, last_update=datetime.utcnow()
+            )
+            if res and update_task_status:
+                self._update_task_status_on_removal_from_queue(
+                    company_id=company_id,
+                    user_id=user_id,
+                    task_ids=[task_id],
+                    queue_id=queue_id,
+                    reason=f"Task was removed from the queue {queue_id}",
+                )
+
+            queue.reload()
+            self.metrics.log_queue_metrics_to_es(company_id=company_id, queues=[queue])
+
+            return len(entries_to_remove) if res else 0
+
+    def reposition_task(
+        self, company_id: str, queue_id: str, task_id: str, move_count: Union[int, str],
+    ) -> int:
+        """
+        Moves the task in the queue to the position calculated by pos_func
+        Returns the updated task position in the queue
+        """
+
+        def get_queue_and_task_position():
+            q = self.get_queue_with_task(
+                company_id=company_id, queue_id=queue_id, task_id=task_id
+            )
+            return q, next(i for i, e in enumerate(q.entries) if e.task == task_id)
+
+        with translate_errors_context():
+            queue, position = get_queue_and_task_position()
+            if move_count == MOVE_FIRST:
+                new_position = 0
+            elif move_count == MOVE_LAST:
+                new_position = len(queue.entries) - 1
+            else:
+                new_position = position + move_count
+            if new_position == position:
+                return new_position
+
+            without_entry = {
+                "$filter": {
+                    "input": "$entries",
+                    "as": "entry",
+                    "cond": {"$ne": ["$$entry.task", task_id]},
+                }
+            }
+            task_entry = {
+                "$filter": {
+                    "input": "$entries",
+                    "as": "entry",
+                    "cond": {"$eq": ["$$entry.task", task_id]},
+                }
+            }
+            if move_count == MOVE_FIRST:
+                operations = [
+                    {
+                        "$set": {
+                            "entries": {"$concatArrays": [task_entry, without_entry]}
+                        }
+                    }
+                ]
+            elif move_count == MOVE_LAST:
+                operations = [
+                    {
+                        "$set": {
+                            "entries": {"$concatArrays": [without_entry, task_entry]}
+                        }
+                    }
+                ]
+            else:
+                operations = [
+                    {
+                        "$set": {
+                            "new_pos": {
+                                "$add": [
+                                    {"$indexOfArray": ["$entries.task", task_id]},
+                                    move_count,
+                                ]
+                            },
+                            "without_entry": without_entry,
+                            "task_entry": task_entry,
+                        }
+                    },
+                    {
+                        "$set": {
+                            "entries": {
+                                "$switch": {
+                                    "branches": [
+                                        {
+                                            "case": {"$lte": ["$new_pos", 0]},
+                                            "then": {
+                                                "$concatArrays": [
+                                                    "$task_entry",
+                                                    "$without_entry",
+                                                ]
+                                            },
+                                        },
+                                        {
+                                            "case": {
+                                                "$gte": [
+                                                    "$new_pos",
+                                                    {"$size": "$without_entry"},
+                                                ]
+                                            },
+                                            "then": {
+                                                "$concatArrays": [
+                                                    "$without_entry",
+                                                    "$task_entry",
+                                                ]
+                                            },
+                                        },
+                                    ],
+                                    "default": {
+                                        "$concatArrays": [
+                                            {"$slice": ["$without_entry", "$new_pos"]},
+                                            "$task_entry",
+                                            {
+                                                "$slice": [
+                                                    "$without_entry",
+                                                    "$new_pos",
+                                                    {"$size": "$without_entry"},
+                                                ]
+                                            },
+                                        ]
+                                    },
+                                }
+                            }
+                        }
+                    },
+                    {"$unset": ["new_pos", "without_entry", "task_entry"]},
+                ]
+
+            updated = Queue.objects(
+                id=queue_id, company=company_id, entries__task=task_id
+            ).update_one(__raw__=operations)
+
+            if not updated:
+                raise errors.bad_request.FailedAddingDuringReposition(task=task_id)
+
+            return get_queue_and_task_position()[1]
+
+    def count_entries(self, company: str, queue_id: str) -> Optional[int]:
+        res = next(
+            Queue.aggregate(
+                [
+                    {
+                        "$match": {
+                            "company": {"$in": ["", company]},
+                            "_id": queue_id,
+                        }
+                    },
+                    {"$project": {"count": {"$size": "$entries"}}},
+                ]
+            ),
+            None,
+        )
+        if res is None:
+            raise errors.bad_request.InvalidQueueId(queue_id=queue_id)
+        return int(res.get("count"))

apiserver/bll/queue/queue_metrics.py

@@ -1,35 +1,42 @@
+import json
 from collections import defaultdict
 from datetime import datetime
+from time import sleep
 from typing import Sequence
 
-import elasticsearch.helpers
+from boltons.typeutils import classproperty
 from elasticsearch import Elasticsearch
 
-import es_factory
-from apierrors.errors import bad_request
-from bll.query import Builder as QueryBuilder
-from config import config
-from database.errors import translate_errors_context
-from database.model.queue import Queue, Entry
-from timing_context import TimingContext
+from apiserver.es_factory import es_factory
+from apiserver.apierrors.errors import bad_request
+from apiserver.bll.query import Builder as QueryBuilder
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.queue import Queue, Entry
+from apiserver.redis_manager import redman
+from apiserver.utilities.threads_manager import ThreadsManager
 
 log = config.logger(__file__)
+_conf = config.get("services.queues")
+_queue_metrics_key_pattern = "queue_metrics_{queue}"
+redis = redman.connection("apiserver")
+
+
+class EsKeys:
+    WAITING_TIME_FIELD = "average_waiting_time"
+    QUEUE_LENGTH_FIELD = "queue_length"
+    TIMESTAMP_FIELD = "timestamp"
+    QUEUE_FIELD = "queue"
 
 
 class QueueMetrics:
-    class EsKeys:
-        WAITING_TIME_FIELD = "average_waiting_time"
-        QUEUE_LENGTH_FIELD = "queue_length"
-        TIMESTAMP_FIELD = "timestamp"
-        QUEUE_FIELD = "queue"
-
     def __init__(self, es: Elasticsearch):
         self.es = es
 
     @staticmethod
     def _queue_metrics_prefix_for_company(company_id: str) -> str:
         """Returns the es index prefix for the company"""
-        return f"queue_metrics_{company_id}_"
+        return f"queue_metrics_{company_id.lower()}_"
 
     @staticmethod
     def _get_es_index_suffix():
@@ -49,7 +56,7 @@ class QueueMetrics:
         total_waiting_in_secs = sum((now - e.added).total_seconds() for e in entries)
         return total_waiting_in_secs / len(entries)
 
-    def log_queue_metrics_to_es(self, company_id: str, queues: Sequence[Queue]) -> bool:
+    def log_queue_metrics_to_es(self, company_id: str, queues: Sequence[Queue]) -> int:
         """
         Calculate and write queue statistics (avg waiting time and queue length) to Elastic
         :return: True if the write to es was successful, false otherwise
@@ -63,23 +70,22 @@ class QueueMetrics:
 
         def make_doc(queue: Queue) -> dict:
             entries = [e for e in queue.entries if e.added]
-            return dict(
-                _index=es_index,
-                _source={
-                    self.EsKeys.TIMESTAMP_FIELD: timestamp,
-                    self.EsKeys.QUEUE_FIELD: queue.id,
-                    self.EsKeys.WAITING_TIME_FIELD: self._calc_avg_waiting_time(
-                        entries
-                    ),
-                    self.EsKeys.QUEUE_LENGTH_FIELD: len(entries),
-                },
-            )
+            return {
+                EsKeys.TIMESTAMP_FIELD: timestamp,
+                EsKeys.QUEUE_FIELD: queue.id,
+                EsKeys.WAITING_TIME_FIELD: self._calc_avg_waiting_time(entries),
+                EsKeys.QUEUE_LENGTH_FIELD: len(entries),
+            }
 
-        actions = list(map(make_doc, queues))
+        logged = 0
+        for q in queues:
+            queue_doc = make_doc(q)
+            self.es.index(index=es_index, document=queue_doc)
+            redis_key = _queue_metrics_key_pattern.format(queue=q.id)
+            redis.set(redis_key, json.dumps(queue_doc))
+            logged += 1
 
-        es_res = elasticsearch.helpers.bulk(self.es, actions)
-        added, errors = es_res[:2]
-        return (added == len(actions)) and not errors
+        return logged
 
     def _log_current_metrics(self, company_id: str, queue_ids=Sequence[str]):
         query = dict(company=company_id)
@@ -90,8 +96,7 @@ class QueueMetrics:
 
     def _search_company_metrics(self, company_id: str, es_req: dict) -> dict:
         return self.es.search(
-            index=f"{self._queue_metrics_prefix_for_company(company_id)}*",
-            body=es_req,
+            index=f"{self._queue_metrics_prefix_for_company(company_id)}*", body=es_req,
         )
 
     @classmethod
@@ -105,13 +110,13 @@ class QueueMetrics:
         return {
             "dates": {
                 "date_histogram": {
-                    "field": cls.EsKeys.TIMESTAMP_FIELD,
+                    "field": EsKeys.TIMESTAMP_FIELD,
                     "fixed_interval": f"{interval}s",
                     "min_doc_count": 1,
                 },
                 "aggs": {
                     "queues": {
-                        "terms": {"field": cls.EsKeys.QUEUE_FIELD},
+                        "terms": {"field": EsKeys.QUEUE_FIELD},
                         "aggs": cls._get_top_waiting_agg(),
                     }
                 },
@@ -128,13 +133,13 @@ class QueueMetrics:
             "top_avg_waiting": {
                 "top_hits": {
                     "sort": [
-                        {cls.EsKeys.WAITING_TIME_FIELD: {"order": "desc"}},
-                        {cls.EsKeys.QUEUE_LENGTH_FIELD: {"order": "desc"}},
+                        {EsKeys.WAITING_TIME_FIELD: {"order": "desc"}},
+                        {EsKeys.QUEUE_LENGTH_FIELD: {"order": "desc"}},
                     ],
                     "_source": {
                         "includes": [
-                            cls.EsKeys.WAITING_TIME_FIELD,
-                            cls.EsKeys.QUEUE_LENGTH_FIELD,
+                            EsKeys.WAITING_TIME_FIELD,
+                            EsKeys.QUEUE_LENGTH_FIELD,
                         ]
                     },
                     "size": 1,
@@ -149,6 +154,7 @@ class QueueMetrics:
         to_date: float,
         interval: int,
         queue_ids: Sequence[str],
+        refresh: bool = False,
     ) -> dict:
         """
         Get the company queue metrics in the specified time range.
@@ -158,7 +164,8 @@ class QueueMetrics:
         In case no queue ids are specified the avg across all the
         company queues is calculated for each metric
         """
-        # self._log_current_metrics(company, queue_ids=queue_ids)
+        if refresh:
+            self._log_current_metrics(company_id, queue_ids=queue_ids)
 
         if from_date >= to_date:
             raise bad_request.FieldsValueError("from_date must be less than to_date")
@@ -174,7 +181,7 @@ class QueueMetrics:
             "aggs": self._get_dates_agg(interval),
         }
 
-        with translate_errors_context(), TimingContext("es", "get_queue_metrics"):
+        with translate_errors_context():
             res = self._search_company_metrics(company_id, es_req)
 
         if "aggregations" not in res:
@@ -256,7 +263,52 @@ class QueueMetrics:
                 continue
             res = queue_data["top_avg_waiting"]["hits"]["hits"][0]["_source"]
             queue_metrics[queue_data["key"]] = {
-                "queue_length": res[cls.EsKeys.QUEUE_LENGTH_FIELD],
-                "avg_waiting_time": res[cls.EsKeys.WAITING_TIME_FIELD],
+                "queue_length": res[EsKeys.QUEUE_LENGTH_FIELD],
+                "avg_waiting_time": res[EsKeys.WAITING_TIME_FIELD],
             }
         return queue_metrics
+
+
+class MetricsRefresher:
+    threads = ThreadsManager()
+
+    @classproperty
+    def watch_interval_sec(self):
+        return _conf.get("metrics_refresh_interval_sec", 300)
+
+    @classmethod
+    @threads.register("queue_metrics_refresh_watchdog", daemon=True)
+    def start(cls, queue_metrics: QueueMetrics = None):
+        if not cls.watch_interval_sec:
+            return
+
+        if not queue_metrics:
+            from .queue_bll import QueueBLL
+
+            queue_metrics = QueueBLL().metrics
+
+        sleep(10)
+        while True:
+            try:
+                for queue in Queue.objects():
+                    timestamp = es_factory.get_timestamp_millis()
+                    doc_time = 0
+                    try:
+                        redis_key = _queue_metrics_key_pattern.format(queue=queue.id)
+                        data = redis.get(redis_key)
+                        if data:
+                            queue_doc = json.loads(data)
+                            doc_time = int(queue_doc.get(EsKeys.TIMESTAMP_FIELD))
+                    except Exception as ex:
+                        log.exception(
+                            f"Error reading queue metrics data for queue {queue.id}: {str(ex)}"
+                        )
+
+                    if (
+                        not doc_time
+                        or (timestamp - doc_time) > cls.watch_interval_sec * 1000
+                    ):
+                        queue_metrics.log_queue_metrics_to_es(queue.company, [queue])
+            except Exception as ex:
+                log.exception(f"Failed collecting queue metrics: {str(ex)}")
+            sleep(60)

apiserver/bll/redis_cache_manager.py

@@ -3,8 +3,7 @@ from typing import Optional, TypeVar, Generic, Type, Callable
 
 from redis import StrictRedis
 
-import database
-from timing_context import TimingContext
+from apiserver import database
 
 T = TypeVar("T")
 
@@ -31,24 +30,36 @@ class RedisCacheManager(Generic[T]):
 
     def set_state(self, state: T) -> None:
         redis_key = self._get_redis_key(state.id)
-        with TimingContext("redis", "cache_set_state"):
-            self.redis.set(redis_key, state.to_json())
-            self.redis.expire(redis_key, self.expiration_interval)
+        self.redis.set(redis_key, state.to_json())
+        self.redis.expire(redis_key, self.expiration_interval)
 
     def get_state(self, state_id) -> Optional[T]:
         redis_key = self._get_redis_key(state_id)
-        with TimingContext("redis", "cache_get_state"):
-            response = self.redis.get(redis_key)
+        response = self.redis.get(redis_key)
         if response:
             return self.state_class.from_json(response)
 
     def delete_state(self, state_id) -> None:
-        with TimingContext("redis", "cache_delete_state"):
-            self.redis.delete(self._get_redis_key(state_id))
+        self.redis.delete(self._get_redis_key(state_id))
 
     def _get_redis_key(self, state_id):
         return f"{self.state_class}/{state_id}"
 
+    def get_or_create_state_core(
+        self,
+        state_id=None,
+        init_state: Callable[[T], None] = _do_nothing,
+        validate_state: Callable[[T], None] = _do_nothing,
+    ) -> T:
+        state = self.get_state(state_id) if state_id else None
+        if state:
+            validate_state(state)
+        else:
+            state = self.state_class(id=database.utils.id())
+            init_state(state)
+
+        return state
+
     @contextmanager
     def get_or_create_state(
         self,
@@ -66,12 +77,9 @@ class RedisCacheManager(Generic[T]):
         :param validate_state: user callback to validate the state if retrieved from cache
         Should throw an exception if the state is not valid. If not passed then no validation is done
         """
-        state = self.get_state(state_id) if state_id else None
-        if state:
-            validate_state(state)
-        else:
-            state = self.state_class(id=database.utils.id())
-            init_state(state)
+        state = self.get_or_create_state_core(
+            state_id=state_id, init_state=init_state, validate_state=validate_state
+        )
 
         try:
             yield state

apiserver/bll/serving/__init__.py

@@ -0,0 +1,376 @@
+from datetime import datetime, timedelta, timezone
+from enum import Enum, auto
+from operator import attrgetter
+from time import time
+from typing import Optional, Sequence, Union
+
+import attr
+from boltons.iterutils import chunked_iter, bucketize
+from pyhocon import ConfigTree
+
+from apiserver.apimodels.serving import (
+    ServingContainerEntry,
+    RegisterRequest,
+    StatusReportRequest,
+)
+from apiserver.apimodels.workers import MachineStats
+from apiserver.apierrors import errors
+from apiserver.config_repo import config
+from apiserver.redis_manager import redman
+from .stats import ServingStats
+
+
+log = config.logger(__file__)
+
+
+class ServingBLL:
+    def __init__(self, redis=None):
+        self.conf = config.get("services.serving", ConfigTree())
+        self.redis = redis or redman.connection("workers")
+
+    @staticmethod
+    def _get_url_key(company: str, url: str):
+        return f"serving_url_{company}_{url}"
+
+    @staticmethod
+    def _get_container_key(company: str, container_id: str) -> str:
+        """Build redis key from company and container_id"""
+        return f"serving_container_{company}_{container_id}"
+
+    def _save_serving_container_entry(self, entry: ServingContainerEntry):
+        self.redis.setex(
+            entry.key, timedelta(seconds=entry.register_timeout), entry.to_json()
+        )
+
+        url_key = self._get_url_key(entry.company_id, entry.endpoint_url)
+        expiration = int(time()) + entry.register_timeout
+        container_item = {entry.key: expiration}
+        self.redis.zadd(url_key, container_item)
+        # make sure that url set will not get stuck in redis
+        # indefinitely in case no more containers report to it
+        self.redis.expire(url_key, max(3600, entry.register_timeout))
+
+    def _get_serving_container_entry(
+        self, company_id: str, container_id: str
+    ) -> Optional[ServingContainerEntry]:
+        """
+        Get a container entry for the provided container ID.
+        """
+        key = self._get_container_key(company_id, container_id)
+        data = self.redis.get(key)
+        if not data:
+            return
+
+        try:
+            entry = ServingContainerEntry.from_json(data)
+            return entry
+        except Exception as e:
+            msg = "Failed parsing container entry"
+            log.exception(f"{msg}: {str(e)}")
+
+    def register_serving_container(
+        self,
+        company_id: str,
+        request: RegisterRequest,
+        ip: str = "",
+    ) -> ServingContainerEntry:
+        """
+        Register a serving container
+        """
+        now = datetime.now(timezone.utc)
+        key = self._get_container_key(company_id, request.container_id)
+        entry = ServingContainerEntry(
+            **request.to_struct(),
+            key=key,
+            company_id=company_id,
+            ip=ip,
+            register_time=now,
+            register_timeout=request.timeout,
+            last_activity_time=now,
+        )
+        self._save_serving_container_entry(entry)
+        return entry
+
+    def unregister_serving_container(
+        self,
+        company_id: str,
+        container_id: str,
+    ) -> None:
+        """
+        Unregister a serving container
+        """
+        entry = self._get_serving_container_entry(company_id, container_id)
+        if entry:
+            url_key = self._get_url_key(entry.company_id, entry.endpoint_url)
+            self.redis.zrem(url_key, entry.key)
+
+        key = self._get_container_key(company_id, container_id)
+        res = self.redis.delete(key)
+        if res:
+            return
+
+        if not self.conf.get("container_auto_unregister", True):
+            raise errors.bad_request.ContainerNotRegistered(container=container_id)
+
+    def container_status_report(
+        self,
+        company_id: str,
+        report: StatusReportRequest,
+        ip: str = "",
+    ) -> None:
+        """
+        Serving container status report
+        """
+        container_id = report.container_id
+        now = datetime.now(timezone.utc)
+        entry = self._get_serving_container_entry(company_id, container_id)
+        if entry:
+            ip = ip or entry.ip
+            register_time = entry.register_time
+            register_timeout = entry.register_timeout
+        else:
+            if not self.conf.get("container_auto_register", True):
+                raise errors.bad_request.ContainerNotRegistered(container=container_id)
+            ip = ip
+            register_time = now
+            register_timeout = int(
+                self.conf.get("default_container_timeout_sec", 10 * 60)
+            )
+
+        key = self._get_container_key(company_id, container_id)
+        entry = ServingContainerEntry(
+            **report.to_struct(),
+            key=key,
+            company_id=company_id,
+            ip=ip,
+            register_time=register_time,
+            register_timeout=register_timeout,
+            last_activity_time=now,
+        )
+        self._save_serving_container_entry(entry)
+        ServingStats.log_stats_to_es(entry)
+
+    def _get_all(
+        self,
+        company_id: str,
+    ) -> Sequence[ServingContainerEntry]:
+        keys = list(self.redis.scan_iter(self._get_container_key(company_id, "*")))
+        entries = []
+        for keys in chunked_iter(keys, 1000):
+            data = self.redis.mget(keys)
+            if not data:
+                continue
+            for d in data:
+                try:
+                    entries.append(ServingContainerEntry.from_json(d))
+                except Exception as ex:
+                    log.error(f"Failed parsing container entry {str(ex)}")
+
+        return entries
+
+    @attr.s(auto_attribs=True)
+    class Counter:
+        class AggType(Enum):
+            avg = auto()
+            max = auto()
+            total = auto()
+            count = auto()
+
+        name: str
+        field: str
+        agg_type: AggType
+        float_precision: int = None
+
+        _max: Union[int, float, datetime] = attr.field(init=False, default=None)
+        _total: Union[int, float] = attr.field(init=False, default=0)
+        _count: int = attr.field(init=False, default=0)
+
+        def add(self, entry: ServingContainerEntry):
+            value = getattr(entry, self.field, None)
+            if value is None:
+                return
+
+            self._count += 1
+            if self.agg_type == self.AggType.max:
+                self._max = value if self._max is None else max(self._max, value)
+            else:
+                self._total += value
+
+        def __call__(self):
+            if self.agg_type == self.AggType.count:
+                return self._count
+
+            if self.agg_type == self.AggType.max:
+                return self._max
+
+            if self.agg_type == self.AggType.total:
+                return self._total
+
+            if not self._count:
+                return None
+            avg = self._total / self._count
+            return (
+                round(avg, self.float_precision) if self.float_precision else round(avg)
+            )
+
+    def _get_summary(self, entries: Sequence[ServingContainerEntry]) -> dict:
+        counters = [
+            self.Counter(
+                name="uptime_sec",
+                field="uptime_sec",
+                agg_type=self.Counter.AggType.max,
+            ),
+            self.Counter(
+                name="requests",
+                field="requests_num",
+                agg_type=self.Counter.AggType.total,
+            ),
+            self.Counter(
+                name="requests_min",
+                field="requests_min",
+                agg_type=self.Counter.AggType.avg,
+                float_precision=2,
+            ),
+            self.Counter(
+                name="latency_ms",
+                field="latency_ms",
+                agg_type=self.Counter.AggType.avg,
+            ),
+            self.Counter(
+                name="last_update",
+                field="last_activity_time",
+                agg_type=self.Counter.AggType.max,
+            ),
+        ]
+        for entry in entries:
+            for counter in counters:
+                counter.add(entry)
+
+        first_entry = entries[0]
+        ret = {
+            "endpoint": first_entry.endpoint_name,
+            "model": first_entry.model_name,
+            "url": first_entry.endpoint_url,
+            "instances": len(entries),
+            **{counter.name: counter() for counter in counters},
+        }
+        ret["last_update"] = ret.get("last_update")
+        return ret
+
+    def get_endpoints(self, company_id: str):
+        """
+        Group instances by urls and return a summary for each url
+        Do not return data for "loading" instances that have no url
+        """
+        entries = self._get_all(company_id)
+        by_url = bucketize(entries, key=attrgetter("endpoint_url"))
+        by_url.pop(None, None)
+        return [self._get_summary(url_entries) for url_entries in by_url.values()]
+
+    def _get_endpoint_entries(
+        self, company_id, endpoint_url: Union[str, None]
+    ) -> Sequence[ServingContainerEntry]:
+        url_key = self._get_url_key(company_id, endpoint_url)
+        timestamp = int(time())
+        self.redis.zremrangebyscore(url_key, min=0, max=timestamp)
+        container_keys = {key.decode() for key in self.redis.zrange(url_key, 0, -1)}
+        if not container_keys:
+            return []
+
+        entries = []
+        found_keys = set()
+        data = self.redis.mget(container_keys) or []
+        for d in data:
+            try:
+                entry = ServingContainerEntry.from_json(d)
+                if entry.endpoint_url == endpoint_url:
+                    entries.append(entry)
+                    found_keys.add(entry.key)
+            except Exception as ex:
+                log.error(f"Failed parsing container entry {str(ex)}")
+
+        missing_keys = container_keys - found_keys
+        if missing_keys:
+            self.redis.zrem(url_key, *missing_keys)
+
+        return entries
+
+    def get_loading_instances(self, company_id: str):
+        entries = self._get_endpoint_entries(company_id, None)
+        return [
+            {
+                "id": entry.container_id,
+                "endpoint": entry.endpoint_name,
+                "url": entry.endpoint_url,
+                "model": entry.model_name,
+                "model_source": entry.model_source,
+                "model_version": entry.model_version,
+                "preprocess_artifact": entry.preprocess_artifact,
+                "input_type": entry.input_type,
+                "input_size": entry.input_size,
+                "uptime_sec": entry.uptime_sec,
+                "age_sec": int((datetime.now(timezone.utc) - entry.register_time).total_seconds()),
+                "last_update": entry.last_activity_time,
+            }
+            for entry in entries
+        ]
+
+    def get_endpoint_details(self, company_id, endpoint_url: str) -> dict:
+        entries = self._get_endpoint_entries(company_id, endpoint_url)
+        if not entries:
+            raise errors.bad_request.NoContainersForUrl(url=endpoint_url)
+
+        instances = []
+        entry: ServingContainerEntry
+        for entry in entries:
+            instances.append(
+                {
+                    "endpoint": entry.endpoint_name,
+                    "model": entry.model_name,
+                    "url": entry.endpoint_url,
+                }
+            )
+
+        def get_machine_stats_data(machine_stats: MachineStats) -> dict:
+            ret = {"cpu_count": 0, "gpu_count": 0}
+            if not machine_stats:
+                return ret
+
+            for value, field in (
+                (machine_stats.cpu_usage, "cpu_count"),
+                (machine_stats.gpu_usage, "gpu_count"),
+            ):
+                if value is None:
+                    continue
+                ret[field] = len(value) if isinstance(value, (list, tuple)) else 1
+
+            return ret
+
+        first_entry = entries[0]
+        return {
+            "endpoint": first_entry.endpoint_name,
+            "model": first_entry.model_name,
+            "url": first_entry.endpoint_url,
+            "preprocess_artifact": first_entry.preprocess_artifact,
+            "input_type": first_entry.input_type,
+            "input_size": first_entry.input_size,
+            "model_source": first_entry.model_source,
+            "model_version": first_entry.model_version,
+            "uptime_sec": max(e.uptime_sec for e in entries),
+            "last_update": max(e.last_activity_time for e in entries),
+            "instances": [
+                {
+                    "id": entry.container_id,
+                    "uptime_sec": entry.uptime_sec,
+                    "requests": entry.requests_num,
+                    "requests_min": entry.requests_min,
+                    "latency_ms": entry.latency_ms,
+                    "last_update": entry.last_activity_time,
+                    "reference": [ref.to_struct() for ref in entry.reference]
+                    if isinstance(entry.reference, list)
+                    else entry.reference,
+                    **get_machine_stats_data(entry.machine_stats),
+                }
+                for entry in entries
+            ],
+        }

apiserver/bll/serving/stats.py

@@ -0,0 +1,340 @@
+from collections import defaultdict
+from datetime import datetime, timezone
+from enum import Enum
+
+from typing import Tuple, Optional, Sequence
+
+from elasticsearch import Elasticsearch
+
+from apiserver.apimodels.serving import (
+    ServingContainerEntry,
+    GetEndpointMetricsHistoryRequest,
+    MetricType,
+)
+from apiserver.apierrors import errors
+from apiserver.utilities.dicts import nested_get
+from apiserver.bll.query import Builder as QueryBuilder
+from apiserver.config_repo import config
+from apiserver.es_factory import es_factory
+
+
+class _AggregationType(Enum):
+    avg = "avg"
+    sum = "sum"
+
+
+class ServingStats:
+    min_chart_interval = config.get("services.serving.min_chart_interval_sec", 40)
+    es: Elasticsearch = es_factory.connect("workers")
+
+    @classmethod
+    def _serving_stats_prefix(cls, company_id: str) -> str:
+        """Returns the es index prefix for the company"""
+        return f"serving_stats_{company_id.lower()}_"
+
+    @staticmethod
+    def _get_es_index_suffix():
+        """Get the index name suffix for storing current month data"""
+        return datetime.now(timezone.utc).strftime("%Y-%m")
+
+    @staticmethod
+    def _get_average_value(value) -> Tuple[Optional[float], Optional[int]]:
+        if value is None:
+            return None, None
+
+        if isinstance(value, (list, tuple)):
+            count = len(value)
+            if not count:
+                return None, None
+            return sum(value) / count, count
+
+        return value, 1
+
+    @classmethod
+    def log_stats_to_es(
+        cls,
+        entry: ServingContainerEntry,
+    ) -> int:
+        """
+        Actually writing the worker statistics to Elastic
+        :return: The amount of logged documents
+        """
+        company_id = entry.company_id
+        es_index = (
+            f"{cls._serving_stats_prefix(company_id)}" f"{cls._get_es_index_suffix()}"
+        )
+
+        entry_data = entry.to_struct()
+        doc = {
+            "timestamp": es_factory.get_timestamp_millis(),
+            **{
+                field: entry_data.get(field)
+                for field in (
+                    "container_id",
+                    "company_id",
+                    "endpoint_url",
+                    "requests_num",
+                    "requests_min",
+                    "uptime_sec",
+                    "latency_ms",
+                )
+            },
+        }
+
+        stats = entry_data.get("machine_stats")
+        if stats:
+            for category in ("cpu", "gpu"):
+                usage, num = cls._get_average_value(stats.get(f"{category}_usage"))
+                doc.update({f"{category}_usage": usage, f"{category}_num": num})
+
+            for category in ("memory", "gpu_memory"):
+                free, _ = cls._get_average_value(stats.get(f"{category}_free"))
+                used, _ = cls._get_average_value(stats.get(f"{category}_used"))
+                doc.update(
+                    {
+                        f"{category}_free": free,
+                        f"{category}_used": used,
+                        f"{category}_total": round((free or 0) + (used or 0), 3),
+                    }
+                )
+
+            doc.update(
+                {
+                    field: stats.get(field)
+                    for field in ("disk_free_home", "network_rx", "network_tx")
+                }
+            )
+
+        cls.es.index(index=es_index, document=doc)
+
+        return 1
+
+    @staticmethod
+    def round_series(values: Sequence, koeff) -> list:
+        return [round(v * koeff, 2) if v else 0 for v in values]
+
+    _mb_to_gb = 1 / 1024
+    agg_fields = {
+        MetricType.requests: (
+            "requests_num",
+            "Number of Requests",
+            _AggregationType.sum,
+            None,
+        ),
+        MetricType.requests_min: (
+            "requests_min",
+            "Requests per Minute",
+            _AggregationType.sum,
+            None,
+        ),
+        MetricType.latency_ms: (
+            "latency_ms",
+            "Average Latency (ms)",
+            _AggregationType.avg,
+            None,
+        ),
+        MetricType.cpu_count: ("cpu_num", "CPU Count", _AggregationType.sum, None),
+        MetricType.gpu_count: ("gpu_num", "GPU Count", _AggregationType.sum, None),
+        MetricType.cpu_util: (
+            "cpu_usage",
+            "Average CPU Load (%)",
+            _AggregationType.avg,
+            None,
+        ),
+        MetricType.gpu_util: (
+            "gpu_usage",
+            "Average GPU Utilization (%)",
+            _AggregationType.avg,
+            None,
+        ),
+        MetricType.ram_total: (
+            "memory_total",
+            "RAM Total (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.ram_used: (
+            "memory_used",
+            "RAM Used (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.ram_free: (
+            "memory_free",
+            "RAM Free (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.gpu_ram_total: (
+            "gpu_memory_total",
+            "GPU RAM Total (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.gpu_ram_used: (
+            "gpu_memory_used",
+            "GPU RAM Used (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.gpu_ram_free: (
+            "gpu_memory_free",
+            "GPU RAM Free (GB)",
+            _AggregationType.sum,
+            _mb_to_gb,
+        ),
+        MetricType.network_rx: (
+            "network_rx",
+            "Network Throughput RX (MBps)",
+            _AggregationType.sum,
+            None,
+        ),
+        MetricType.network_tx: (
+            "network_tx",
+            "Network Throughput TX (MBps)",
+            _AggregationType.sum,
+            None,
+        ),
+    }
+
+    @classmethod
+    def get_endpoint_metrics(
+        cls,
+        company_id: str,
+        metrics_request: GetEndpointMetricsHistoryRequest,
+    ) -> dict:
+        from_date = metrics_request.from_date
+        to_date = metrics_request.to_date
+        if from_date >= to_date:
+            raise errors.bad_request.FieldsValueError(
+                "from_date must be less than to_date"
+            )
+
+        metric_type = metrics_request.metric_type
+        agg_data = cls.agg_fields.get(metric_type)
+        if not agg_data:
+            raise NotImplemented(f"Charts for {metric_type} not implemented")
+
+        agg_field, title, agg_type, multiplier = agg_data
+        if agg_type == _AggregationType.sum:
+            instance_sum_type = "sum_bucket"
+        else:
+            instance_sum_type = "avg_bucket"
+
+        interval = max(metrics_request.interval, cls.min_chart_interval)
+        endpoint_url = metrics_request.endpoint_url
+        hist_ret = {
+            "computed_interval": interval,
+            "total": {
+                "title": title,
+                "dates": [],
+                "values": [],
+            },
+            "instances": {},
+        }
+        must_conditions = [
+            QueryBuilder.term("company_id", company_id),
+            QueryBuilder.term("endpoint_url", endpoint_url),
+            QueryBuilder.dates_range(from_date, to_date),
+        ]
+        query = {"bool": {"must": must_conditions}}
+        es_index = f"{cls._serving_stats_prefix(company_id)}*"
+        res = cls.es.search(
+            index=es_index,
+            size=0,
+            query=query,
+            aggs={"instances": {"terms": {"field": "container_id"}}},
+        )
+        instance_buckets = nested_get(res, ("aggregations", "instances", "buckets"))
+        if not instance_buckets:
+            return hist_ret
+
+        instance_keys = {ib["key"] for ib in instance_buckets}
+        must_conditions.append(QueryBuilder.terms("container_id", instance_keys))
+        query = {"bool": {"must": must_conditions}}
+        sample_func = "avg" if metric_type != MetricType.requests else "max"
+        aggs = {
+            "instances": {
+                "terms": {
+                    "field": "container_id",
+                    "size": max(len(instance_keys), 10),
+                },
+                "aggs": {
+                    "sample": {sample_func: {"field": agg_field}},
+                },
+            },
+            "total_instances": {
+                instance_sum_type: {
+                    "gap_policy": "insert_zeros",
+                    "buckets_path": "instances>sample",
+                }
+            },
+        }
+        hist_params = {}
+        if metric_type == MetricType.requests:
+            hist_params["min_doc_count"] = 1
+        else:
+            hist_params["extended_bounds"] = {
+                "min": int(from_date) * 1000,
+                "max": int(to_date) * 1000,
+            }
+        aggs = {
+            "dates": {
+                "date_histogram": {
+                    "field": "timestamp",
+                    "fixed_interval": f"{interval}s",
+                    **hist_params,
+                },
+                "aggs": aggs,
+            }
+        }
+
+        filter_path = None
+        if not metrics_request.instance_charts:
+            filter_path = "aggregations.dates.buckets.total_instances"
+
+        data = cls.es.search(
+            index=es_index,
+            size=0,
+            query=query,
+            aggs=aggs,
+            filter_path=filter_path,
+        )
+        agg_res = data.get("aggregations")
+        if not agg_res:
+            return hist_ret
+
+        dates_ = []
+        total = []
+        instances = defaultdict(list)
+        # remove last interval if it's incomplete. Allow 10% tolerance
+        last_valid_timestamp = (to_date - 0.9 * interval) * 1000
+        for point in agg_res["dates"]["buckets"]:
+            date_ = point["key"]
+            if date_ > last_valid_timestamp:
+                break
+            dates_.append(date_)
+            total.append(nested_get(point, ("total_instances", "value"), 0))
+            if metrics_request.instance_charts:
+                found_keys = set()
+                for instance in nested_get(point, ("instances", "buckets"), []):
+                    instances[instance["key"]].append(
+                        nested_get(instance, ("sample", "value"), 0)
+                    )
+                    found_keys.add(instance["key"])
+                for missing_key in instance_keys - found_keys:
+                    instances[missing_key].append(0)
+
+        koeff = multiplier if multiplier else 1.0
+        hist_ret["total"]["dates"] = dates_
+        hist_ret["total"]["values"] = cls.round_series(total, koeff)
+        hist_ret["instances"] = {
+            key: {
+                "title": key,
+                "dates": dates_,
+                "values": cls.round_series(values, koeff),
+            }
+            for key, values in sorted(instances.items(), key=lambda p: p[0])
+        }
+
+        return hist_ret

apiserver/bll/statistics/resource_monitor.py

@@ -0,0 +1,97 @@
+from datetime import datetime
+import operator
+from threading import Lock
+from time import sleep
+
+import attr
+import psutil
+
+from apiserver.utilities.threads_manager import ThreadsManager
+
+
+stat_threads = ThreadsManager("Statistics")
+
+
+@attr.s(auto_attribs=True)
+class Sample:
+    cpu_usage: float = 0.0
+    mem_used_gb: float = 0
+    mem_free_gb: float = 0
+
+    @classmethod
+    def _apply(cls, op, *samples):
+        return cls(
+            **{
+                field: op(*(getattr(sample, field) for sample in samples))
+                for field in attr.fields_dict(cls)
+            }
+        )
+
+    def min(self, sample):
+        return self._apply(min, self, sample)
+
+    def max(self, sample):
+        return self._apply(max, self, sample)
+
+    def avg(self, sample, count):
+        res = self._apply(lambda x: x * count, self)
+        res = self._apply(operator.add, res, sample)
+        res = self._apply(lambda x: x / (count + 1), res)
+        return res
+
+    @classmethod
+    def get_current_sample(cls) -> "Sample":
+        return cls(
+            cpu_usage=psutil.cpu_percent(),
+            mem_used_gb=psutil.virtual_memory().used / (1024 ** 3),
+            mem_free_gb=psutil.virtual_memory().free / (1024 ** 3),
+        )
+
+
+class ResourceMonitor:
+    class Accumulator:
+        def __init__(self):
+            sample = Sample.get_current_sample()
+            self.avg = sample
+            self.min = sample
+            self.max = sample
+            self.time = datetime.utcnow()
+            self.count = 1
+
+        def add_sample(self, sample: Sample):
+            self.min = self.min.min(sample)
+            self.max = self.max.max(sample)
+            self.avg = self.avg.avg(sample, self.count)
+            self.count += 1
+
+    sample_interval_sec = 5
+    _lock = Lock()
+    accumulator = Accumulator()
+
+    @classmethod
+    @stat_threads.register("resource_monitor", daemon=True)
+    def start(cls):
+        while True:
+            sleep(cls.sample_interval_sec)
+            sample = Sample.get_current_sample()
+            with cls._lock:
+                cls.accumulator.add_sample(sample)
+
+    @classmethod
+    def get_stats(cls) -> dict:
+        """ Returns current resource statistics and clears internal resource statistics """
+        with cls._lock:
+            min_ = attr.asdict(cls.accumulator.min)
+            max_ = attr.asdict(cls.accumulator.max)
+            avg = attr.asdict(cls.accumulator.avg)
+            interval = datetime.utcnow() - cls.accumulator.time
+            cls.accumulator = cls.Accumulator()
+
+        return {
+            "interval_sec": interval.total_seconds(),
+            "num_cores": psutil.cpu_count(),
+            **{
+                k: {"min": v, "max": max_[k], "avg": avg[k]}
+                for k, v in min_.items()
+            }
+        }

apiserver/bll/statistics/stats_reporter.py

@@ -8,22 +8,20 @@ from typing import Sequence, Optional
 
 import dpath
 import requests
-from requests.adapters import HTTPAdapter
-from requests.packages.urllib3.util.retry import Retry
+from requests.adapters import HTTPAdapter, Retry
 
-from bll.query import Builder as QueryBuilder
-from bll.util import get_server_uuid
-from bll.workers import WorkerStats, WorkerBLL
-from config import config
-from config.info import get_deployment_type
-from database.model import Company, User
-from database.model.queue import Queue
-from database.model.task.task import Task
-from tools import safe_get
-from utilities.json import dumps
-from utilities.threads_manager import ThreadsManager
-from version import __version__ as current_version
-from .resource_monitor import ResourceMonitor
+from apiserver.bll.query import Builder as QueryBuilder
+from apiserver.bll.util import get_server_uuid
+from apiserver.bll.workers import WorkerStats, WorkerBLL
+from apiserver.config_repo import config
+from apiserver.config.info import get_deployment_type
+from apiserver.database.model import Company, User
+from apiserver.database.model.queue import Queue
+from apiserver.database.model.task.task import Task
+from apiserver.utilities.dicts import nested_get
+from apiserver.utilities.json import dumps
+from apiserver.version import __version__ as current_version
+from .resource_monitor import ResourceMonitor, stat_threads
 
 log = config.logger(__file__)
 
@@ -31,21 +29,23 @@ worker_bll = WorkerBLL()
 
 
 class StatisticsReporter:
-    threads = ThreadsManager("Statistics", resource_monitor=ResourceMonitor)
     send_queue = queue.Queue()
     supported = config.get("apiserver.statistics.supported", True)
 
     @classmethod
     def start(cls):
+        if not cls.supported:
+            return
+        ResourceMonitor.start()
         cls.start_sender()
         cls.start_reporter()
 
     @classmethod
-    @threads.register("reporter", daemon=True)
+    @stat_threads.register("reporter", daemon=True)
     def start_reporter(cls):
         """
         Periodically send statistics reports for companies who have opted in.
-        Note: in trains we usually have only a single company
+        Note: in clearml we usually have only a single company
         """
         if not cls.supported:
             return
@@ -54,7 +54,7 @@ class StatisticsReporter:
             hours=config.get("apiserver.statistics.report_interval_hours", 24)
         )
         sleep(report_interval.total_seconds())
-        while not ThreadsManager.terminating:
+        while True:
             try:
                 for company in Company.objects(
                     defaults__stats_option__enabled=True
@@ -68,7 +68,7 @@ class StatisticsReporter:
             sleep(report_interval.total_seconds())
 
     @classmethod
-    @threads.register("sender", daemon=True)
+    @stat_threads.register("sender", daemon=True)
     def start_sender(cls):
         if not cls.supported:
             return
@@ -85,7 +85,7 @@ class StatisticsReporter:
 
         WarningFilter.attach()
 
-        while not ThreadsManager.terminating:
+        while True:
             try:
                 report = cls.send_queue.get()
 
@@ -111,7 +111,7 @@ class StatisticsReporter:
                 "uuid": get_server_uuid(),
                 "queues": {"count": Queue.objects(company=company_id).count()},
                 "users": {"count": User.objects(company=company_id).count()},
-                "resources": cls.threads.resource_monitor.get_stats(),
+                "resources": ResourceMonitor.get_stats(),
                 "experiments": next(
                     iter(cls._get_experiments_stats(company_id).values()), {}
                 ),
@@ -162,7 +162,7 @@ class StatisticsReporter:
         def _get_cardinality_fields(categories: Sequence[dict]) -> dict:
             names = {"cpu": "num_cores"}
             return {
-                names[c["key"]]: safe_get(c, "count/value")
+                names[c["key"]]: nested_get(c, ("count", "value"))
                 for c in categories
                 if c["key"] in names
             }
@@ -175,21 +175,21 @@ class StatisticsReporter:
             }
             return {
                 names[m["key"]]: {
-                    "min": safe_get(m, "min/value"),
-                    "max": safe_get(m, "max/value"),
-                    "avg": safe_get(m, "avg/value"),
+                    "min": nested_get(m, ("min", "value")),
+                    "max": nested_get(m, ("max", "value")),
+                    "avg": nested_get(m, ("avg", "value")),
                 }
                 for m in metrics
                 if m["key"] in names
             }
 
-        buckets = safe_get(res, "aggregations/workers/buckets", default=[])
+        buckets = nested_get(res, ("aggregations", "workers", "buckets"), default=[])
         return {
             b["key"]: {
                 key: {
                     "interval_sec": agent_resource_threshold_sec,
-                    **_get_cardinality_fields(safe_get(b, "categories/buckets", [])),
-                    **_get_metric_fields(safe_get(b, "metrics/buckets", [])),
+                    **_get_cardinality_fields(nested_get(b, ("categories", "buckets"), [])),
+                    **_get_metric_fields(nested_get(b, ("metrics", "buckets"), [])),
                 }
             }
             for b in buckets
@@ -227,7 +227,7 @@ class StatisticsReporter:
             },
         }
         res = cls._run_worker_stats_query(company_id, es_req)
-        buckets = safe_get(res, "aggregations/workers/buckets", default=[])
+        buckets = nested_get(res, ("aggregations", "workers", "buckets"), default=[])
         return {
             b["key"]: {"last_activity_time": b["last_activity_time"]["value"]}
             for b in buckets
@@ -254,6 +254,14 @@ class StatisticsReporter:
                     **({"last_worker": {"$in": workers}} if workers else {}),
                 }
             },
+            {
+                "$project": {
+                    "last_worker": 1,
+                    "last_update": 1,
+                    "started": 1,
+                    "last_iteration": 1,
+                }
+            },
             {
                 "$group": {
                     "_id": "$last_worker" if workers else None,

apiserver/bll/storage/__init__.py

@@ -0,0 +1,273 @@
+import json
+import os
+import tempfile
+from copy import copy
+from datetime import datetime
+from typing import Optional, Sequence
+
+import attr
+from boltons.cacheutils import cachedproperty
+from clearml.backend_config.bucket_config import (
+    S3BucketConfigurations,
+    AzureContainerConfigurations,
+    GSBucketConfigurations,
+    AzureContainerConfig,
+    GSBucketConfig,
+    S3BucketConfig,
+)
+
+from apiserver.apierrors import errors
+from apiserver.apimodels.storage import SetSettingsRequest
+from apiserver.config_repo import config
+from apiserver.database.model.storage_settings import (
+    StorageSettings,
+    GoogleBucketSettings,
+    AWSSettings,
+    AzureStorageSettings,
+    GoogleStorageSettings,
+)
+from apiserver.database.utils import id as db_id
+
+log = config.logger(__file__)
+
+
+class StorageBLL:
+    default_aws_configs: S3BucketConfigurations = None
+    conf = config.get("services.storage_credentials")
+
+    @cachedproperty
+    def _default_aws_configs(self) -> S3BucketConfigurations:
+        return S3BucketConfigurations.from_config(self.conf.get("aws.s3"))
+
+    @cachedproperty
+    def _default_azure_configs(self) -> AzureContainerConfigurations:
+        return AzureContainerConfigurations.from_config(self.conf.get("azure.storage"))
+
+    @cachedproperty
+    def _default_gs_configs(self) -> GSBucketConfigurations:
+        return GSBucketConfigurations.from_config(self.conf.get("google.storage"))
+
+    def get_azure_settings_for_company(
+        self,
+        company_id: str,
+        db_settings: StorageSettings = None,
+        query_db: bool = True,
+    ) -> AzureContainerConfigurations:
+        if not db_settings and query_db:
+            db_settings = (
+                StorageSettings.objects(company=company_id).only("azure").first()
+            )
+
+        if not db_settings or not db_settings.azure:
+            return copy(self._default_azure_configs)
+
+        azure = db_settings.azure
+        return AzureContainerConfigurations(
+            container_configs=[
+                AzureContainerConfig(**entry.to_proper_dict())
+                for entry in (azure.containers or [])
+            ]
+        )
+
+    def get_gs_settings_for_company(
+        self,
+        company_id: str,
+        db_settings: StorageSettings = None,
+        query_db: bool = True,
+        json_string: bool = False,
+    ) -> GSBucketConfigurations:
+        if not db_settings and query_db:
+            db_settings = (
+                StorageSettings.objects(company=company_id).only("google").first()
+            )
+
+        if not db_settings or not db_settings.google:
+            if not json_string:
+                return copy(self._default_gs_configs)
+
+            if self._default_gs_configs._buckets:
+                buckets = [
+                    attr.evolve(
+                        b,
+                        credentials_json=self._assure_json_string(b.credentials_json),
+                    )
+                    for b in self._default_gs_configs._buckets
+                ]
+            else:
+                buckets = self._default_gs_configs._buckets
+
+            return GSBucketConfigurations(
+                buckets=buckets,
+                default_project=self._default_gs_configs._default_project,
+                default_credentials=self._assure_json_string(
+                    self._default_gs_configs._default_credentials
+                ),
+            )
+
+        def get_bucket_config(bc: GoogleBucketSettings) -> GSBucketConfig:
+            data = bc.to_proper_dict()
+            if not json_string and bc.credentials_json:
+                data["credentials_json"] = self._assure_json_file(bc.credentials_json)
+            return GSBucketConfig(**data)
+
+        google = db_settings.google
+        buckets_configs = [get_bucket_config(b) for b in (google.buckets or [])]
+        return GSBucketConfigurations(
+            buckets=buckets_configs,
+            default_project=google.project,
+            default_credentials=google.credentials_json
+            if json_string
+            else self._assure_json_file(google.credentials_json),
+        )
+
+    def get_aws_settings_for_company(
+        self,
+        company_id: str,
+        db_settings: StorageSettings = None,
+        query_db: bool = True,
+    ) -> S3BucketConfigurations:
+        if not db_settings and query_db:
+            db_settings = (
+                StorageSettings.objects(company=company_id).only("aws").first()
+            )
+        if not db_settings or not db_settings.aws:
+            return copy(self._default_aws_configs)
+
+        aws = db_settings.aws
+        buckets_configs = S3BucketConfig.from_list(
+            [b.to_proper_dict() for b in (aws.buckets or [])]
+        )
+        return S3BucketConfigurations(
+            buckets=buckets_configs,
+            default_key=aws.key,
+            default_secret=aws.secret,
+            default_region=aws.region,
+            default_use_credentials_chain=aws.use_credentials_chain,
+            default_token=aws.token,
+            default_extra_args={},
+        )
+
+    def _assure_json_file(self, name_or_content: str) -> str:
+        if not name_or_content:
+            return name_or_content
+
+        if name_or_content.endswith(".json") or os.path.exists(name_or_content):
+            return name_or_content
+
+        try:
+            json.loads(name_or_content)
+        except Exception:
+            return name_or_content
+
+        with tempfile.NamedTemporaryFile(
+            mode="wt", delete=False, suffix=".json"
+        ) as tmp:
+            tmp.write(name_or_content)
+
+        return tmp.name
+
+    def _assure_json_string(self, name_or_content: str) -> Optional[str]:
+        if not name_or_content:
+            return name_or_content
+
+        try:
+            json.loads(name_or_content)
+            return name_or_content
+        except Exception:
+            pass
+
+        try:
+            with open(name_or_content) as fp:
+                return fp.read()
+        except Exception:
+            return None
+
+    def get_company_settings(self, company_id: str) -> dict:
+        db_settings = StorageSettings.objects(company=company_id).first()
+        aws = self.get_aws_settings_for_company(company_id, db_settings, query_db=False)
+        aws_dict = {
+            "key": aws._default_key,
+            "secret": aws._default_secret,
+            "token": aws._default_token,
+            "region": aws._default_region,
+            "use_credentials_chain": aws._default_use_credentials_chain,
+            "buckets": [attr.asdict(b) for b in aws._buckets],
+        }
+
+        gs = self.get_gs_settings_for_company(
+            company_id, db_settings, query_db=False, json_string=True
+        )
+        gs_dict = {
+            "project": gs._default_project,
+            "credentials_json": gs._default_credentials or None,
+            "buckets": [attr.asdict(b) for b in gs._buckets],
+        }
+
+        azure = self.get_azure_settings_for_company(company_id, db_settings)
+        azure_dict = {
+            "containers": [attr.asdict(ac) for ac in azure._container_configs],
+        }
+
+        return {
+            "aws": aws_dict,
+            "google": gs_dict,
+            "azure": azure_dict,
+            "last_update": db_settings.last_update if db_settings else None,
+        }
+
+    def set_company_settings(
+        self, company_id: str, settings: SetSettingsRequest
+    ) -> int:
+        update_dict = {}
+        if settings.aws:
+            update_dict["aws"] = {
+                **{
+                    k: v
+                    for k, v in settings.aws.to_struct().items()
+                    if k in AWSSettings.get_fields()
+                }
+            }
+
+        if settings.azure:
+            update_dict["azure"] = {
+                **{
+                    k: v
+                    for k, v in settings.azure.to_struct().items()
+                    if k in AzureStorageSettings.get_fields()
+                }
+            }
+
+        if settings.google:
+            update_dict["google"] = {
+                **{
+                    k: v
+                    for k, v in settings.google.to_struct().items()
+                    if k in GoogleStorageSettings.get_fields()
+                }
+            }
+            cred_json = update_dict["google"].get("credentials_json")
+            if cred_json:
+                try:
+                    json.loads(cred_json)
+                except Exception as ex:
+                    raise errors.bad_request.ValidationError(
+                        f"Invalid json credentials: {str(ex)}"
+                    )
+
+        if not update_dict:
+            raise errors.bad_request.ValidationError("No settings were provided")
+
+        settings = StorageSettings.objects(company=company_id).only("id").first()
+        settings_id = settings.id if settings else db_id()
+        return StorageSettings.objects(id=settings_id).update(
+            upsert=True,
+            id=settings_id,
+            company=company_id,
+            last_update=datetime.utcnow(),
+            **update_dict,
+        )
+
+    def reset_company_settings(self, company_id: str, keys: Sequence[str]) -> int:
+        return StorageSettings.objects(company=company_id).update(
+            last_update=datetime.utcnow(), **{f"unset__{k}": 1 for k in keys}
+        )

apiserver/bll/task/__init__.py

@@ -1,7 +1,5 @@
 from .task_bll import TaskBLL
 from .utils import (
     ChangeStatusRequest,
-    update_project_time,
     validate_status_change,
-    split_by,
 )

apiserver/bll/task/artifacts.py

@@ -0,0 +1,93 @@
+from operator import itemgetter
+from typing import Sequence
+
+from apiserver.apimodels.tasks import Artifact as ApiArtifact, ArtifactId
+from apiserver.bll.task.utils import get_task_for_update, update_task
+from apiserver.database.model.task.task import DEFAULT_ARTIFACT_MODE, Artifact
+from apiserver.database.utils import hash_field_name
+from apiserver.service_repo.auth import Identity
+from apiserver.utilities.dicts import nested_get, nested_set
+from apiserver.utilities.parameter_key_escaper import mongoengine_safe
+
+
+def get_artifact_id(artifact: dict):
+    """
+    Calculate id from 'key' and 'mode' fields
+    Return hash on on the id so that it will not contain mongo illegal characters
+    """
+    key_hash: str = hash_field_name(artifact["key"])
+    mode: str = artifact.get("mode", DEFAULT_ARTIFACT_MODE)
+    return f"{key_hash}_{mode}"
+
+
+def artifacts_prepare_for_save(fields: dict):
+    artifacts_field = ("execution", "artifacts")
+    artifacts = nested_get(fields, artifacts_field)
+    if artifacts is None:
+        return
+
+    nested_set(
+        fields, artifacts_field, value={get_artifact_id(a): a for a in artifacts}
+    )
+
+
+def artifacts_unprepare_from_saved(fields):
+    artifacts_field = ("execution", "artifacts")
+    artifacts = nested_get(fields, artifacts_field)
+    if artifacts is None:
+        return
+
+    nested_set(
+        fields,
+        artifacts_field,
+        value=sorted(artifacts.values(), key=itemgetter("key")),
+    )
+
+
+class Artifacts:
+    @classmethod
+    def add_or_update_artifacts(
+        cls,
+        company_id: str,
+        identity: Identity,
+        task_id: str,
+        artifacts: Sequence[ApiArtifact],
+        force: bool,
+    ) -> int:
+        task = get_task_for_update(
+            company_id=company_id, task_id=task_id, force=force, identity=identity
+        )
+
+        artifacts = {
+            get_artifact_id(a): Artifact(**a)
+            for a in (api_artifact.to_struct() for api_artifact in artifacts)
+        }
+
+        update_cmds = {
+            f"set__execution__artifacts__{mongoengine_safe(name)}": value
+            for name, value in artifacts.items()
+        }
+        return update_task(task, user_id=identity.user, update_cmds=update_cmds)
+
+    @classmethod
+    def delete_artifacts(
+        cls,
+        company_id: str,
+        identity: Identity,
+        task_id: str,
+        artifact_ids: Sequence[ArtifactId],
+        force: bool,
+    ) -> int:
+        task = get_task_for_update(
+            company_id=company_id, task_id=task_id, force=force, identity=identity
+        )
+
+        artifact_ids = [
+            get_artifact_id(a)
+            for a in (artifact_id.to_struct() for artifact_id in artifact_ids)
+        ]
+        delete_cmds = {
+            f"unset__execution__artifacts__{id_}": 1 for id_ in set(artifact_ids)
+        }
+
+        return update_task(task, user_id=identity.user, update_cmds=delete_cmds)

apiserver/bll/task/hyperparams.py

@@ -1,21 +1,25 @@
-from datetime import datetime
 from itertools import chain
 from operator import attrgetter
 from typing import Sequence, Dict
 
 from boltons import iterutils
 
-from apierrors import errors
-from apimodels.tasks import (
+from apiserver.apierrors import errors
+from apiserver.apimodels.tasks import (
     HyperParamKey,
     HyperParamItem,
     ReplaceHyperparams,
     Configuration,
 )
-from bll.task import TaskBLL
-from config import config
-from database.model.task.task import ParamsItem, Task, ConfigurationItem, TaskStatus
-from utilities.parameter_key_escaper import ParameterKeyEscaper
+from apiserver.bll.task import TaskBLL
+from apiserver.bll.task.utils import get_task_for_update, update_task
+from apiserver.config_repo import config
+from apiserver.database.model.task.task import ParamsItem, Task, ConfigurationItem
+from apiserver.service_repo.auth import Identity
+from apiserver.utilities.parameter_key_escaper import (
+    ParameterKeyEscaper,
+    mongoengine_safe,
+)
 
 log = config.logger(__file__)
 task_bll = TaskBLL()
@@ -28,7 +32,10 @@ class HyperParams:
     def get_params(cls, company_id: str, task_ids: Sequence[str]) -> Dict[str, dict]:
         only = ("id", "hyperparams")
         tasks = task_bll.assert_exists(
-            company_id=company_id, task_ids=task_ids, only=only, allow_public=True,
+            company_id=company_id,
+            task_ids=task_ids,
+            only=only,
+            allow_public=True,
         )
 
         return {
@@ -58,11 +65,20 @@ class HyperParams:
 
     @classmethod
     def delete_params(
-        cls, company_id: str, task_id: str, hyperparams=Sequence[HyperParamKey]
+        cls,
+        company_id: str,
+        identity: Identity,
+        task_id: str,
+        hyperparams: Sequence[HyperParamKey],
+        force: bool,
     ) -> int:
         properties_only = cls._normalize_params(hyperparams)
-        task = cls._get_task_for_update(
-            company=company_id, id=task_id, allow_all_statuses=properties_only
+        task = get_task_for_update(
+            company_id=company_id,
+            task_id=task_id,
+            allow_all_statuses=properties_only,
+            force=force,
+            identity=identity,
         )
 
         with_param, without_param = iterutils.partition(
@@ -83,19 +99,30 @@ class HyperParams:
             name = ParameterKeyEscaper.escape(item.name)
             delete_cmds[f"unset__hyperparams__{section}__{name}"] = 1
 
-        return task.update(**delete_cmds, last_update=datetime.utcnow())
+        return update_task(
+            task,
+            user_id=identity.user,
+            update_cmds=delete_cmds,
+            set_last_update=not properties_only,
+        )
 
     @classmethod
     def edit_params(
         cls,
         company_id: str,
+        identity: Identity,
         task_id: str,
         hyperparams: Sequence[HyperParamItem],
         replace_hyperparams: str,
+        force: bool,
     ) -> int:
         properties_only = cls._normalize_params(hyperparams)
-        task = cls._get_task_for_update(
-            company=company_id, id=task_id, allow_all_statuses=properties_only
+        task = get_task_for_update(
+            company_id=company_id,
+            task_id=task_id,
+            allow_all_statuses=properties_only,
+            force=force,
+            identity=identity,
         )
 
         update_cmds = dict()
@@ -104,13 +131,20 @@ class HyperParams:
             update_cmds["set__hyperparams"] = hyperparams
         elif replace_hyperparams == ReplaceHyperparams.section:
             for section, value in hyperparams.items():
-                update_cmds[f"set__hyperparams__{section}"] = value
+                update_cmds[f"set__hyperparams__{mongoengine_safe(section)}"] = value
         else:
             for section, section_params in hyperparams.items():
                 for name, value in section_params.items():
-                    update_cmds[f"set__hyperparams__{section}__{name}"] = value
+                    update_cmds[
+                        f"set__hyperparams__{section}__{mongoengine_safe(name)}"
+                    ] = value
 
-        return task.update(**update_cmds, last_update=datetime.utcnow())
+        return update_task(
+            task,
+            user_id=identity.user,
+            update_cmds=update_cmds,
+            set_last_update=not properties_only,
+        )
 
     @classmethod
     def _db_dicts_from_list(cls, items: Sequence[HyperParamItem]) -> Dict[str, dict]:
@@ -135,7 +169,10 @@ class HyperParams:
         else:
             only.append("configuration")
         tasks = task_bll.assert_exists(
-            company_id=company_id, task_ids=task_ids, only=only, allow_public=True,
+            company_id=company_id,
+            task_ids=task_ids,
+            only=only,
+            allow_public=True,
         )
 
         return {
@@ -150,17 +187,19 @@ class HyperParams:
 
     @classmethod
     def get_configuration_names(
-        cls, company_id: str, task_ids: Sequence[str]
+        cls, company_id: str, task_ids: Sequence[str], skip_empty: bool
     ) -> Dict[str, list]:
+        skip_empty_condition = {"$match": {"items.v.value": {"$nin": [None, ""]}}}
         pipeline = [
             {
                 "$match": {
-                    "company": {"$in": [None, "", company_id]},
+                    "company": {"$in": ["", company_id]},
                     "_id": {"$in": task_ids},
                 }
             },
             {"$project": {"items": {"$objectToArray": "$configuration"}}},
             {"$unwind": "$items"},
+            *([skip_empty_condition] if skip_empty else []),
             {"$group": {"_id": "$_id", "names": {"$addToSet": "$items.k"}}},
         ]
 
@@ -179,11 +218,15 @@ class HyperParams:
     def edit_configuration(
         cls,
         company_id: str,
+        identity: Identity,
         task_id: str,
         configuration: Sequence[Configuration],
         replace_configuration: bool,
+        force: bool,
     ) -> int:
-        task = cls._get_task_for_update(company=company_id, id=task_id)
+        task = get_task_for_update(
+            company_id=company_id, task_id=task_id, force=force, identity=identity
+        )
 
         update_cmds = dict()
         configuration = {
@@ -194,36 +237,26 @@ class HyperParams:
             update_cmds["set__configuration"] = configuration
         else:
             for name, value in configuration.items():
-                update_cmds[f"set__configuration__{name}"] = value
+                update_cmds[f"set__configuration__{mongoengine_safe(name)}"] = value
 
-        return task.update(**update_cmds, last_update=datetime.utcnow())
+        return update_task(task, user_id=identity.user, update_cmds=update_cmds)
 
     @classmethod
     def delete_configuration(
-        cls, company_id: str, task_id: str, configuration=Sequence[str]
+        cls,
+        company_id: str,
+        identity: Identity,
+        task_id: str,
+        configuration: Sequence[str],
+        force: bool,
     ) -> int:
-        task = cls._get_task_for_update(company=company_id, id=task_id)
+        task = get_task_for_update(
+            company_id=company_id, task_id=task_id, force=force, identity=identity
+        )
 
         delete_cmds = {
             f"unset__configuration__{ParameterKeyEscaper.escape(name)}": 1
             for name in set(configuration)
         }
 
-        return task.update(**delete_cmds, last_update=datetime.utcnow())
-
-    @staticmethod
-    def _get_task_for_update(
-        company: str, id: str, allow_all_statuses: bool = False
-    ) -> Task:
-        task = Task.get_for_writing(company=company, id=id, _only=("id", "status"))
-        if not task:
-            raise errors.bad_request.InvalidTaskId(id=id)
-
-        if allow_all_statuses:
-            return task
-
-        if task.status != TaskStatus.created:
-            raise errors.bad_request.InvalidTaskStatus(
-                expected=TaskStatus.created, status=task.status
-            )
-        return task
+        return update_task(task, user_id=identity.user, update_cmds=delete_cmds)

apiserver/bll/task/non_responsive_tasks_watchdog.py

@@ -1,11 +1,10 @@
 from datetime import timedelta, datetime
 from time import sleep
 
-from apierrors import errors
-from bll.task import ChangeStatusRequest
-from config import config
-from database.model.task.task import TaskStatus, Task
-from utilities.threads_manager import ThreadsManager
+from apiserver.bll.util import update_project_time
+from apiserver.config_repo import config
+from apiserver.database.model.task.task import TaskStatus, Task
+from apiserver.utilities.threads_manager import ThreadsManager
 
 log = config.logger(__file__)
 
@@ -40,7 +39,7 @@ class NonResponsiveTasksWatchdog:
     @threads.register("non_responsive_tasks_watchdog", daemon=True)
     def start(cls):
         sleep(cls.settings.watch_interval_sec)
-        while not ThreadsManager.terminating:
+        while True:
             watch_interval = cls.settings.watch_interval_sec
             if cls.settings.enabled:
                 try:
@@ -71,19 +70,30 @@ class NonResponsiveTasksWatchdog:
             return 0
 
         err_count = 0
+        project_ids = set()
+        now = datetime.utcnow()
         for task in tasks:
             log.info(
                 f"Stopping {task.id} ({task.name}), last updated at {task.last_update}"
             )
+            # noinspection PyBroadException
             try:
-                ChangeStatusRequest(
-                    task=task,
-                    new_status=TaskStatus.stopped,
+                updated = Task.objects(id=task.id, status=task.status).update(
+                    status=TaskStatus.stopped,
                     status_reason="Forced stop (non-responsive)",
                     status_message="Forced stop (non-responsive)",
-                    force=True,
-                ).execute()
-            except errors.bad_request.FailedChangingTaskStatus:
-                err_count += 1
+                    status_changed=now,
+                    last_update=now,
+                    last_change=now,
+                    last_changed_by="__apiserver__",
+                )
+                if updated:
+                    project_ids.add(task.project)
+                else:
+                    err_count += 1
+            except Exception as ex:
+                log.error("Failed setting status: %s", str(ex))
+
+        update_project_time(list(project_ids))
 
         return len(tasks) - err_count

apiserver/bll/task/param_utils.py

@@ -1,12 +1,11 @@
 import itertools
-from typing import Sequence, Tuple
+from typing import Sequence, Tuple, Optional
 
-import dpath
 
-from apierrors import errors
-from database.model.task.task import Task
-from tools import safe_get
-from utilities.parameter_key_escaper import ParameterKeyEscaper
+from apiserver.apierrors import errors
+from apiserver.database.model.task.task import Task
+from apiserver.utilities.dicts import nested_get, nested_delete, nested_set
+from apiserver.utilities.parameter_key_escaper import ParameterKeyEscaper
 
 
 hyperparams_default_section = "Args"
@@ -14,7 +13,7 @@ hyperparams_legacy_type = "legacy"
 tf_define_section = "TF_DEFINE"
 
 
-def split_param_name(full_name: str, default_section: str) -> Tuple[str, str]:
+def split_param_name(full_name: str, default_section: str) -> Tuple[Optional[str], str]:
     """
     Return parameter section and name. The section is either TF_DEFINE or the default one
     """
@@ -62,7 +61,7 @@ def _remove_legacy_params(data: dict, with_sections: bool = False) -> int:
     return removed
 
 
-def _get_legacy_params(data: dict, with_sections: bool = False) -> Sequence[str]:
+def _get_legacy_params(data: dict, with_sections: bool = False) -> Sequence[dict]:
     """
     Remove the legacy params from the data dict and return the number of removed params
     If the path not found then return 0
@@ -71,8 +70,10 @@ def _get_legacy_params(data: dict, with_sections: bool = False) -> Sequence[str]
         return []
 
     if with_sections:
-        return itertools.chain.from_iterable(
-            _get_legacy_params(section_data) for section_data in data.values()
+        return list(
+            itertools.chain.from_iterable(
+                _get_legacy_params(section_data) for section_data in data.values()
+            )
         )
 
     return [
@@ -86,15 +87,15 @@ def params_prepare_for_save(fields: dict, previous_task: Task = None):
     Escape all the section and param names for hyper params and configuration to make it mongo sage
     """
     for old_params_field, new_params_field, default_section in (
-        ("execution/parameters", "hyperparams", hyperparams_default_section),
-        ("execution/model_desc", "configuration", None),
+        (("execution", "parameters"), "hyperparams", hyperparams_default_section),
+        (("execution", "model_desc"), "configuration", None),
     ):
-        legacy_params = safe_get(fields, old_params_field)
+        legacy_params = nested_get(fields, old_params_field)
         if legacy_params is None:
             continue
 
         if (
-            not safe_get(fields, new_params_field)
+            not fields.get(new_params_field)
             and previous_task
             and previous_task[new_params_field]
         ):
@@ -117,21 +118,34 @@ def params_prepare_for_save(fields: dict, previous_task: Task = None):
             new_param = dict(name=name, type=hyperparams_legacy_type, value=str(value))
             if section is not None:
                 new_param["section"] = section
-            dpath.new(fields, new_path, new_param)
-        dpath.delete(fields, old_params_field)
+            nested_set(fields, new_path, new_param)
+        nested_delete(fields, old_params_field)
 
-    for param_field in ("hyperparams", "configuration"):
-        params = safe_get(fields, param_field)
-        if params:
-            escaped_params = {
-                ParameterKeyEscaper.escape(key): {
-                    ParameterKeyEscaper.escape(k): v for k, v in value.items()
-                }
-                if isinstance(value, dict)
-                else value
-                for key, value in params.items()
+    def ensure_non_empty(k: str, desc: str) -> str:
+        if not k:
+            raise errors.bad_request.ValidationError(
+                f"Empty {desc} name is not allowed"
+            )
+        return k
+
+    params = fields.get("hyperparams")
+    if params:
+        escaped_params = {
+            ParameterKeyEscaper.escape(ensure_non_empty(key, "section")): {
+                ParameterKeyEscaper.escape(ensure_non_empty(k, "parameter")): v
+                for k, v in value.items()
             }
-            dpath.set(fields, param_field, escaped_params)
+            for key, value in params.items()
+        }
+        fields["hyperparams"] = escaped_params
+
+    params = fields.get("configuration")
+    if params:
+        escaped_params = {
+            ParameterKeyEscaper.escape(ensure_non_empty(key, "configuration")): value
+            for key, value in params.items()
+        }
+        fields["configuration"] = escaped_params
 
 
 def params_unprepare_from_saved(fields, copy_to_legacy=False):
@@ -140,7 +154,7 @@ def params_unprepare_from_saved(fields, copy_to_legacy=False):
     If copy_to_legacy is set then copy hyperparams and configuration data to the legacy location for the old clients
     """
     for param_field in ("hyperparams", "configuration"):
-        params = safe_get(fields, param_field)
+        params = fields.get(param_field)
         if params:
             unescaped_params = {
                 ParameterKeyEscaper.unescape(key): {
@@ -150,18 +164,18 @@ def params_unprepare_from_saved(fields, copy_to_legacy=False):
                 else value
                 for key, value in params.items()
             }
-            dpath.set(fields, param_field, unescaped_params)
+            fields[param_field] = unescaped_params
 
     if copy_to_legacy:
         for new_params_field, old_params_field, use_sections in (
-            (f"hyperparams", "execution/parameters", True),
-            (f"configuration", "execution/model_desc", False),
+            ("hyperparams", ("execution", "parameters"), True),
+            ("configuration", ("execution", "model_desc"), False),
         ):
             legacy_params = _get_legacy_params(
-                safe_get(fields, new_params_field), with_sections=use_sections
+                fields.get(new_params_field), with_sections=use_sections
             )
             if legacy_params:
-                dpath.new(
+                nested_set(
                     fields,
                     old_params_field,
                     {_get_full_param_name(p): p["value"] for p in legacy_params},
@@ -174,7 +188,7 @@ def _process_path(path: str):
     Need to unescape and apply a full mongo escaping
     """
     parts = path.split(".")
-    if len(parts) < 2 or len(parts) > 3:
+    if len(parts) < 2 or len(parts) > 4:
         raise errors.bad_request.ValidationError("invalid task field", path=path)
     return ".".join(
         ParameterKeyEscaper.escape(ParameterKeyEscaper.unescape(p)) for p in parts
@@ -184,7 +198,8 @@ def _process_path(path: str):
 def escape_paths(paths: Sequence[str]) -> Sequence[str]:
     for old_prefix, new_prefix in (
         ("execution.parameters", f"hyperparams.{hyperparams_default_section}"),
-        ("execution.model_desc", f"configuration"),
+        ("execution.model_desc", "configuration"),
+        ("execution.docker_cmd", "container"),
     ):
         path: str
         paths = [path.replace(old_prefix, new_prefix) for path in paths]

apiserver/bll/task/task_bll.py

@@ -0,0 +1,540 @@
+from datetime import datetime
+from typing import Collection, Sequence, Tuple, Optional, Dict
+
+import six
+from mongoengine import Q
+from redis import StrictRedis
+from six import string_types
+
+import apiserver.database.utils as dbutils
+from apiserver.apierrors import errors, APIError
+from apiserver.apimodels.tasks import TaskInputModel
+from apiserver.bll.queue import QueueBLL
+from apiserver.bll.organization import OrgBLL, Tags
+from apiserver.bll.project import ProjectBLL
+from apiserver.bll.util import update_project_time
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.model import Model
+from apiserver.database.model.project import Project
+from apiserver.database.model.task.metrics import EventStats, MetricEventStats
+from apiserver.database.model.task.output import Output
+from apiserver.database.model.task.task import (
+    Task,
+    TaskStatus,
+    TaskSystemTags,
+    ArtifactModes,
+    ModelItem,
+    Models,
+    DEFAULT_ARTIFACT_MODE,
+    TaskModelNames,
+    TaskModelTypes,
+)
+from apiserver.database.model import EntityVisibility
+from apiserver.database.model.queue import Queue
+from apiserver.database.utils import (
+    get_company_or_none_constraint,
+    id as create_id,
+)
+from apiserver.es_factory import es_factory
+from apiserver.redis_manager import redman
+from apiserver.services.utils import validate_tags, escape_dict_field, escape_dict
+from apiserver.utilities.dicts import nested_set
+from .artifacts import artifacts_prepare_for_save
+from .param_utils import params_prepare_for_save
+from .utils import (
+    ChangeStatusRequest,
+    deleted_prefix,
+    get_last_metric_updates,
+)
+
+log = config.logger(__file__)
+org_bll = OrgBLL()
+queue_bll = QueueBLL()
+project_bll = ProjectBLL()
+
+
+class TaskBLL:
+    def __init__(self, events_es=None, redis=None):
+        self.events_es = events_es or es_factory.connect("events")
+        self.redis: StrictRedis = redis or redman.connection("apiserver")
+
+    @staticmethod
+    def get_by_id(
+        company_id,
+        task_id,
+        required_status=None,
+        only_fields=None,
+        allow_public=False,
+    ):
+        if only_fields:
+            if isinstance(only_fields, string_types):
+                only_fields = [only_fields]
+            else:
+                only_fields = list(only_fields)
+            only_fields = only_fields + ["status"]
+
+        tasks = Task.get_many(
+            company=company_id,
+            query=Q(id=task_id),
+            allow_public=allow_public,
+            override_projection=only_fields,
+            return_dicts=False,
+        )
+        task = None if not tasks else tasks[0]
+
+        if not task:
+            raise errors.bad_request.InvalidTaskId(id=task_id)
+
+        if required_status and not task.status == required_status:
+            raise errors.bad_request.InvalidTaskStatus(expected=required_status)
+
+        return task
+
+    @staticmethod
+    def assert_exists(
+        company_id, task_ids, only=None, allow_public=False, return_tasks=True
+    ) -> Optional[Sequence[Task]]:
+        task_ids = [task_ids] if isinstance(task_ids, six.string_types) else task_ids
+        with translate_errors_context():
+            ids = set(task_ids)
+            q = Task.get_many(
+                company=company_id,
+                query=Q(id__in=ids),
+                allow_public=allow_public,
+                return_dicts=False,
+            )
+            if only:
+                # Make sure to reset fields filters (some fields are excluded by default) since this
+                # is an internal call and specific fields were requested.
+                q = q.all_fields().only(*only)
+
+            if q.count() != len(ids):
+                raise errors.bad_request.InvalidTaskId(ids=task_ids)
+
+            if return_tasks:
+                return list(q)
+
+    @staticmethod
+    def create(company: str, user: str, fields: dict):
+        now = datetime.utcnow()
+        return Task(
+            id=create_id(),
+            user=user,
+            company=company,
+            created=now,
+            last_update=now,
+            last_change=now,
+            last_changed_by=user,
+            **fields,
+        )
+
+    @staticmethod
+    def validate_input_models(task, allow_only_public=False):
+        if not task.models.input:
+            return
+
+        company = None if allow_only_public else task.company
+        model_ids = set(m.model for m in task.models.input)
+        models = Model.objects(
+            Q(id__in=model_ids) & get_company_or_none_constraint(company)
+        ).only("id")
+        missing = model_ids - {m.id for m in models}
+        if missing:
+            raise errors.bad_request.InvalidModelId(models=missing)
+
+        return
+
+    @classmethod
+    def clone_task(
+        cls,
+        company_id: str,
+        user_id: str,
+        task_id: str,
+        name: Optional[str] = None,
+        comment: Optional[str] = None,
+        parent: Optional[str] = None,
+        project: Optional[str] = None,
+        tags: Optional[Sequence[str]] = None,
+        system_tags: Optional[Sequence[str]] = None,
+        hyperparams: Optional[dict] = None,
+        configuration: Optional[dict] = None,
+        container: Optional[dict] = None,
+        execution_overrides: Optional[dict] = None,
+        input_models: Optional[Sequence[TaskInputModel]] = None,
+        validate_references: bool = False,
+        new_project_name: str = None,
+        hyperparams_overrides: Optional[dict] = None,
+        configuration_overrides: Optional[dict] = None,
+    ) -> Tuple[Task, dict]:
+        validate_tags(tags, system_tags)
+        task: Task = cls.get_by_id(
+            company_id=company_id, task_id=task_id, allow_public=True
+        )
+
+        params_dict = {}
+        if hyperparams:
+            params_dict["hyperparams"] = hyperparams
+        elif hyperparams_overrides:
+            updated_hyperparams = {
+                sec: {k: value for k, value in sec_data.items()}
+                for sec, sec_data in (task.hyperparams or {}).items()
+            }
+            for section, section_data in hyperparams_overrides.items():
+                for key, value in section_data.items():
+                    nested_set(updated_hyperparams, (section, key), value)
+            params_dict["hyperparams"] = updated_hyperparams
+
+        if configuration:
+            params_dict["configuration"] = configuration
+        elif configuration_overrides:
+            updated_configuration = {
+                k: value for k, value in (task.configuration or {}).items()
+            }
+            for key, value in configuration_overrides.items():
+                updated_configuration[key] = value
+            params_dict["configuration"] = updated_configuration
+
+        now = datetime.utcnow()
+        if input_models:
+            input_models = [
+                ModelItem(model=m.model, name=m.name, updated=now) for m in input_models
+            ]
+
+        execution_dict = task.execution.to_proper_dict() if task.execution else {}
+        if execution_overrides:
+            execution_model = execution_overrides.pop("model", None)
+            if not input_models and execution_model:
+                input_models = [
+                    ModelItem(
+                        model=execution_model,
+                        name=TaskModelNames[TaskModelTypes.input],
+                        updated=now,
+                    )
+                ]
+
+            docker_cmd = execution_overrides.pop("docker_cmd", None)
+            if not container and docker_cmd:
+                image, _, arguments = docker_cmd.partition(" ")
+                container = {"image": image, "arguments": arguments}
+
+            artifacts_prepare_for_save({"execution": execution_overrides})
+
+            params_dict["execution"] = {}
+            for legacy_param in ("parameters", "configuration"):
+                legacy_value = execution_overrides.pop(legacy_param, None)
+                if legacy_value is not None:
+                    params_dict["execution"] = legacy_value
+
+            escape_dict_field(execution_overrides, "model_labels")
+
+            execution_dict.update(execution_overrides)
+
+        params_prepare_for_save(params_dict, previous_task=task)
+
+        artifacts = execution_dict.get("artifacts")
+        if artifacts:
+            execution_dict["artifacts"] = {
+                k: a
+                for k, a in artifacts.items()
+                if a.get("mode", DEFAULT_ARTIFACT_MODE) != ArtifactModes.output
+            }
+        execution_dict.pop("queue", None)
+
+        new_project_data = None
+        if not project and new_project_name:
+            # Use a project with the provided name, or create a new project
+            project = ProjectBLL.find_or_create(
+                project_name=new_project_name,
+                user=user_id,
+                company=company_id,
+                description="",
+            )
+            new_project_data = {"id": project, "name": new_project_name}
+
+        def clean_system_tags(input_tags: Sequence[str]) -> Sequence[str]:
+            if not input_tags:
+                return input_tags
+
+            return [
+                tag
+                for tag in input_tags
+                if tag
+                not in [TaskSystemTags.development, EntityVisibility.archived.value]
+            ]
+
+        def ensure_int_labels(execution: dict) -> dict:
+            if not execution:
+                return execution
+
+            model_labels = execution.get("model_labels")
+            if model_labels:
+                execution["model_labels"] = {k: int(v) for k, v in model_labels.items()}
+
+            return execution
+
+        parent_task = (
+            task.parent
+            if task.parent and not task.parent.startswith(deleted_prefix)
+            else task.id
+        )
+        new_task = Task(
+            id=create_id(),
+            user=user_id,
+            company=company_id,
+            created=now,
+            last_update=now,
+            last_change=now,
+            last_changed_by=user_id,
+            name=name or task.name,
+            comment=comment or task.comment,
+            parent=parent or parent_task,
+            project=project or task.project,
+            tags=tags or task.tags,
+            system_tags=system_tags or clean_system_tags(task.system_tags),
+            type=task.type,
+            script=task.script,
+            output=Output(destination=task.output.destination) if task.output else None,
+            models=Models(input=input_models or task.models.input),
+            container=escape_dict(container) or task.container,
+            execution=ensure_int_labels(execution_dict),
+            configuration=params_dict.get("configuration") or task.configuration,
+            hyperparams=params_dict.get("hyperparams") or task.hyperparams,
+        )
+        cls.validate(
+            new_task,
+            validate_models=validate_references or input_models,
+            validate_parent=validate_references or parent,
+            validate_project=validate_references or project,
+        )
+        new_task.save()
+
+        if task.project == new_task.project:
+            updated_tags = tags
+            updated_system_tags = system_tags
+        else:
+            updated_tags = new_task.tags
+            updated_system_tags = new_task.system_tags
+        org_bll.update_tags(
+            company_id,
+            Tags.Task,
+            projects=[new_task.project],
+            tags=updated_tags,
+            system_tags=updated_system_tags,
+        )
+        update_project_time(new_task.project)
+
+        return new_task, new_project_data
+
+    @classmethod
+    def validate(
+        cls,
+        task: Task,
+        validate_models=True,
+        validate_parent=True,
+        validate_project=True,
+    ):
+        """
+        Validate task properties according to the flag
+        Task project is always checked for being writable
+        in order to disable the modification of public projects
+        """
+        if (
+            validate_parent
+            and task.parent
+            and not task.parent.startswith(deleted_prefix)
+            and not Task.get(
+                company=task.company, id=task.parent, _only=("id",), include_public=True
+            )
+        ):
+            raise errors.bad_request.InvalidTaskId("invalid parent", parent=task.parent)
+
+        if task.project:
+            project = Project.get_for_writing(company=task.company, id=task.project)
+            if validate_project and not project:
+                raise errors.bad_request.InvalidProjectId(id=task.project)
+
+        if validate_models:
+            cls.validate_input_models(task)
+
+    @staticmethod
+    def set_last_update(
+        task_ids: Collection[str],
+        company_id: str,
+        user_id: str,
+        last_update: datetime,
+        **extra_updates,
+    ):
+        tasks = Task.objects(id__in=task_ids, company=company_id).only(
+            "status", "started"
+        )
+        count = 0
+        for task in tasks:
+            updates = extra_updates
+            if task.status == TaskStatus.in_progress and task.started:
+                updates = {
+                    "active_duration": (
+                        datetime.utcnow() - task.started
+                    ).total_seconds(),
+                    **extra_updates,
+                }
+            count += Task.objects(id=task.id, company=company_id).update(
+                upsert=False,
+                last_update=last_update,
+                last_change=last_update,
+                last_changed_by=user_id,
+                **updates,
+            )
+        return count
+
+    @staticmethod
+    def update_statistics(
+        task_id: str,
+        company_id: str,
+        user_id: str,
+        last_update: datetime = None,
+        last_iteration: int = None,
+        last_iteration_max: int = None,
+        last_scalar_events: Dict[str, Dict[str, dict]] = None,
+        last_events: Dict[str, Dict[str, dict]] = None,
+        **extra_updates,
+    ):
+        """
+        Update task statistics
+        :param task_id: Task's ID.
+        :param company_id: Task's company ID.
+        :param last_update: Last update time. If not provided, defaults to datetime.utcnow().
+        :param last_iteration: Last reported iteration. Use this to set a value regardless of current
+            task's last iteration value.
+        :param last_iteration_max: Last reported iteration. Use this to conditionally set a value only
+            if the current task's last iteration value is smaller than the provided value.
+        :param last_scalar_events: Last reported metrics summary for scalar events (value, metric, variant).
+        :param last_events: Last reported metrics summary (value, metric, event type).
+        :param extra_updates: Extra task updates to include in this update call.
+        :return:
+        """
+        last_update = last_update or datetime.utcnow()
+
+        if last_iteration is not None:
+            extra_updates.update(last_iteration=last_iteration)
+        elif last_iteration_max is not None:
+            extra_updates.update(max__last_iteration=last_iteration_max)
+
+        raw_updates = {}
+        if last_scalar_events is not None:
+            get_last_metric_updates(
+                task_id=task_id,
+                last_scalar_events=last_scalar_events,
+                raw_updates=raw_updates,
+                extra_updates=extra_updates,
+            )
+
+        if last_events is not None:
+
+            def events_per_type(metric_data_: Dict[str, dict]) -> Dict[str, EventStats]:
+                return {
+                    event_type: EventStats(last_update=event["timestamp"])
+                    for event_type, event in metric_data_.items()
+                }
+
+            metric_stats = {
+                dbutils.hash_field_name(metric_key): MetricEventStats(
+                    metric=metric_key, event_stats_by_type=events_per_type(metric_data)
+                )
+                for metric_key, metric_data in last_events.items()
+            }
+            extra_updates["metric_stats"] = metric_stats
+
+        ret = TaskBLL.set_last_update(
+            task_ids=[task_id],
+            company_id=company_id,
+            user_id=user_id,
+            last_update=last_update,
+            **extra_updates,
+        )
+        if ret and raw_updates:
+            Task.objects(id=task_id).update_one(__raw__=[{"$set": raw_updates}])
+
+        return ret
+
+    @staticmethod
+    def remove_task_from_all_queues(
+        company_id: str, task_id: str, exclude: str = None
+    ) -> int:
+        more = {}
+        if exclude:
+            more["id__ne"] = exclude
+        return Queue.objects(company=company_id, entries__task=task_id, **more).update(
+            pull__entries__task=task_id, last_update=datetime.utcnow()
+        )
+
+    @classmethod
+    def dequeue_and_change_status(
+        cls,
+        task: Task,
+        company_id: str,
+        user_id: str,
+        status_message: str,
+        status_reason: str,
+        remove_from_all_queues=False,
+        new_status=None,
+        new_status_for_aborted_task=None,
+    ):
+        try:
+            cls.dequeue(task, company_id=company_id, user_id=user_id, silent_fail=True)
+        except APIError:
+            # dequeue may fail if the queue was deleted
+            pass
+
+        if remove_from_all_queues:
+            cls.remove_task_from_all_queues(company_id=company_id, task_id=task.id)
+
+        if task.status not in [TaskStatus.queued, TaskStatus.in_progress]:
+            return {"updated": 0}
+
+        if new_status_for_aborted_task and task.status == TaskStatus.in_progress:
+            new_status = new_status_for_aborted_task
+
+        return ChangeStatusRequest(
+            task=task,
+            new_status=new_status or task.enqueue_status or TaskStatus.created,
+            status_reason=status_reason,
+            status_message=status_message,
+            user_id=user_id,
+            force=True,
+        ).execute(enqueue_status=None)
+
+    @classmethod
+    def dequeue(cls, task: Task, company_id: str, user_id: str, silent_fail=False):
+        """
+        Dequeue the task from the queue
+        :param task: task to dequeue
+        :param company_id: task's company ID.
+        :param silent_fail: do not throw exceptions. APIError is still thrown
+        :raise errors.bad_request.InvalidTaskId: if the task's status is not queued
+        :raise errors.bad_request.MissingRequiredFields: if the task is not queued
+        :raise APIError or errors.server_error.TransactionError: if internal call to queues.remove_task fails
+        :return: the result of queues.remove_task call. None in case of silent failure
+        """
+        if task.status not in (TaskStatus.queued,):
+            if silent_fail:
+                return
+            raise errors.bad_request.InvalidTaskId(
+                status=task.status, expected=TaskStatus.queued
+            )
+
+        if not task.execution or not task.execution.queue:
+            if silent_fail:
+                return
+            raise errors.bad_request.MissingRequiredFields(
+                "task has no queue value", field="execution.queue"
+            )
+
+        return {
+            "removed": queue_bll.remove_task(
+                company_id=company_id,
+                user_id=user_id,
+                queue_id=task.execution.queue,
+                task_id=task.id,
+            )
+        }

apiserver/bll/task/task_cleanup.py

@@ -0,0 +1,374 @@
+from datetime import datetime
+from itertools import chain
+from operator import attrgetter
+from typing import Sequence, Set, Tuple, Union
+
+import attr
+from boltons.iterutils import partition, bucketize, first, chunked_iter
+from furl import furl
+from mongoengine import NotUniqueError
+from pymongo.errors import DuplicateKeyError
+
+from apiserver.apierrors import errors
+from apiserver.bll.event import EventBLL
+from apiserver.bll.event.event_bll import PlotFields
+from apiserver.bll.task.utils import deleted_prefix
+from apiserver.config_repo import config
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task, TaskStatus, ArtifactModes
+from apiserver.database.model.url_to_delete import (
+    StorageType,
+    UrlToDelete,
+    FileType,
+    DeletionStatus,
+)
+from apiserver.database.utils import id as db_id
+
+log = config.logger(__file__)
+event_bll = EventBLL()
+
+
+@attr.s(auto_attribs=True)
+class TaskUrls:
+    model_urls: Sequence[str]
+    artifact_urls: Sequence[str]
+    event_urls: Sequence[str] = []  # left here is in order not to break the api
+
+    def __add__(self, other: "TaskUrls"):
+        if not other:
+            return self
+
+        return TaskUrls(
+            model_urls=list(set(self.model_urls) | set(other.model_urls)),
+            artifact_urls=list(set(self.artifact_urls) | set(other.artifact_urls)),
+        )
+
+
+@attr.s(auto_attribs=True)
+class CleanupResult:
+    """
+    Counts of objects modified in task cleanup operation
+    """
+
+    updated_children: int
+    updated_models: int
+    deleted_models: int
+    deleted_model_ids: Set[str]
+    urls: TaskUrls = None
+
+    def to_res_dict(self, return_file_urls: bool) -> dict:
+        remove_fields = ["deleted_model_ids"]
+        if not return_file_urls:
+            remove_fields.append("urls")
+
+        # noinspection PyTypeChecker
+        res = attr.asdict(
+            self, filter=lambda attrib, value: attrib.name not in remove_fields
+        )
+        if not return_file_urls:
+            res["urls"] = None
+
+        return res
+
+    def __add__(self, other: "CleanupResult"):
+        if not other:
+            return self
+
+        return CleanupResult(
+            updated_children=self.updated_children + other.updated_children,
+            updated_models=self.updated_models + other.updated_models,
+            deleted_models=self.deleted_models + other.deleted_models,
+            urls=self.urls + other.urls if self.urls else other.urls,
+            deleted_model_ids=self.deleted_model_ids | other.deleted_model_ids,
+        )
+
+    @staticmethod
+    def empty():
+        return CleanupResult(
+            updated_children=0,
+            updated_models=0,
+            deleted_models=0,
+            deleted_model_ids=set(),
+        )
+
+
+def collect_plot_image_urls(
+    company: str, task_or_model: Union[str, Sequence[str]]
+) -> Set[str]:
+    urls = set()
+    task_ids = task_or_model if isinstance(task_or_model, list) else [task_or_model]
+    for tasks in chunked_iter(task_ids, 100):
+        next_scroll_id = None
+        while True:
+            events, next_scroll_id = event_bll.get_plot_image_urls(
+                company_id=company, task_ids=tasks, scroll_id=next_scroll_id
+            )
+            if not events:
+                break
+            for event in events:
+                event_urls = event.get(PlotFields.source_urls)
+                if event_urls:
+                    urls.update(set(event_urls))
+
+    return urls
+
+
+def collect_debug_image_urls(
+    company: str, task_or_model: Union[str, Sequence[str]]
+) -> Set[str]:
+    """
+    Return the set of unique image urls
+    Uses DebugImagesIterator to make sure that we do not retrieve recycled urls
+    """
+    urls = set()
+    task_ids = task_or_model if isinstance(task_or_model, list) else [task_or_model]
+    for tasks in chunked_iter(task_ids, 100):
+        after_key = None
+        while True:
+            res, after_key = event_bll.get_debug_image_urls(
+                company_id=company,
+                task_ids=tasks,
+                after_key=after_key,
+            )
+            urls.update(res)
+            if not after_key:
+                break
+
+    return urls
+
+
+supported_storage_types = {
+    "s3://": StorageType.s3,
+    "azure://": StorageType.azure,
+    "gs://": StorageType.gs,
+}
+
+supported_storage_types.update(
+    {
+        p: StorageType.fileserver
+        for p in config.get(
+            "services.async_urls_delete.fileserver.url_prefixes",
+            ["https://", "http://"],
+        )
+    }
+)
+
+
+def schedule_for_delete(
+    company: str,
+    user: str,
+    task_id: str,
+    urls: Set[str],
+    can_delete_folders: bool,
+) -> Set[str]:
+    urls_per_storage = bucketize(
+        urls,
+        key=lambda u: first(
+            type_
+            for prefix, type_ in supported_storage_types.items()
+            if u.startswith(prefix)
+        ),
+    )
+    urls_per_storage.pop(None, None)
+
+    processed_urls = set()
+    for storage_type, storage_urls in urls_per_storage.items():
+        delete_folders = (storage_type == StorageType.fileserver) and can_delete_folders
+        scheduled_to_delete = set()
+        for url in storage_urls:
+            folder = None
+            if delete_folders:
+                try:
+                    parsed = furl(url)
+                    if parsed.path and len(parsed.path.segments) > 1:
+                        folder = parsed.remove(
+                            args=True, fragment=True, path=parsed.path.segments[-1]
+                        ).url.rstrip("/")
+                except Exception as ex:
+                    pass
+
+            to_delete = folder or url
+            if to_delete in scheduled_to_delete:
+                processed_urls.add(url)
+                continue
+
+            try:
+                UrlToDelete(
+                    id=db_id(),
+                    company=company,
+                    user=user,
+                    url=to_delete,
+                    task=task_id,
+                    created=datetime.utcnow(),
+                    storage_type=storage_type,
+                    type=FileType.folder if folder else FileType.file,
+                ).save()
+            except (DuplicateKeyError, NotUniqueError):
+                existing = UrlToDelete.objects(company=company, url=to_delete).first()
+                if existing:
+                    existing.update(
+                        user=user,
+                        task=task_id,
+                        created=datetime.utcnow(),
+                        retry_count=0,
+                        unset__last_failure_time=1,
+                        unset__last_failure_reason=1,
+                        status=DeletionStatus.created,
+                    )
+            processed_urls.add(url)
+            scheduled_to_delete.add(to_delete)
+
+    return processed_urls
+
+
+def delete_task_events_and_collect_urls(
+    company: str, task_ids: Sequence[str], wait_for_delete: bool, model=False
+) -> Set[str]:
+    event_urls = collect_debug_image_urls(company, task_ids) | collect_plot_image_urls(
+        company, task_ids
+    )
+
+    event_bll.delete_task_events(
+        company, task_ids, model=model, wait_for_delete=wait_for_delete
+    )
+
+    return event_urls
+
+
+def cleanup_task(
+    company: str,
+    user: str,
+    task: Task,
+    force: bool = False,
+    update_children=True,
+    delete_output_models=True,
+) -> CleanupResult:
+    """
+    Validate task deletion and delete/modify all its output.
+    :param task: task object
+    :param force: whether to delete task with published outputs
+    :return: count of delete and modified items
+    """
+    published_models, draft_models, in_use_model_ids = verify_task_children_and_ouptuts(
+        task, force
+    )
+    artifact_urls = (
+        {
+            a.uri
+            for a in task.execution.artifacts.values()
+            if a.mode == ArtifactModes.output and a.uri
+        }
+        if task.execution and task.execution.artifacts
+        else {}
+    )
+    model_urls = {m.uri for m in draft_models if m.uri and m.id not in in_use_model_ids}
+
+    deleted_task_id = f"{deleted_prefix}{task.id}"
+    updated_children = 0
+    now = datetime.utcnow()
+    if update_children:
+        updated_children = Task.objects(parent=task.id).update(
+            parent=deleted_task_id,
+            last_change=now,
+            last_changed_by=user,
+        )
+
+    deleted_models = 0
+    updated_models = 0
+    deleted_model_ids = set()
+    for models, allow_delete in ((draft_models, True), (published_models, False)):
+        if not models:
+            continue
+        if delete_output_models and allow_delete:
+            model_ids = list({m.id for m in models if m.id not in in_use_model_ids})
+            if model_ids:
+                deleted_models += Model.objects(id__in=model_ids).delete()
+                deleted_model_ids.update(model_ids)
+
+            if in_use_model_ids:
+                Model.objects(id__in=list(in_use_model_ids)).update(
+                    unset__task=1,
+                    set__last_change=now,
+                    set__last_changed_by=user,
+                )
+            continue
+
+        if update_children:
+            updated_models += Model.objects(id__in=[m.id for m in models]).update(
+                task=deleted_task_id,
+                last_change=now,
+                last_changed_by=user,
+            )
+        else:
+            Model.objects(id__in=[m.id for m in models]).update(
+                unset__task=1,
+                set__last_change=now,
+                set__last_changed_by=user,
+            )
+
+    return CleanupResult(
+        deleted_models=deleted_models,
+        updated_children=updated_children,
+        updated_models=updated_models,
+        urls=TaskUrls(
+            artifact_urls=list(artifact_urls),
+            model_urls=list(model_urls),
+        ),
+        deleted_model_ids=deleted_model_ids,
+    )
+
+
+def verify_task_children_and_ouptuts(
+    task, force: bool
+) -> Tuple[Sequence[Model], Sequence[Model], Set[str]]:
+    if not force:
+        published_children_count = Task.objects(
+            parent=task.id, status=TaskStatus.published
+        ).count()
+        if published_children_count:
+            raise errors.bad_request.TaskCannotBeDeleted(
+                "has children, use force=True",
+                task=task.id,
+                children=published_children_count,
+            )
+
+    model_fields = ["id", "ready", "uri"]
+    published_models, draft_models = partition(
+        Model.objects(task=task.id).only(*model_fields),
+        key=attrgetter("ready"),
+    )
+    if not force and published_models:
+        raise errors.bad_request.TaskCannotBeDeleted(
+            "has output models, use force=True",
+            task=task.id,
+            models=len(published_models),
+        )
+
+    if task.models and task.models.output:
+        model_ids = [m.model for m in task.models.output]
+        for output_model in Model.objects(id__in=model_ids).only(*model_fields):
+            if output_model.ready:
+                if not force:
+                    raise errors.bad_request.TaskCannotBeDeleted(
+                        "has output model, use force=True",
+                        task=task.id,
+                        model=output_model.id,
+                    )
+                published_models.append(output_model)
+            else:
+                draft_models.append(output_model)
+
+    in_use_model_ids = {}
+    if draft_models:
+        model_ids = {m.id for m in draft_models}
+        dependent_tasks = Task.objects(models__input__model__in=list(model_ids)).only(
+            "id", "models"
+        )
+        in_use_model_ids = model_ids & {
+            m.model
+            for m in chain.from_iterable(
+                t.models.input for t in dependent_tasks if t.models
+            )
+        }
+
+    return published_models, draft_models, in_use_model_ids

apiserver/bll/task/task_operations.py

@@ -0,0 +1,612 @@
+from datetime import datetime
+from typing import Callable, Any, Tuple, Union, Sequence
+
+from apiserver.apierrors import errors, APIError
+from apiserver.bll.queue import QueueBLL
+from apiserver.bll.task import (
+    TaskBLL,
+    validate_status_change,
+    ChangeStatusRequest,
+)
+from apiserver.bll.task.task_cleanup import cleanup_task, CleanupResult
+from apiserver.bll.task.utils import get_task_with_write_access
+from apiserver.bll.util import update_project_time
+from apiserver.config_repo import config
+from apiserver.database.model import EntityVisibility
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.output import Output
+from apiserver.database.model.task.task import (
+    TaskStatus,
+    Task,
+    TaskSystemTags,
+    TaskStatusMessage,
+    ArtifactModes,
+    Execution,
+    DEFAULT_LAST_ITERATION,
+    TaskType,
+)
+from apiserver.database.utils import get_options
+from apiserver.service_repo.auth import Identity
+from apiserver.utilities.dicts import nested_set
+
+log = config.logger(__file__)
+queue_bll = QueueBLL()
+
+
+def _get_pipeline_steps_for_controller_task(
+    task: Task, company_id: str, only: Sequence[str] = None
+) -> Sequence[Task]:
+    if not task or task.type != TaskType.controller:
+        return []
+
+    query = Task.objects(company=company_id, parent=task.id)
+    if only:
+        query = query.only(*only)
+
+    return list(query)
+
+
+def archive_task(
+    task: Union[str, Task],
+    company_id: str,
+    identity: Identity,
+    status_message: str,
+    status_reason: str,
+    include_pipeline_steps: bool,
+) -> int:
+    """
+    Deque and archive task
+    Return 1 if successful
+    """
+    user_id = identity.user
+    fields = (
+        "id",
+        "company",
+        "execution",
+        "status",
+        "project",
+        "system_tags",
+        "enqueue_status",
+        "type",
+    )
+    if isinstance(task, str):
+        task = get_task_with_write_access(
+            task,
+            company_id=company_id,
+            identity=identity,
+            only=fields,
+        )
+
+    def archive_task_core(task_: Task) -> int:
+        try:
+            TaskBLL.dequeue_and_change_status(
+                task_,
+                company_id=company_id,
+                user_id=user_id,
+                status_message=status_message,
+                status_reason=status_reason,
+                remove_from_all_queues=True,
+                new_status_for_aborted_task=TaskStatus.stopped,
+            )
+        except APIError:
+            # dequeue may fail if the task was not enqueued
+            pass
+
+        return task_.update(
+            status_message=status_message,
+            status_reason=status_reason,
+            add_to_set__system_tags=EntityVisibility.archived.value,
+            last_change=datetime.utcnow(),
+            last_changed_by=user_id,
+        )
+
+    if include_pipeline_steps and (
+        step_tasks := _get_pipeline_steps_for_controller_task(
+            task, company_id, only=fields
+        )
+    ):
+        for step in step_tasks:
+            archive_task_core(step)
+
+    return archive_task_core(task)
+
+
+def unarchive_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    status_message: str,
+    status_reason: str,
+    include_pipeline_steps: bool,
+) -> int:
+    """
+    Unarchive task. Return 1 if successful
+    """
+    fields = ("id", "type")
+    task = get_task_with_write_access(
+        task_id,
+        company_id=company_id,
+        identity=identity,
+        only=fields,
+    )
+
+    def unarchive_task_core(task_: Task) -> int:
+        return task_.update(
+            status_message=status_message,
+            status_reason=status_reason,
+            pull__system_tags=EntityVisibility.archived.value,
+            last_change=datetime.utcnow(),
+            last_changed_by=identity.user,
+        )
+
+    if include_pipeline_steps and (
+        step_tasks := _get_pipeline_steps_for_controller_task(
+            task, company_id, only=fields
+        )
+    ):
+        for step in step_tasks:
+            unarchive_task_core(step)
+
+    return unarchive_task_core(task)
+
+
+def dequeue_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    status_message: str,
+    status_reason: str,
+    remove_from_all_queues: bool = False,
+    new_status=None,
+) -> Tuple[int, dict]:
+    if new_status and new_status not in get_options(TaskStatus):
+        raise errors.bad_request.ValidationError(f"Invalid task status: {new_status}")
+
+    # get the task without write access to make sure that it actually exists
+    task = Task.get(
+        id=task_id,
+        company=company_id,
+        _only=("id",),
+        include_public=True,
+    )
+    if not task:
+        TaskBLL.remove_task_from_all_queues(company_id, task_id=task_id)
+        return 1, {"updated": 0}
+
+    user_id = identity.user
+    task = get_task_with_write_access(
+        task_id,
+        company_id=company_id,
+        identity=identity,
+        only=(
+            "id",
+            "company",
+            "execution",
+            "status",
+            "project",
+            "enqueue_status",
+        ),
+    )
+
+    res = TaskBLL.dequeue_and_change_status(
+        task,
+        company_id=company_id,
+        user_id=user_id,
+        status_message=status_message,
+        status_reason=status_reason,
+        remove_from_all_queues=remove_from_all_queues,
+        new_status=new_status,
+    )
+    return 1, res
+
+
+def enqueue_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    queue_id: str,
+    status_message: str,
+    status_reason: str,
+    queue_name: str = None,
+    validate: bool = False,
+    force: bool = False,
+    update_execution_queue: bool = True,
+) -> Tuple[int, dict]:
+    if queue_id and queue_name:
+        raise errors.bad_request.ValidationError(
+            "Either queue id or queue name should be provided"
+        )
+
+    task = get_task_with_write_access(
+        task_id=task_id, company_id=company_id, identity=identity
+    )
+    if not update_execution_queue:
+        if not (
+            task.status == TaskStatus.queued and task.execution and task.execution.queue
+        ):
+            raise errors.bad_request.ValidationError(
+                "Cannot skip setting execution queue for a task "
+                "that is not enqueued or does not have execution queue set"
+            )
+
+    if queue_name:
+        queue = queue_bll.get_by_name(
+            company_id=company_id, queue_name=queue_name, only=("id",)
+        )
+        if not queue:
+            queue = queue_bll.create(company_id=company_id, name=queue_name)
+        queue_id = queue.id
+
+    if not queue_id:
+        # try to get default queue
+        queue_id = queue_bll.get_default(company_id).id
+
+    user_id = identity.user
+    if validate:
+        TaskBLL.validate(task)
+
+    before_enqueue_status = task.status
+    if task.status == TaskStatus.queued and task.enqueue_status:
+        before_enqueue_status = task.enqueue_status
+    res = ChangeStatusRequest(
+        task=task,
+        new_status=TaskStatus.queued,
+        status_reason=status_reason,
+        status_message=status_message,
+        force=force,
+        user_id=user_id,
+    ).execute(enqueue_status=before_enqueue_status)
+
+    try:
+        queue_bll.add_task(company_id=company_id, queue_id=queue_id, task_id=task.id)
+    except Exception:
+        # failed enqueueing, revert to previous state
+        ChangeStatusRequest(
+            task=task,
+            current_status_override=TaskStatus.queued,
+            new_status=task.status,
+            force=True,
+            status_reason="failed enqueueing",
+            user_id=user_id,
+        ).execute(enqueue_status=None)
+        raise
+
+    # set the current queue ID in the task
+    if update_execution_queue:
+        if task.execution:
+            Task.objects(id=task_id).update(execution__queue=queue_id, multi=False)
+        else:
+            Task.objects(id=task_id).update(
+                execution=Execution(queue=queue_id), multi=False
+            )
+        nested_set(res, ("fields", "execution.queue"), queue_id)
+
+    # make sure that the task is not queued in any other queue
+    TaskBLL.remove_task_from_all_queues(
+        company_id=company_id, task_id=task_id, exclude=queue_id
+    )
+    return 1, res
+
+
+def move_tasks_to_trash(tasks: Sequence[str]) -> int:
+    try:
+        collection_name = Task._get_collection_name()
+        trash_collection_name = f"{collection_name}__trash"
+        Task.aggregate(
+            [
+                {"$match": {"_id": {"$in": tasks}}},
+                {
+                    "$merge": {
+                        "into": trash_collection_name,
+                        "on": "_id",
+                        "whenMatched": "replace",
+                        "whenNotMatched": "insert",
+                    }
+                },
+            ],
+            allow_disk_use=True,
+        )
+    except Exception as ex:
+        log.error(f"Error copying tasks to trash {str(ex)}")
+
+    return Task.objects(id__in=tasks).delete()
+
+
+def delete_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    move_to_trash: bool,
+    force: bool,
+    delete_output_models: bool,
+    status_message: str,
+    status_reason: str,
+    include_pipeline_steps: bool,
+) -> Tuple[int, Task, CleanupResult]:
+    user_id = identity.user
+    task = get_task_with_write_access(task_id, company_id=company_id, identity=identity)
+
+    if (
+        task.status != TaskStatus.created
+        and EntityVisibility.archived.value not in task.system_tags
+        and not force
+    ):
+        raise errors.bad_request.TaskCannotBeDeleted(
+            "due to status, use force=True",
+            task=task.id,
+            expected=TaskStatus.created,
+            current=task.status,
+        )
+
+    def delete_task_core(task_: Task, force_: bool) -> CleanupResult:
+        try:
+            TaskBLL.dequeue_and_change_status(
+                task_,
+                company_id=company_id,
+                user_id=user_id,
+                status_message=status_message,
+                status_reason=status_reason,
+                remove_from_all_queues=True,
+            )
+        except APIError:
+            # dequeue may fail if the task was not enqueued
+            pass
+
+        res = cleanup_task(
+            company=company_id,
+            user=user_id,
+            task=task_,
+            force=force_,
+            delete_output_models=delete_output_models,
+        )
+
+        if move_to_trash:
+            # make sure that whatever changes were done to the task are saved
+            # the task itself will be deleted later in the move_tasks_to_trash operation
+            task_.last_update = datetime.utcnow()
+            task_.save()
+        else:
+            task_.delete()
+
+        return res
+
+    task_ids = [task.id]
+    cleanup_res = CleanupResult.empty()
+    if include_pipeline_steps and (
+        step_tasks := _get_pipeline_steps_for_controller_task(task, company_id)
+    ):
+        for step in step_tasks:
+            cleanup_res += delete_task_core(step, True)
+            task_ids.append(step.id)
+
+    cleanup_res = delete_task_core(task, force)
+    if move_to_trash:
+        move_tasks_to_trash(task_ids)
+
+    update_project_time(task.project)
+    return 1, task, cleanup_res
+
+
+def reset_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    force: bool,
+    delete_output_models: bool,
+    clear_all: bool,
+) -> Tuple[dict, CleanupResult, dict]:
+    user_id = identity.user
+    task = get_task_with_write_access(task_id, company_id=company_id, identity=identity)
+
+    if not force and task.status == TaskStatus.published:
+        raise errors.bad_request.InvalidTaskStatus(task_id=task.id, status=task.status)
+
+    dequeued = {}
+    updates = {}
+
+    try:
+        dequeued = TaskBLL.dequeue(
+            task, company_id=company_id, user_id=user_id, silent_fail=True
+        )
+    except APIError:
+        # dequeue may fail if the task was not enqueued
+        pass
+
+    TaskBLL.remove_task_from_all_queues(company_id=company_id, task_id=task.id)
+
+    cleaned_up = cleanup_task(
+        company=company_id,
+        user=user_id,
+        task=task,
+        force=force,
+        update_children=False,
+        delete_output_models=delete_output_models,
+    )
+
+    updates.update(
+        set__last_iteration=DEFAULT_LAST_ITERATION,
+        set__last_metrics={},
+        set__unique_metrics=[],
+        set__metric_stats={},
+        set__models__output=[],
+        set__runtime={},
+        unset__output__result=1,
+        unset__output__error=1,
+        unset__last_worker=1,
+        unset__last_worker_report=1,
+        unset__started=1,
+        unset__completed=1,
+        unset__published=1,
+        unset__active_duration=1,
+        unset__enqueue_status=1,
+    )
+
+    if clear_all:
+        updates.update(
+            set__execution=Execution(),
+            unset__script=1,
+        )
+    else:
+        updates.update(unset__execution__queue=1)
+        if task.execution and task.execution.artifacts:
+            updates.update(
+                set__execution__artifacts={
+                    key: artifact
+                    for key, artifact in task.execution.artifacts.items()
+                    if artifact.mode == ArtifactModes.input
+                }
+            )
+
+    res = ChangeStatusRequest(
+        task=task,
+        new_status=TaskStatus.created,
+        force=force,
+        status_reason="reset",
+        status_message="reset",
+        user_id=user_id,
+    ).execute(
+        **updates,
+    )
+
+    return dequeued, cleaned_up, res
+
+
+def publish_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    force: bool,
+    publish_model_func: Callable[[str, str, Identity], Any] = None,
+    status_message: str = "",
+    status_reason: str = "",
+) -> dict:
+    user_id = identity.user
+    task = get_task_with_write_access(task_id, company_id=company_id, identity=identity)
+    if not force:
+        validate_status_change(task.status, TaskStatus.published)
+
+    previous_task_status = task.status
+    output = task.output or Output()
+    publish_failed = False
+
+    try:
+        # set state to publishing
+        task.status = TaskStatus.publishing
+        task.save()
+
+        # publish task models
+        if task.models and task.models.output and publish_model_func:
+            model_id = task.models.output[-1].model
+            model = (
+                Model.objects(id=model_id, company=company_id)
+                .only("id", "ready")
+                .first()
+            )
+            if model and not model.ready:
+                publish_model_func(model.id, company_id, identity)
+
+        # set task status to published, and update (or set) it's new output (view and models)
+        return ChangeStatusRequest(
+            task=task,
+            new_status=TaskStatus.published,
+            force=force,
+            status_reason=status_reason,
+            status_message=status_message,
+            user_id=user_id,
+        ).execute(published=datetime.utcnow(), output=output)
+
+    except Exception as ex:
+        publish_failed = True
+        raise ex
+    finally:
+        if publish_failed:
+            task.status = previous_task_status
+            task.save()
+
+
+def stop_task(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    user_name: str,
+    status_reason: str,
+    force: bool,
+    include_pipeline_steps: bool,
+) -> dict:
+    """
+    Stop a running task. Requires task status 'in_progress' and
+    execution_progress 'running', or force=True. Development task or
+    task that has no associated worker is stopped immediately.
+    For a non-development task with worker only the status message
+    is set to 'stopping' to allow the worker to stop the task and report by itself
+    :return: updated task fields
+    """
+    user_id = identity.user
+    fields = (
+        "status",
+        "project",
+        "tags",
+        "system_tags",
+        "last_worker",
+        "last_update",
+        "execution.queue",
+        "type",
+    )
+    task = get_task_with_write_access(
+        task_id,
+        company_id=company_id,
+        identity=identity,
+        only=fields,
+    )
+
+    def is_run_by_worker(t: Task) -> bool:
+        """Checks if there is an active worker running the task"""
+        update_timeout = config.get("apiserver.workers.task_update_timeout", 600)
+        return (
+            t.last_worker
+            and t.last_update
+            and (datetime.utcnow() - t.last_update).total_seconds() < update_timeout
+        )
+
+    def stop_task_core(task_: Task, force_: bool):
+        is_queued = task_.status == TaskStatus.queued
+        set_stopped = (
+            is_queued
+            or TaskSystemTags.development in task_.system_tags
+            or not is_run_by_worker(task_)
+        )
+
+        if set_stopped:
+            if is_queued:
+                try:
+                    TaskBLL.dequeue(
+                        task_, company_id=company_id, user_id=user_id, silent_fail=True
+                    )
+                except APIError:
+                    # dequeue may fail if the task was not enqueued
+                    pass
+
+            new_status = TaskStatus.stopped
+            status_message = f"Stopped by {user_name}"
+        else:
+            new_status = task_.status
+            status_message = TaskStatusMessage.stopping
+
+        return ChangeStatusRequest(
+            task=task_,
+            new_status=new_status,
+            status_reason=status_reason,
+            status_message=status_message,
+            force=force_,
+            user_id=user_id,
+        ).execute()
+
+    if include_pipeline_steps and (
+        step_tasks := _get_pipeline_steps_for_controller_task(
+            task, company_id, only=fields
+        )
+    ):
+        for step in step_tasks:
+            stop_task_core(step, True)
+
+    return stop_task_core(task, force)

apiserver/bll/task/utils.py

@@ -0,0 +1,411 @@
+from datetime import datetime
+from typing import Sequence
+
+import attr
+import six
+from mongoengine import Q
+from mongoengine.base import UPDATE_OPERATORS
+
+from apiserver.apierrors import errors
+from apiserver.bll.util import update_project_time
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.model import Model
+from apiserver.database.model.task.task import Task, TaskStatus, TaskSystemTags
+from apiserver.database.utils import get_options
+from apiserver.service_repo.auth import Identity
+from apiserver.utilities.attrs import typed_attrs
+
+valid_statuses = get_options(TaskStatus)
+deleted_prefix = "__DELETED__"
+
+
+@typed_attrs
+class ChangeStatusRequest(object):
+    task = attr.ib(type=Task)
+    new_status = attr.ib(
+        type=six.string_types, validator=attr.validators.in_(valid_statuses)
+    )
+    status_reason = attr.ib(type=six.string_types, default="")
+    status_message = attr.ib(type=six.string_types, default="")
+    force = attr.ib(type=bool, default=False)
+    allow_same_state_transition = attr.ib(type=bool, default=True)
+    current_status_override = attr.ib(default=None)
+    user_id = attr.ib(type=str, default=None)
+
+    def execute(self, **kwargs):
+        current_status = self.current_status_override or self.task.status
+        project_id = self.task.project
+
+        # Verify new status is allowed from current status (will throw exception if not valid)
+        self.validate_transition(current_status)
+
+        control = dict(upsert=False, multi=False, write_concern=None, full_result=False)
+
+        now = datetime.utcnow()
+        fields = dict(
+            status=self.new_status,
+            status_reason=self.status_reason,
+            status_message=self.status_message,
+            status_changed=now,
+            last_update=now,
+            last_change=now,
+            last_changed_by=self.user_id,
+        )
+
+        if self.new_status == TaskStatus.queued:
+            fields["pull__system_tags"] = TaskSystemTags.development
+
+        def safe_mongoengine_key(key):
+            return f"__{key}" if key in control else key
+
+        fields.update({safe_mongoengine_key(k): v for k, v in kwargs.items()})
+
+        with translate_errors_context():
+            # atomic change of task status by querying the task with the EXPECTED status before modifying it
+            params = fields.copy()
+            params.update(control)
+            updated = Task.objects(id=self.task.id, status=current_status).update(
+                **params
+            )
+
+        if not updated:
+            # failed to change status (someone else beat us to it?)
+            raise errors.bad_request.FailedChangingTaskStatus(
+                task_id=self.task.id,
+                current_status=current_status,
+                new_status=self.new_status,
+            )
+
+        update_project_time(project_id)
+
+        def is_mongo_operator(field: str) -> bool:
+            head, _, tail = field.partition("__")
+            return tail and (head in UPDATE_OPERATORS)
+
+        # make sure to not return _raw_ queries or any of the update operators
+        fields = {
+            key: value
+            for key, value in fields.items()
+            if not (key == "__raw__" or is_mongo_operator(key))
+        }
+
+        return dict(updated=updated, fields=fields)
+
+    def validate_transition(self, current_status):
+        if self.force:
+            return
+        if self.new_status != current_status:
+            validate_status_change(current_status, self.new_status)
+        elif not self.allow_same_state_transition:
+            raise errors.bad_request.InvalidTaskStatus(
+                "Task already in requested status",
+                current_status=current_status,
+                new_status=self.new_status,
+            )
+
+
+def validate_status_change(current_status, new_status):
+    assert current_status in valid_statuses
+    assert new_status in valid_statuses
+
+    allowed_statuses = get_possible_status_changes(current_status)
+    if new_status not in allowed_statuses:
+        raise errors.bad_request.InvalidTaskStatus(
+            "Invalid status change",
+            current_status=current_status,
+            new_status=new_status,
+        )
+
+
+state_machine = {
+    TaskStatus.created: {TaskStatus.queued, TaskStatus.in_progress},
+    TaskStatus.queued: {TaskStatus.created, TaskStatus.in_progress, TaskStatus.stopped},
+    TaskStatus.in_progress: {
+        TaskStatus.stopped,
+        TaskStatus.failed,
+        TaskStatus.created,
+        TaskStatus.completed,
+    },
+    TaskStatus.stopped: {
+        TaskStatus.closed,
+        TaskStatus.created,
+        TaskStatus.failed,
+        TaskStatus.queued,
+        TaskStatus.in_progress,
+        TaskStatus.published,
+        TaskStatus.publishing,
+        TaskStatus.completed,
+    },
+    TaskStatus.closed: {
+        TaskStatus.created,
+        TaskStatus.failed,
+        TaskStatus.published,
+        TaskStatus.publishing,
+        TaskStatus.stopped,
+    },
+    TaskStatus.failed: {
+        TaskStatus.created,
+        TaskStatus.stopped,
+        TaskStatus.published,
+        TaskStatus.queued,
+    },
+    TaskStatus.publishing: {TaskStatus.published},
+    TaskStatus.published: set(),
+    TaskStatus.completed: {
+        TaskStatus.published,
+        TaskStatus.in_progress,
+        TaskStatus.created,
+    },
+}
+
+
+def get_possible_status_changes(current_status):
+    """
+    :param current_status:
+    :return possible states from current state
+    """
+    possible = state_machine.get(current_status)
+    if possible is None:
+        raise errors.server_error.InternalError(
+            f"Current status {current_status} not supported by state machine"
+        )
+
+    return possible
+
+
+def get_many_tasks_for_writing(
+    company_id: str,
+    identity: Identity,
+    query: Q = None,
+    only: Sequence = None,
+    throw_on_forbidden: bool = True,
+) -> Sequence[Task]:
+    if only:
+        missing = [f for f in ("company",) if f not in only]
+        if missing:
+            only = [*only, *missing]
+
+    result = list(
+        Task.get_many(
+            company=company_id,
+            query=query,
+            override_projection=only,
+            allow_public=True,
+            return_dicts=False,
+        )
+    )
+
+    if not company_id:
+        return result
+
+    forbidden_tasks = {task.id for task in result if not task.company}
+    if forbidden_tasks:
+        if throw_on_forbidden:
+            raise errors.forbidden.NoWritePermission(
+                f"cannot modify public task(s), ids={tuple(forbidden_tasks)}"
+            )
+        result = [task for task in result if task.id not in forbidden_tasks]
+
+    return result
+
+
+def get_task_with_write_access(
+    task_id: str,
+    company_id: str,
+    identity: Identity,
+    only=None,
+) -> Task:
+    """
+    Gets a task that has a required write access
+    :except errors.bad_request.InvalidTaskId: if the task is not found
+    :except errors.forbidden.NoWritePermission: if write_access was required and the task cannot be modified
+    """
+    query = dict(id=task_id, company=company_id)
+
+    task = Task.get_for_writing(_only=only, **query)
+    if not task:
+        raise errors.bad_request.InvalidTaskId(**query)
+
+    return task
+
+
+def get_task_for_update(
+    company_id: str,
+    task_id: str,
+    identity: Identity,
+    allow_all_statuses: bool = False,
+    force: bool = False,
+) -> Task:
+    """
+    Loads only task id and return the task only if it is updatable (status == 'created')
+    """
+    task = get_task_with_write_access(
+        task_id=task_id,
+        company_id=company_id,
+        only=("id", "status"),
+        identity=identity,
+    )
+
+    if allow_all_statuses:
+        return task
+
+    allowed_statuses = (
+        [TaskStatus.created, TaskStatus.in_progress] if force else [TaskStatus.created]
+    )
+    if task.status not in allowed_statuses:
+        raise errors.bad_request.InvalidTaskStatus(
+            expected=TaskStatus.created, status=task.status
+        )
+    return task
+
+
+def update_task(
+    task: Task, user_id: str, update_cmds: dict, set_last_update: bool = True
+):
+    now = datetime.utcnow()
+    last_updates = dict(last_change=now, last_changed_by=user_id)
+    if set_last_update:
+        last_updates.update(last_update=now)
+    return task.update(**update_cmds, **last_updates)
+
+
+def get_last_metric_updates(
+    task_id: str,
+    last_scalar_events: dict,
+    raw_updates: dict,
+    extra_updates: dict,
+    model_events: bool = False,
+):
+    max_values = config.get("services.tasks.max_last_metrics", 2000)
+    total_metrics = set()
+    if max_values:
+        query = dict(id=task_id)
+        to_add = sum(len(v) for m, v in last_scalar_events.items())
+        if to_add <= max_values:
+            query[f"unique_metrics__{max_values - to_add}__exists"] = True
+        db_cls = Model if model_events else Task
+        task = db_cls.objects(**query).only("unique_metrics").first()
+        if task and task.unique_metrics:
+            total_metrics = set(task.unique_metrics)
+
+    new_metrics = []
+
+    def add_last_metric_mean_update(
+        metric_path: str,
+        metric_count: int,
+        metric_total: float,
+    ):
+        """
+        Update new mean field based on the value in db and new data
+        The count field is updated here too and not with inc__ so that
+        it will not get updated in the db earlier than the corresponding mean
+        """
+        metric_path = metric_path.replace("__", ".")
+        mean_value_field = f"{metric_path}.mean_value"
+        count_field = f"{metric_path}.count"
+        raw_updates[mean_value_field] = {
+            "$round": [
+                {
+                    "$divide": [
+                        {
+                            "$add": [
+                                {
+                                    "$multiply": [
+                                        {"$ifNull": [f"${mean_value_field}", 0]},
+                                        {"$ifNull": [f"${count_field}", 0]},
+                                    ]
+                                },
+                                metric_total,
+                            ]
+                        },
+                        {
+                            "$add": [
+                                {"$ifNull": [f"${count_field}", 0]},
+                                metric_count,
+                            ]
+                        },
+                    ]
+                },
+                2,
+            ]
+        }
+        raw_updates[count_field] = {
+            "$add": [
+                {"$ifNull": [f"${count_field}", 0]},
+                metric_count,
+            ]
+        }
+
+    def add_last_metric_conditional_update(
+        metric_path: str, metric_value, iter_value: int, is_min: bool, is_first: bool
+    ):
+        """
+        Build an aggregation for an atomic update of the min or max value and the corresponding iteration
+        """
+        if is_first:
+            field_prefix = "first"
+            op = None
+        elif is_min:
+            field_prefix = "min"
+            op = "$gt"
+        else:
+            field_prefix = "max"
+            op = "$lt"
+
+        value_field = f"{metric_path}__{field_prefix}_value".replace("__", ".")
+        exists = {"$lte": [f"${value_field}", None]}
+        if op:
+            condition = {
+                "$or": [
+                    exists,
+                    {op: [f"${value_field}", metric_value]},
+                ]
+            }
+        else:
+            condition = exists
+
+        raw_updates[value_field] = {
+            "$cond": [condition, metric_value, f"${value_field}"]
+        }
+
+        value_iteration_field = (
+            f"{metric_path}__{field_prefix}_value_iteration".replace("__", ".")
+        )
+        raw_updates[value_iteration_field] = {
+            "$cond": [condition, iter_value, f"${value_iteration_field}"]
+        }
+
+    for metric_key, metric_data in last_scalar_events.items():
+        for variant_key, variant_data in metric_data.items():
+            metric = f"{variant_data.get('metric')}/{variant_data.get('variant')}"
+            if max_values:
+                if len(total_metrics) >= max_values and metric not in total_metrics:
+                    continue
+                total_metrics.add(metric)
+
+            new_metrics.append(metric)
+            path = f"last_metrics__{metric_key}__{variant_key}"
+            for key, value in variant_data.items():
+                if key in ("min_value", "max_value", "first_value"):
+                    add_last_metric_conditional_update(
+                        metric_path=path,
+                        metric_value=value,
+                        iter_value=variant_data.get(f"{key}_iter", 0),
+                        is_min=(key == "min_value"),
+                        is_first=(key == "first_value"),
+                    )
+                elif key in ("metric", "variant", "value", "x_axis_label"):
+                    extra_updates[f"set__{path}__{key}"] = value
+
+            count = variant_data.get("count")
+            total = variant_data.get("total")
+            if count is not None and total is not None:
+                add_last_metric_mean_update(
+                    metric_path=path,
+                    metric_count=count,
+                    metric_total=total,
+                )
+
+    if new_metrics:
+        extra_updates["add_to_set__unique_metrics"] = new_metrics

apiserver/bll/user/__init__.py

@@ -1,7 +1,10 @@
-from apierrors import errors
-from apimodels.users import CreateRequest
-from database.errors import translate_errors_context
-from database.model.user import User
+from datetime import datetime
+
+from apiserver.apierrors import errors
+from apiserver.apimodels.users import CreateRequest
+from apiserver.config.info import get_version
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.user import User
 
 
 class UserBLL:
@@ -12,7 +15,11 @@ class UserBLL:
             if user_id and User.objects(id=user_id).only("id"):
                 raise errors.bad_request.UserIdExists(id=user_id)
 
-            user = User(**request.to_struct())
+            user = User(
+                **request.to_struct(),
+                created=datetime.utcnow(),
+                created_in_version=get_version(),
+            )
             user.save(force_insert=True)
 
     @staticmethod

apiserver/bll/util.py

@@ -0,0 +1,92 @@
+import functools
+import itertools
+from concurrent.futures.thread import ThreadPoolExecutor
+from datetime import datetime
+from typing import (
+    Optional,
+    Callable,
+    Iterable,
+    Tuple,
+    Sequence,
+    TypeVar,
+    Union,
+)
+
+from boltons import iterutils
+
+from apiserver.apierrors import APIError
+from apiserver.database.model.project import Project
+from apiserver.database.model.settings import Settings
+
+
+@functools.lru_cache()
+def get_server_uuid() -> Optional[str]:
+    return Settings.get_by_key("server.uuid")
+
+
+def parallel_chunked_decorator(func: Callable = None, chunk_size: int = 100):
+    """
+    Decorates a method for parallel chunked execution. The method should have
+    one positional parameter (that is used for breaking into chunks)
+    and arbitrary number of keyword params. The return value should be iterable
+    The results are concatenated in the same order as the passed params
+    """
+    if func is None:
+        return functools.partial(parallel_chunked_decorator, chunk_size=chunk_size)
+
+    @functools.wraps(func)
+    def wrapper(self, iterable: Iterable, **kwargs):
+        assert iterutils.is_collection(
+            iterable
+        ), "The positional parameter should be an iterable for breaking into chunks"
+
+        func_with_params = functools.partial(func, self, **kwargs)
+        with ThreadPoolExecutor() as pool:
+            return list(
+                itertools.chain.from_iterable(
+                    filter(
+                        None,
+                        pool.map(
+                            func_with_params,
+                            iterutils.chunked_iter(iterable, chunk_size),
+                        ),
+                    )
+                ),
+            )
+
+    return wrapper
+
+
+T = TypeVar("T")
+
+
+def run_batch_operation(
+    func: Callable[[str], T], ids: Sequence[str]
+) -> Tuple[Sequence[Tuple[str, T]], Sequence[dict]]:
+    results = list()
+    failures = list()
+    for _id in ids:
+        try:
+            results.append((_id, func(_id)))
+        except APIError as err:
+            failures.append(
+                {
+                    "id": _id,
+                    "error": {
+                        "codes": [err.code, err.subcode],
+                        "msg": err.msg,
+                        "data": err.error_data,
+                    },
+                }
+            )
+    return results, failures
+
+
+def update_project_time(project_ids: Union[str, Sequence[str]]):
+    if not project_ids:
+        return
+
+    if isinstance(project_ids, str):
+        project_ids = [project_ids]
+
+    return Project.objects(id__in=project_ids).update(last_update=datetime.utcnow())

apiserver/bll/workers/__init__.py

@@ -1,15 +1,18 @@
 import itertools
+import re
 from datetime import datetime, timedelta
+from time import time
 from typing import Sequence, Set, Optional
 
 import attr
 import elasticsearch.helpers
+from boltons.iterutils import partition, chunked_iter
+from pyhocon import ConfigTree
 
-import es_factory
-from apierrors import APIError
-from apierrors.errors import bad_request, server_error
-from apimodels.workers import (
-    DEFAULT_TIMEOUT,
+from apiserver.es_factory import es_factory
+from apiserver.apierrors import APIError
+from apiserver.apierrors.errors import bad_request, server_error
+from apiserver.apimodels.workers import (
     IdNameEntry,
     WorkerEntry,
     StatusReportRequest,
@@ -17,24 +20,26 @@ from apimodels.workers import (
     QueueEntry,
     MachineStats,
 )
-from config import config
-from database.errors import translate_errors_context
-from database.model.auth import User
-from database.model.company import Company
-from database.model.project import Project
-from database.model.queue import Queue
-from database.model.task.task import Task
-from redis_manager import redman
-from timing_context import TimingContext
-from tools import safe_get
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+from apiserver.database.model.auth import User
+from apiserver.database.model.company import Company
+from apiserver.database.model.project import Project
+from apiserver.database.model.queue import Queue
+from apiserver.database.model.task.task import Task
+from apiserver.redis_manager import redman
+from apiserver.utilities.dicts import nested_get
 from .stats import WorkerStats
 
 log = config.logger(__file__)
 
 
 class WorkerBLL:
+    _key_regex_trans = str.maketrans({"*": ".*", "?": ".?"})
+
     def __init__(self, es=None, redis=None):
         self.es_client = es or es_factory.connect("workers")
+        self.config = config.get("services.workers", ConfigTree())
         self.redis = redis or redman.connection("workers")
         self._stats = WorkerStats(self.es_client)
 
@@ -51,6 +56,7 @@ class WorkerBLL:
         queues: Sequence[str] = None,
         timeout: int = 0,
         tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
     ) -> WorkerEntry:
         """
         Register a worker
@@ -66,7 +72,7 @@ class WorkerBLL:
         """
         key = WorkerBLL._get_worker_key(company_id, user_id, worker)
 
-        timeout = timeout or DEFAULT_TIMEOUT
+        timeout = timeout or int(self.config.get("default_worker_timeout_sec", 10 * 60))
         queues = queues or []
 
         with translate_errors_context():
@@ -76,7 +82,7 @@ class WorkerBLL:
                 raise bad_request.InvalidUserId(**query)
             company = Company.objects(id=company_id).only("id", "name").first()
             if not company:
-                raise server_error.InternalError("invalid company", company=company_id)
+                raise bad_request.InvalidId("invalid company", company=company_id)
 
             queue_objs = Queue.objects(company=company_id, id__in=queues).only("id")
             if len(queue_objs) < len(queues):
@@ -95,9 +101,10 @@ class WorkerBLL:
                 register_timeout=timeout,
                 last_activity_time=now,
                 tags=tags,
+                system_tags=system_tags,
             )
 
-            self.redis.setex(key, timedelta(seconds=timeout), entry.to_json())
+            self._save_worker_data(entry)
 
         return entry
 
@@ -109,15 +116,20 @@ class WorkerBLL:
         :param worker: worker ID
         :raise bad_request.WorkerNotRegistered: the worker was not previously registered
         """
-        with TimingContext("redis", "workers_unregister"):
-            res = self.redis.delete(
-                company_id, self._get_worker_key(company_id, user_id, worker)
-            )
-        if not res:
+        res = self.redis.delete(
+            company_id, self._get_worker_key(company_id, user_id, worker)
+        )
+        if not res and not config.get("apiserver.workers.auto_unregister", False):
             raise bad_request.WorkerNotRegistered(worker=worker)
 
     def status_report(
-        self, company_id: str, user_id: str, ip: str, report: StatusReportRequest, tags: Sequence[str] = None,
+        self,
+        company_id: str,
+        user_id: str,
+        ip: str,
+        report: StatusReportRequest,
+        tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
     ) -> None:
         """
         Write worker status report
@@ -133,22 +145,23 @@ class WorkerBLL:
 
         try:
             entry.ip = ip
-            now = datetime.utcnow()
-            entry.last_activity_time = now
 
             if tags is not None:
                 entry.tags = tags
+            if system_tags is not None:
+                entry.system_tags = system_tags
 
             if report.machine_stats:
-                self._log_stats_to_es(
+                self.log_stats_to_es(
                     company_id=company_id,
-                    company_name=entry.company.name,
-                    worker=report.worker,
+                    worker_id=report.worker,
                     timestamp=report.timestamp,
                     task=report.task,
                     machine_stats=report.machine_stats,
                 )
 
+            now = datetime.utcnow()
+            entry.last_activity_time = now
             entry.queue = report.queue
 
             if report.queues:
@@ -164,6 +177,8 @@ class WorkerBLL:
                         last_worker=report.worker,
                         last_worker_report=now,
                         last_update=now,
+                        last_change=now,
+                        last_changed_by=user_id,
                     )
                     # modify(new=True, ...) returns the modified object
                     task = Task.objects(**query).modify(new=True, **update)
@@ -175,7 +190,9 @@ class WorkerBLL:
                     if task.project:
                         project = Project.objects(id=task.project).only("name").first()
                         if project:
-                            entry.project = IdNameEntry(id=project.id, name=project.name)
+                            entry.project = IdNameEntry(
+                                id=project.id, name=project.name
+                            )
 
             entry.last_report_time = now
         except APIError:
@@ -187,8 +204,41 @@ class WorkerBLL:
         finally:
             self._save_worker(entry)
 
+    def get_count(
+        self,
+        company_id: str,
+        last_seen: Optional[int] = None,
+        tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
+        worker_pattern: str = None,
+    ):
+        if not last_seen:
+            return len(
+                self._get_keys(
+                    company_id,
+                    user_tags=tags,
+                    system_tags=system_tags,
+                    worker_pattern=worker_pattern,
+                )
+            )
+
+        return len(
+            self.get_all(
+                company_id,
+                last_seen=last_seen,
+                tags=tags,
+                system_tags=system_tags,
+                worker_pattern=worker_pattern,
+            )
+        )
+
     def get_all(
-        self, company_id: str, last_seen: Optional[int] = None
+        self,
+        company_id: str,
+        last_seen: Optional[int] = None,
+        tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
+        worker_pattern: str = None,
     ) -> Sequence[WorkerEntry]:
         """
         Get all the company workers that were active during the last_seen period
@@ -197,7 +247,12 @@ class WorkerBLL:
         :return:
         """
         try:
-            workers = self._get(company_id)
+            workers = self._get(
+                company_id,
+                user_tags=tags,
+                system_tags=system_tags,
+                worker_pattern=worker_pattern,
+            )
         except Exception as e:
             raise server_error.DataError("failed loading worker entries", err=e.args[0])
 
@@ -212,15 +267,23 @@ class WorkerBLL:
         return workers
 
     def get_all_with_projection(
-        self, company_id: str, last_seen: int
+        self,
+        company_id: str,
+        last_seen: int,
+        tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
+        worker_pattern: str = None,
     ) -> Sequence[WorkerResponseEntry]:
-
-        helpers = list(
-            map(
-                WorkerConversionHelper.from_worker_entry,
-                self.get_all(company_id=company_id, last_seen=last_seen),
+        helpers = [
+            WorkerConversionHelper.from_worker_entry(entry)
+            for entry in self.get_all(
+                company_id=company_id,
+                last_seen=last_seen,
+                tags=tags,
+                system_tags=system_tags,
+                worker_pattern=worker_pattern,
             )
-        )
+        ]
 
         task_ids = set(filter(None, (helper.task_id for helper in helpers)))
         all_queues = set(
@@ -234,19 +297,18 @@ class WorkerBLL:
                 {
                     "$project": {
                         "name": 1,
+                        "display_name": 1,
                         "next_entry": {"$arrayElemAt": ["$entries", 0]},
                         "num_entries": {"$size": "$entries"},
                     }
                 },
             ]
-            queues_info = {
-                res["_id"]: res for res in Queue.objects.aggregate(projection)
-            }
+            queues_info = {res["_id"]: res for res in Queue.aggregate(projection)}
             task_ids = task_ids.union(
                 filter(
                     None,
                     (
-                        safe_get(info, "next_entry/task")
+                        nested_get(info, ("next_entry", "task"))
                         for info in queues_info.values()
                     ),
                 )
@@ -257,7 +319,7 @@ class WorkerBLL:
             tasks_info = {
                 task.id: task
                 for task in Task.objects(id__in=task_ids).only(
-                    "name", "started", "last_iteration"
+                    "name", "started", "last_iteration", "active_duration"
                 )
             }
 
@@ -269,8 +331,9 @@ class WorkerBLL:
                 if not info:
                     continue
                 entry.name = info.get("name", None)
+                entry.display_name = info.get("display_name", None)
                 entry.num_tasks = info.get("num_entries", 0)
-                task_id = safe_get(info, "next_entry/task")
+                task_id = nested_get(info, ("next_entry", "task"))
                 if task_id:
                     task = tasks_info.get(task_id, None)
                     entry.next_task = IdNameEntry(
@@ -280,13 +343,9 @@ class WorkerBLL:
         for helper in helpers:
             worker = helper.worker
             if helper.task_id:
-                task = tasks_info.get(helper.task_id, None)
+                task: Task = tasks_info.get(helper.task_id, None)
                 if task:
-                    worker.task.running_time = (
-                        int((datetime.utcnow() - task.started).total_seconds() * 1000)
-                        if task.started
-                        else 0
-                    )
+                    worker.task.running_time = (task.active_duration or 0) * 1000
                     worker.task.last_iteration = task.last_iteration
 
             update_queue_entries(worker.queue)
@@ -313,8 +372,7 @@ class WorkerBLL:
         """
         key = self._get_worker_key(company_id, user_id, worker)
 
-        with TimingContext("redis", "get_worker"):
-            data = self.redis.get(key)
+        data = self.redis.get(key)
 
         if data:
             try:
@@ -341,42 +399,163 @@ class WorkerBLL:
 
         raise bad_request.InvalidWorkerId(worker=worker)
 
+    @staticmethod
+    def _get_tagged_workers_key(company: str, tags_field: str, tag: str) -> str:
+        """Build redis key from company, user and worker_id"""
+        return f"workers.{tags_field}_{company}_{tag}"
+
+    @staticmethod
+    def _get_all_workers_key(company: str) -> str:
+        """Build redis key from company, user and worker_id"""
+        return f"workers_{company}"
+
+    def _save_worker_data(self, entry: WorkerEntry):
+        self.redis.setex(
+            entry.key, timedelta(seconds=entry.register_timeout), entry.to_json()
+        )
+        company_id = entry.company.id
+        expiration = int(time()) + entry.register_timeout
+        worker_item = {entry.key: expiration}
+        self.redis.zadd(self._get_all_workers_key(company_id), worker_item)
+        for tags, tags_field in (
+            (entry.tags, "tags"),
+            (entry.system_tags, "systemtags"),
+        ):
+            for tag in tags:
+                name = self._get_tagged_workers_key(company_id, tags_field, tag)
+                self.redis.zadd(name, worker_item)
+
     def _save_worker(self, entry: WorkerEntry) -> None:
         """Save worker entry in Redis"""
         try:
-            self.redis.setex(
-                entry.key, timedelta(seconds=entry.register_timeout), entry.to_json()
-            )
+            self._save_worker_data(entry)
         except Exception:
             msg = "Failed saving worker entry"
             log.exception(msg)
 
+    def _get_keys(
+        self,
+        company: str,
+        user: str = "*",
+        user_tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
+        worker_pattern: str = None,
+    ) -> Sequence[bytes]:
+        if not (user_tags or system_tags):
+            match = self._get_worker_key(company, user, worker_pattern or "*")
+            return list(self.redis.scan_iter(match))
+
+        def filter_by_user_and_pattern(in_keys: Set[bytes]) -> Set[bytes]:
+            if user != "*":
+                user_bytes = user.encode()
+                in_keys = {k for k in in_keys if user_bytes in k}
+
+            if worker_pattern:
+                worker_pattern_bytes = (
+                    f"{worker_pattern.translate(self._key_regex_trans)}$".encode()
+                )
+                regex = re.compile(worker_pattern_bytes)
+                in_keys = {k for k in in_keys if regex.search(k)}
+
+            return in_keys
+
+        worker_keys = set()
+        for tags, tags_field in (
+            (user_tags, "tags"),
+            (system_tags, "systemtags"),
+        ):
+            if not tags:
+                continue
+
+            timestamp = int(time())
+            include, exclude = partition(tags, key=lambda x: x[0] != "-")
+            if include:
+                tagged_workers = set()
+                for tag in include:
+                    tagged_workers_key = self._get_tagged_workers_key(
+                        company, tags_field, tag
+                    )
+                    self.redis.zremrangebyscore(
+                        tagged_workers_key, min=0, max=timestamp
+                    )
+                    tagged_workers.update(self.redis.zrange(tagged_workers_key, 0, -1))
+
+                tagged_workers = filter_by_user_and_pattern(tagged_workers)
+                worker_keys = (
+                    worker_keys.intersection(tagged_workers)
+                    if worker_keys
+                    else tagged_workers
+                )
+                if not worker_keys:
+                    return []
+
+            if exclude:
+                if not worker_keys:
+                    all_workers_key = self._get_all_workers_key(company)
+                    self.redis.zremrangebyscore(all_workers_key, min=0, max=timestamp)
+                    worker_keys.update(self.redis.zrange(all_workers_key, 0, -1))
+                    worker_keys = filter_by_user_and_pattern(worker_keys)
+                    if not worker_keys:
+                        return []
+
+                for tag in exclude:
+                    tagged_workers_key = self._get_tagged_workers_key(
+                        company, tags_field, tag[1:]
+                    )
+                    self.redis.zremrangebyscore(
+                        tagged_workers_key, min=0, max=timestamp
+                    )
+                    worker_keys.difference_update(
+                        self.redis.zrange(tagged_workers_key, 0, -1)
+                    )
+                    if not worker_keys:
+                        return []
+
+        return list(worker_keys)
+
     def _get(
-        self, company: str, user: str = "*", worker_id: str = "*"
+        self,
+        company: str,
+        user: str = "*",
+        user_tags: Sequence[str] = None,
+        system_tags: Sequence[str] = None,
+        worker_pattern: str = None,
     ) -> Sequence[WorkerEntry]:
         """Get worker entries matching the company and user, worker patterns"""
-        match = self._get_worker_key(company, user, worker_id)
-        with TimingContext("redis", "workers_get_all"):
-            res = self.redis.scan_iter(match)
-        return [WorkerEntry.from_json(self.redis.get(r)) for r in res]
+
+        entries = []
+        for keys in chunked_iter(
+            self._get_keys(
+                company,
+                user=user,
+                user_tags=user_tags,
+                system_tags=system_tags,
+                worker_pattern=worker_pattern,
+            ),
+            1000,
+        ):
+            data = self.redis.mget(keys)
+            if data:
+                entries.extend(WorkerEntry.from_json(d) for d in data if d)
+
+        return entries
 
     @staticmethod
     def _get_es_index_suffix():
         """Get the index name suffix for storing current month data"""
         return datetime.utcnow().strftime("%Y-%m")
 
-    def _log_stats_to_es(
+    def log_stats_to_es(
         self,
         company_id: str,
-        company_name: str,
-        worker: str,
+        worker_id: str,
         timestamp: int,
         task: str,
         machine_stats: MachineStats,
-    ) -> bool:
+    ) -> int:
         """
         Actually writing the worker statistics to Elastic
-        :return: True if successful, False otherwise
+        :return: The amount of logged documents
         """
         es_index = (
             f"{self._stats.worker_stats_prefix_for_company(company_id)}"
@@ -388,8 +567,7 @@ class WorkerBLL:
                 _index=es_index,
                 _source=dict(
                     timestamp=timestamp,
-                    worker=worker,
-                    company=company_name,
+                    worker=worker_id,
                     task=task,
                     category=category,
                     metric=metric,
@@ -414,7 +592,7 @@ class WorkerBLL:
 
         es_res = elasticsearch.helpers.bulk(self.es_client, actions)
         added, errors = es_res[:2]
-        return (added == len(actions)) and not errors
+        return added
 
 
 @attr.s(auto_attribs=True)

apiserver/bll/workers/stats.py

@@ -0,0 +1,292 @@
+from collections import defaultdict
+from concurrent.futures.thread import ThreadPoolExecutor
+from functools import partial
+from typing import Optional, Sequence
+
+from apiserver.apierrors import errors
+from apiserver.apierrors.errors import bad_request
+from apiserver.apimodels.workers import AggregationType, GetStatsRequest, StatItem
+from apiserver.bll.query import Builder as QueryBuilder
+from apiserver.config_repo import config
+from apiserver.database.errors import translate_errors_context
+
+log = config.logger(__file__)
+
+
+class WorkerStats:
+    min_chart_interval = config.get("services.workers.min_chart_interval_sec", 40)
+    _max_metrics_concurrency = config.get("services.events.events_retrieval.max_metrics_concurrency", 4)
+
+    def __init__(self, es):
+        self.es = es
+
+    @staticmethod
+    def worker_stats_prefix_for_company(company_id: str) -> str:
+        """Returns the es index prefix for the company"""
+        return f"worker_stats_{company_id.lower()}_"
+
+    def search_company_stats(self, company_id: str, es_req: dict) -> dict:
+        return self.es.search(
+            index=f"{self.worker_stats_prefix_for_company(company_id)}*",
+            body=es_req,
+        )
+
+    def get_worker_stats_keys(
+        self, company_id: str, worker_ids: Optional[Sequence[str]]
+    ) -> dict:
+        """
+        Get dictionary of metric types grouped by categories
+        :param company_id: company id
+        :param worker_ids: optional list of workers to get metric types from.
+        If not specified them metrics for all the company workers returned
+        :return:
+        """
+        es_req = {
+            "size": 0,
+            "aggs": {
+                "categories": {
+                    "terms": {"field": "category"},
+                    "aggs": {"metrics": {"terms": {"field": "metric"}}},
+                }
+            },
+        }
+        if worker_ids:
+            es_req["query"] = QueryBuilder.terms("worker", worker_ids)
+
+        res = self.search_company_stats(company_id, es_req)
+
+        if not res["hits"]["total"]["value"]:
+            raise bad_request.WorkerStatsNotFound(
+                f"No statistic metrics found for the company {company_id} and workers {worker_ids}"
+            )
+
+        return {
+            category["key"]: [
+                metric["key"] for metric in category["metrics"]["buckets"]
+            ]
+            for category in res["aggregations"]["categories"]["buckets"]
+        }
+
+    def _get_worker_stats_per_metric(
+        self,
+        metric_item: StatItem,
+        company_id: str,
+        from_date: float,
+        to_date: float,
+        interval: int,
+        split_by_resource: bool,
+        worker_ids: Sequence[str],
+    ):
+        agg_types_to_es = {
+            AggregationType.avg: "avg",
+            AggregationType.min: "min",
+            AggregationType.max: "max",
+        }
+        agg = {
+            metric_item.aggregation.value: {
+                agg_types_to_es[metric_item.aggregation]: {"field": "value", "missing": 0.0 }
+            }
+        }
+        split_by_resource = split_by_resource and metric_item.key.startswith("gpu_")
+        if split_by_resource:
+            split_aggs = {"split": {"terms": {"field": "variant"}, "aggs": agg}}
+        else:
+            split_aggs = {}
+
+        es_req = {
+            "size": 0,
+            "aggs": {
+                "workers": {
+                    "terms": {"field": "worker"},
+                    "aggs": {
+                        "dates": {
+                            "date_histogram": {
+                                "field": "timestamp",
+                                "fixed_interval": f"{interval}s",
+                                "extended_bounds": {
+                                    "min": int(from_date) * 1000,
+                                    "max": int(to_date) * 1000,
+                                },
+                            },
+                            "aggs": {
+                                **agg,
+                                **split_aggs,
+                            },
+                        }
+                    },
+                }
+            },
+        }
+
+        query_terms = [
+            QueryBuilder.dates_range(from_date, to_date),
+            QueryBuilder.term("metric", metric_item.key),
+        ]
+        if worker_ids:
+            query_terms.append(QueryBuilder.terms("worker", worker_ids))
+        es_req["query"] = {"bool": {"must": query_terms}}
+
+        with translate_errors_context():
+            data = self.search_company_stats(company_id, es_req)
+
+        cutoff_date = (
+            to_date - 0.9 * interval
+        ) * 1000  # do not return the point for the incomplete last interval
+        return self._extract_results(
+            data, metric_item, split_by_resource, cutoff_date
+        )
+
+    def get_worker_stats(self, company_id: str, request: GetStatsRequest) -> dict:
+        """
+        Get statistics for company workers metrics in the specified time range
+        Returned as date histograms for different aggregation types
+        grouped by worker, metric type (and optionally metric variant)
+        Buckets with no metrics are not returned
+        Note: all the statistics are retrieved as one ES query
+        """
+        from_date = request.from_date
+        to_date = request.to_date
+        if from_date >= to_date:
+            raise errors.bad_request.FieldsValueError(
+                "from_date must be less than to_date"
+            )
+
+        interval = max(request.interval, self.min_chart_interval)
+        with ThreadPoolExecutor(self._max_metrics_concurrency) as pool:
+            res = list(
+                pool.map(
+                    partial(
+                        self._get_worker_stats_per_metric,
+                        company_id=company_id,
+                        from_date=from_date,
+                        to_date=to_date,
+                        interval=interval,
+                        split_by_resource=request.split_by_resource,
+                        worker_ids=request.worker_ids,
+                    ),
+                    request.items,
+                )
+            )
+
+        ret = defaultdict(lambda: defaultdict(dict))
+        for workers in res:
+            for worker, metrics in workers.items():
+                for metric, stats in metrics.items():
+                    ret[worker][metric].update(stats)
+
+        return ret
+
+    @staticmethod
+    def _extract_results(
+        data: dict,
+        metric_item: StatItem,
+        split_by_resource: bool,
+        cutoff_date,
+    ) -> dict:
+        """
+        Clean results returned from elastic search (remove "aggregations", "buckets" etc.),
+        leave only aggregation types requested by the user and return a clean dictionary
+        :param data: aggregation data retrieved from ES
+        """
+        if "aggregations" not in data:
+            return {}
+
+        def extract_metric_results(metric: dict) -> dict:
+            aggregation = metric_item.aggregation.value
+            date_buckets = metric["dates"]["buckets"]
+            length = len(date_buckets)
+            while length > 0 and date_buckets[length - 1]["key"] >= cutoff_date:
+                length -= 1
+
+            dates = [None] * length
+            agg_values = [0.0] * length
+            resource_series = defaultdict(lambda: [0.0] * length)
+
+            for idx in range(0, length):
+                date = date_buckets[idx]
+                dates[idx] = date["key"]
+                if aggregation in date:
+                    agg_values[idx] = date[aggregation]["value"] or 0.0
+
+                if split_by_resource and "split" in date:
+                    for resource in date["split"]["buckets"]:
+                        series = resource_series[resource["key"]]
+                        if aggregation in resource:
+                            series[idx] = resource[aggregation]["value"] or 0.0
+
+            if len(resource_series) == 1:
+                resource_series = {}
+
+            return {
+                "dates": dates,
+                "values": agg_values,
+                **(
+                    {"resource_series": resource_series} if resource_series else {}
+                ),
+            }
+
+        return {
+            worker["key"]: {
+                metric_item.key: {
+                    metric_item.aggregation.value: extract_metric_results(worker)
+                }
+            }
+            for worker in data["aggregations"]["workers"]["buckets"]
+        }
+
+    def get_activity_report(
+        self,
+        company_id: str,
+        from_date: float,
+        to_date: float,
+        interval: int,
+        active_only: bool,
+    ) -> Sequence[dict]:
+        """
+        Get statistics for company workers metrics in the specified time range
+        Returned as date histograms for different aggregation types
+        grouped by worker, metric type (and optionally metric variant)
+        Note: all the statistics are retrieved using one ES query
+        """
+        if from_date >= to_date:
+            raise bad_request.FieldsValueError("from_date must be less than to_date")
+        interval = max(interval, self.min_chart_interval)
+
+        must = [QueryBuilder.dates_range(from_date, to_date)]
+        if active_only:
+            must.append({"exists": {"field": "task"}})
+
+        es_req = {
+            "size": 0,
+            "aggs": {
+                "dates": {
+                    "date_histogram": {
+                        "field": "timestamp",
+                        "fixed_interval": f"{interval}s",
+                        "extended_bounds": {
+                          "min": int(from_date) * 1000,
+                          "max": int(to_date) * 1000,
+                        }
+                    },
+                    "aggs": {"workers_count": {"cardinality": {"field": "worker"}}},
+                }
+            },
+            "query": {"bool": {"must": must}},
+        }
+
+        with translate_errors_context():
+            data = self.search_company_stats(company_id, es_req)
+
+        if "aggregations" not in data:
+            return {}
+
+        ret = [
+            dict(date=date["key"], count=date["workers_count"]["value"])
+            for date in data["aggregations"]["dates"]["buckets"]
+        ]
+
+        if ret and ret[-1]["date"] > (to_date - 0.9 * interval):
+            # remove last interval if it's incomplete. Allow 10% tolerance
+            ret.pop()
+
+        return ret

apiserver/config/__init__.py

@@ -0,0 +1 @@
+from .basic import BasicConfig, ConfigurationError

apiserver/config/basic.py

@@ -0,0 +1,238 @@
+import logging
+import logging.config
+import os
+import platform
+from functools import reduce
+from os import getenv
+from os.path import expandvars
+from pathlib import Path
+from typing import List, Any, TypeVar, Sequence, Set
+
+from boltons.iterutils import first
+from pyhocon import ConfigTree, ConfigFactory, ConfigValues
+from pyparsing import (
+    ParseFatalException,
+    ParseException,
+    RecursiveGrammarException,
+    ParseSyntaxException,
+)
+
+from apiserver.utilities import json
+
+EXTRA_CONFIG_PATHS = ("/opt/trains/config", "/opt/clearml/config")
+DEFAULT_PREFIXES = ("clearml", "trains")
+EXTRA_CONFIG_PATH_SEP = ":" if platform.system() != "Windows" else ";"
+
+
+class BasicConfig:
+    NotSet = object()
+
+    extra_config_values_env_key_sep = "__"
+    default_config_dir = "default"
+
+    def __init__(
+        self,
+        folder: str = None,
+        verbose: bool = True,
+        prefix: Sequence[str] = DEFAULT_PREFIXES,
+        exclude_files_from_base_folder: Sequence[str] = None,
+    ):
+        folder = (
+            Path(folder)
+            if folder
+            else Path(__file__).with_name(self.default_config_dir)
+        )
+        if not folder.is_dir():
+            raise ValueError("Invalid configuration folder")
+
+        self.exclude_files_from_base_folder = (
+            set(exclude_files_from_base_folder)
+            if exclude_files_from_base_folder
+            else set()
+        )
+        self.verbose = verbose
+
+        self.extra_config_path_override_var = [
+            f"{p.upper()}_CONFIG_DIR" for p in prefix
+        ]
+
+        self.prefix = prefix[0]
+        self.extra_config_values_env_key_prefix = [
+            f"{p.upper()}{self.extra_config_values_env_key_sep}"
+            for p in reversed(prefix)
+        ]
+
+        self._paths = [folder, *self._get_paths()]
+        self._config = self._reload()
+
+    def __getitem__(self, key):
+        return self._config[key]
+
+    def get(self, key: str, default: Any = NotSet) -> Any:
+        value = self._config.get(key, default)
+        if value is self.NotSet:
+            raise KeyError(
+                f"Unable to find value for key '{key}' and default value was not provided."
+            )
+        return value
+
+    def to_dict(self) -> dict:
+        return self._config.as_plain_ordered_dict()
+
+    def as_json(self) -> str:
+        return json.dumps(self.to_dict(), indent=2)
+
+    def logger(self, name: str) -> logging.Logger:
+        if Path(name).is_file():
+            name = Path(name).stem
+            if name == "__init__" and Path(name).parent.stem:
+                name = Path(name).parent.stem
+        path = ".".join((self.prefix, name))
+        return logging.getLogger(path)
+
+    def _read_extra_env_config_values(self) -> ConfigTree:
+        """Loads extra configuration from environment-injected values"""
+        result = ConfigTree()
+
+        for prefix in self.extra_config_values_env_key_prefix:
+            keys = sorted(k for k in os.environ if k.startswith(prefix))
+            for key in keys:
+                path = (
+                    key[len(prefix) :]
+                    .replace(self.extra_config_values_env_key_sep, ".")
+                    .lower()
+                )
+                result = self._merge_configs(
+                    result, ConfigFactory.parse_string(f"{path}: {os.environ[key]}")
+                )
+
+        return result
+
+    def _get_paths(self) -> List[Path]:
+        default_paths = EXTRA_CONFIG_PATH_SEP.join(EXTRA_CONFIG_PATHS)
+        value = first(map(getenv, self.extra_config_path_override_var), default_paths)
+
+        paths = [
+            Path(expandvars(v)).expanduser() for v in value.split(EXTRA_CONFIG_PATH_SEP)
+        ]
+
+        if value is not default_paths:
+            invalid = [path for path in paths if not path.is_dir()]
+            if invalid:
+                print(
+                    f"WARNING: Invalid paths in {self.extra_config_path_override_var} env var: {' '.join(map(str, invalid))}"
+                )
+
+        return [path for path in paths if path.is_dir()]
+
+    def reload(self):
+        self._config = self._reload()
+
+    def _reload(self) -> ConfigTree:
+        extra_config_values = self._read_extra_env_config_values()
+
+        configs = [
+            self._read_recursive(
+                path,
+                exclude_files=(
+                    self.exclude_files_from_base_folder if idx == 0 else None
+                ),
+            )
+            for idx, path in enumerate(self._paths)
+        ]
+
+        return reduce(
+            lambda last, config: self._merge_configs(last, config, copy_trees=True),
+            configs + [extra_config_values],
+            ConfigTree(),
+        )
+
+    @classmethod
+    def _merge_configs(cls, a, b, copy_trees=False, override_prefix="-"):
+        """Based on pyhocon.ConfigTree.merge_configs, with dict override support using a `-` key prefix"""
+        for key, value in b.items():
+            override = key.startswith(override_prefix)
+            if override:
+                key = key[len(override_prefix) :]
+            # if key is in both a and b and both values are dictionary then merge it otherwise override it
+            if (
+                not override
+                and key in a
+                and isinstance(a[key], ConfigTree)
+                and isinstance(b[key], ConfigTree)
+            ):
+                if copy_trees:
+                    a[key] = a[key].copy()
+                cls._merge_configs(a[key], b[key], copy_trees=copy_trees)
+            else:
+                if isinstance(value, ConfigValues):
+                    value.parent = a
+                    value.key = key
+                    if key in a:
+                        value.overriden_value = a[key]
+                a[key] = value
+                if a.root:
+                    if b.root:
+                        a.history[key] = a.history.get(key, []) + b.history.get(
+                            key, [value]
+                        )
+                    else:
+                        a.history[key] = a.history.get(key, []) + [value]
+
+        return a
+
+    def _read_recursive(self, conf_root, exclude_files: Set[str]) -> ConfigTree:
+        conf = ConfigTree()
+
+        if not conf_root:
+            return conf
+
+        if not conf_root.is_dir():
+            if self.verbose:
+                if not conf_root.exists():
+                    print(f"No config in {conf_root}")
+                else:
+                    print(f"Not a directory: {conf_root}")
+            return conf
+
+        if self.verbose:
+            print(f"Loading config from {conf_root}")
+
+        for file in conf_root.rglob("*.conf"):
+            if exclude_files and file.name in exclude_files:
+                continue
+            key = ".".join(file.relative_to(conf_root).with_suffix("").parts)
+            conf.put(key, self._read_single_file(file))
+
+        return conf
+
+    def _read_single_file(self, file_path):
+        if self.verbose:
+            print(f"Loading config from file {file_path}")
+
+        try:
+            return ConfigFactory.parse_file(file_path)
+        except ParseSyntaxException as ex:
+            msg = f"Failed parsing {file_path} ({ex.__class__.__name__}): (at char {ex.loc}, line:{ex.lineno}, col:{ex.column})"
+            raise ConfigurationError(msg, file_path=file_path) from ex
+        except (ParseException, ParseFatalException, RecursiveGrammarException) as ex:
+            msg = f"Failed parsing {file_path} ({ex.__class__.__name__}): {ex}"
+            raise ConfigurationError(msg) from ex
+        except Exception as ex:
+            print(f"Failed loading {file_path}: {ex}")
+            raise
+
+    def initialize_logging(self):
+        logging_config = self.get("logging", None)
+        if not logging_config:
+            return
+        logging.config.dictConfig(logging_config)
+
+
+class ConfigurationError(Exception):
+    def __init__(self, msg, file_path=None, *args):
+        super().__init__(msg, *args)
+        self.file_path = file_path
+
+
+ConfigType = TypeVar("ConfigType", bound=BasicConfig)

apiserver/config/default/apiserver.conf

@@ -2,8 +2,8 @@
     watch: false            # Watch for changes (dev only)
     debug: false            # Debug mode
     pretty_json: false      # prettify json response
-    return_stack: true      # return stack trace on error
-    log_calls: true         # Log API Calls
+    return_stack: false      # return stack trace on error
+    return_stack_to_caller: false # top-level control on whether to return stack trace in an API response
 
     # if 'return_stack' is true and error contains a status code, return stack trace only for these status codes
     # valid values are:
@@ -41,10 +41,7 @@
         # controls whether FieldDoesNotExist exception will be raised for any extra attribute existing in stored data
         # but not declared in a data model
         strict: false
-
-        aggregate {
-            allow_disk_use: true
-        }
+        ensure_db_version_on_startup: true
     }
 
     elastic {
@@ -62,6 +59,9 @@
         # verify user tokens
         verify_user_tokens: false
 
+        # If set then users that were created from secure credentials or fixed user settings and are no longer in these settings will be deleted on startup
+        delete_missing_autocreated_users: true
+
         # max token expiration timeout in seconds (1 year)
         max_expiration_sec: 31536000
 
@@ -69,19 +69,27 @@
         default_expiration_sec: 2592000
 
         # cookie containing auth token, for requests arriving from a web-browser
-        session_auth_cookie_name: "trains_token_basic"
+        session_auth_cookie_name: "clearml_token_basic"
 
         # cookie configuration for authorization cookies generated by auth.login
         cookies {
             httponly: true  # allow only http to access the cookies (no JS etc)
             secure: false   # not using HTTPS
             domain: null    # Limit to localhost is not supported
+            samesite: Lax
             max_age: 99999999999
         }
 
+        # provide a cookie domain override per company
+#        cookies_domain_override {
+#            <company-id>: <domain>
+#        }
+
 #        # A list of fixed users
+#        # Note: password may be bcrypt-hashed (generate using `python -c 'import bcrypt; print(bcrypt.hashpw("password", bcrypt.gensalt()))'`)
 #        fixed_users {
 #            enabled: true
+#            pass_hashed: false
 #            users: [
 #                {
 #                    username: "john"
@@ -105,9 +113,15 @@
     workers {
         # Auto-register unknown workers on status reports and other calls
         auto_register: true
+        # Assume unknow workers have unregistered (i.e. do not raise unregistered error)
+        auto_unregister: true
         # Timeout in seconds on task status update. If exceeded
         # then task can be stopped without communicating to the worker
         task_update_timeout: 600
+
+        # Timeout in seconds for worker registration (or status report). If a worker did not report for this long,
+        # it is discarded from the server's table
+        default_timeout: 600
     }
 
     check_for_updates {
@@ -116,9 +130,9 @@
         # Check for updates every 24 hours
         check_interval_sec: 86400
 
-        url: "https://updates.trains.allegro.ai/updates"
+        url: "https://updates.clear.ml/updates"
 
-        component_name: "trains-server"
+        component_name: "clearml-server"
 
         # GET request timeout
         request_timeout_sec: 3.0
@@ -128,7 +142,7 @@
         # Note: statistics are sent ONLY if the user has actively opted-in
         supported: true
 
-        url: "https://updates.trains.allegro.ai/stats"
+        url: "https://updates.clear.ml/stats"
 
         report_interval_hours: 24
         agent_relevant_threshold_days: 30
@@ -137,4 +151,11 @@
         max_backoff_sec: 5
     }
 
+    getting_started_info {
+        "agentName": "clearml",
+        "configure": "clearml-init",
+        "install": "pip install clearml",
+        "packageName": "clearml"
+    }
+
 }

apiserver/config/default/hosts.conf

@@ -1,9 +1,10 @@
+fileserver = "http://localhost:8081"
+
 elastic {
   events {
-    hosts: [{host: "127.0.0.1", port: 9211}]
+    hosts: [{host: "127.0.0.1", port: 9200, scheme: http}]
     args {
       timeout: 60
-      dead_timeout: 10
       max_retries: 3
       retry_on_timeout: true
     }
@@ -11,10 +12,9 @@ elastic {
   }
 
   workers {
-    hosts: [{host:"127.0.0.1", port:9211}]
+    hosts: [{host:"127.0.0.1", port:9200, scheme: http}]
     args {
       timeout: 60
-      dead_timeout: 10
       max_retries: 3
       retry_on_timeout: true
     }

apiserver/config/default/logging.conf

@@ -16,7 +16,7 @@
             backupCount: 3
             maxBytes: 10240000,
             class: "logging.handlers.RotatingFileHandler",
-            filename: "/var/log/trains/apiserver.log"
+            filename: "/var/log/clearml/apiserver.log"
         }
     }
     root {

apiserver/config/default/secure.conf

@@ -0,0 +1,44 @@
+{
+    http {
+        session_secret {
+            apiserver: "V8gcW3EneNDcNfO7G_TSUsWe7uLozyacc9_I33o7bxUo8rCN31VLRg"
+        }
+    }
+
+    auth {
+        # token sign secret
+        token_secret: "Rq8FW84sSqVgq7WvBB_4EzNl9y8z8IGiDXX3C345_a5AZfcwZcwCIA"
+    }
+
+    credentials {
+        # system credentials as they appear in the auth DB, used for intra-service communications
+        apiserver {
+            role: "system"
+            user_key: "62T8CP7HGBC6647XF9314C2VY67RJO"
+            user_secret: "gaOfhDX2-bpkeI7-cwEcaMuGijxaG2UG3jbIvg4DxmVGF0LNI7rgvCb1-ne38IlBo1w"
+        }
+        fileserver {
+            role: "system"
+            user_key: "GSQWPEKSKNKF354LC9V6BHXKTYFD5I"
+            user_secret: "tuBXcGQBECsEhcNiK2kiWi750z9r8Z85XrQ9V0c24huTuCb2xf2X1nKG"
+        }
+        webserver {
+            role: "system"
+            user_key: "EYVQ385RW7Y2QQUH88CZ7DWIQ1WUHP"
+            user_secret: "XhkH6a6ds9JBnM_MrahYyYdO-wS2bqFSm8gl-V0UZXH26Ydd6Eyi28TeBEoSr6Z3Bes"
+            revoke_in_fixed_mode: true
+        }
+        services_agent {
+            role: "admin"
+            user_key: ""
+            user_secret: ""
+        }
+        tests {
+            role: "user"
+            display_name: "Default User"
+            user_key: "EGRTCO8JMSIGI6S39GTP43NFWXDQOW"
+            user_secret: "LPEJbGJ6bK4tujQcmrD3i1dbMBDdwUwelVa-LG0K0FFmY9bzH_H0Sw"
+            revoke_in_fixed_mode: true
+        }
+    }
+}

apiserver/config/default/services/_mongo.conf

@@ -0,0 +1,9 @@
+max_page_size: 500
+
+# expiration time in seconds for the redis scroll states in get_many family of apis
+scroll_state_expiration_seconds: 600
+
+allow_disk_use {
+    sort: true
+    aggregate: true
+}

apiserver/config/default/services/async_urls_delete.conf

@@ -0,0 +1,13 @@
+# if set to true then on task delete/reset external file urls for known storage types are scheduled for async delete
+# otherwise they are returned to a client for the client side delete
+enabled: true
+max_retries: 3
+retry_timeout_sec: 60
+
+fileserver {
+    # fileserver url prefixes. Evaluated in the order of priority
+    # Can be in the form <schema>://host:port/path or /path
+    url_prefixes: ["https://files.community-master.hosted.allegro.ai/"]
+    timeout_sec: 300
+    token_expiration_sec: 600
+}

apiserver/config/default/services/events.conf

@@ -0,0 +1,47 @@
+es_index_prefix: "events"
+
+ignore_iteration {
+    metrics: [":monitor:machine", ":monitor:gpu"]
+}
+
+
+events_retrieval {
+    state_expiration_sec: 3600
+
+    # max number of concurrent queries to ES when calculating events metrics
+    # should not exceed the amount of concurrent connections set in the ES driver
+    max_metrics_concurrency: 4
+
+    # If set then max_metrics_count and max_variants_count are calculated dynamically on user data
+    dynamic_metrics_count: true
+
+    # The percentage from the ES aggs limit (10000) to use for the max_metrics and max_variants calculation
+    dynamic_metrics_count_threshold: 80
+
+    # the max amount of metrics to aggregate on
+    max_metrics_count: 100
+
+    # the max amount of variants to aggregate on
+    max_variants_count: 100
+
+    debug_images {
+        # Allow to return the debug images for the variants with uninitialized valid iterations border
+        allow_uninitialized_variants: true
+    }
+
+    max_raw_scalars_size: 200000
+
+    scroll_id_key: "cTN5VEtWEC6QrHvUl0FTx9kNyO0CcCK1p57akxma"
+
+    multi_plots_batch_size: 1000
+}
+
+# if set then plot str will be checked for the valid json on plot add
+# and the result of the check is written to the db
+validate_plot_str: false
+
+# If not 0 then the plots equal or greater to the size will be stored compressed in the DB
+plot_compression_threshold: 100000
+
+# async events delete threshold
+max_async_deleted_events_per_sec: 1000

apiserver/config/default/services/models.conf

@@ -0,0 +1,4 @@
+metadata_values {
+    # cache ttl sec
+    cache_ttl_sec: 86400
+}

apiserver/config/default/services/organization.conf

@@ -0,0 +1,9 @@
+tags_cache {
+  expiration_seconds: 3600
+}
+download {
+    redis_timeout_sec: 300
+    batch_size: 500
+    max_download_items: 50000
+    max_project_name_length: 60
+}

apiserver/config/default/services/projects.conf

@@ -0,0 +1,18 @@
+# Order of featured projects, by name or ID
+featured {
+  order: [
+    #  {id: "<project-id>"}
+    #  OR
+    #  {name: "<project-name>"}
+    #  OR
+    #  {name_regex: "<python-regex>"}
+  ]
+
+  # default featured index for public projects not specified in the order
+  public_default: 9999
+}
+
+sub_projects {
+    # the max sub project depth
+    max_depth: 10
+}

apiserver/config/default/services/queues.conf

@@ -0,0 +1,8 @@
+{
+    metrics_before_from_date: 3600
+    # interval in seconds to update queue metrics. Put 0 to disable
+    metrics_refresh_interval_sec: 300
+    # the queues with these tags will not be returned from get_all/get_all_ex unless id or name specified
+    # or search_hidden is set
+    hidden_tags: [k8s-glue]
+}

apiserver/config/default/services/serving.conf

@@ -0,0 +1,7 @@
+default_container_timeout_sec: 600
+# Auto-register unknown serving containers on status reports and other calls
+container_auto_register: true
+# Assume unknow serving containers have unregistered (i.e. do not raise unregistered error)
+container_auto_unregister: true
+# The minimal sampling interval for serving model monitor chars
+min_chart_interval_sec: 40

apiserver/config/default/services/storage_credentials.conf

@@ -0,0 +1,54 @@
+aws {
+    s3 {
+        # S3 credentials, used for read/write access by various SDK elements
+        # default, used for any bucket not specified below
+        key: ""
+        secret: ""
+        region: ""
+        use_credentials_chain: false
+        # Additional ExtraArgs passed to boto3 when uploading files. Can also be set per-bucket under "credentials".
+        extra_args: {}
+        credentials: [
+            # specifies key/secret credentials to use when handling s3 urls (read or write)
+            # {
+            #     bucket: "my-bucket-name"
+            #     key: "my-access-key"
+            #     secret: "my-secret-key"
+            # },
+//            {
+//                # This will apply to all buckets in this host (unless key/value is specifically provided for a given bucket)
+//                host: "localhost:9000"
+//                key: "minioadmin"
+//                secret: "minioadmin"
+//                # region: my-server
+//                multipart: false
+//                secure: false
+//            }
+        ]
+    }
+}
+google.storage {
+    # Default project and credentials file
+    # Will be used when no bucket configuration is found
+//    project: "clearml"
+//    credentials_json: "/path/to/credentials.json"
+//
+//    # Specific credentials per bucket and sub directory
+//    credentials = [
+//        {
+//            bucket: "my-bucket"
+//            subdir: "path/in/bucket" # Not required
+//            project: "clearml"
+//            credentials_json: "/path/to/credentials.json"
+//        },
+//    ]
+}
+azure.storage {
+    # containers: [
+    #     {
+    #         account_name: "clearml"
+    #         account_key: "secret"
+    #         # container_name:
+    #     }
+    # ]
+}

apiserver/config/default/services/tasks.conf

@@ -0,0 +1,28 @@
+non_responsive_tasks_watchdog {
+    enabled: true
+
+    # In-progress tasks older than this value in seconds will be stopped by the watchdog
+    threshold_sec: 7200
+
+    # Watchdog will sleep for this number of seconds after each cycle
+    watch_interval_sec: 900
+}
+
+multi_task_histogram_limit: 100
+
+hyperparam_values {
+    # max allowed outdate time for the cashed result
+    cache_allowed_outdate_sec: 60
+
+    # cache ttl sec
+    cache_ttl_sec: 86400
+}
+
+# the maximum amount of unique last metrics/variants combinations
+# for which the last values are stored in a task
+max_last_metrics: 2000
+
+# if set then call to tasks.delete/cleanup does not wait for ES events deletion
+async_events_delete: true
+# do not use async_delete if the deleted task has amount of events lower than this threshold
+async_events_delete_threshold: 100000

apiserver/config/default/services/workers.conf

@@ -0,0 +1,9 @@
+default_worker_timeout_sec: 600
+default_cluster_timeout_sec: 600
+
+# The minimal sampling interval for resource dashboard and worker activity charts
+min_chart_interval_sec: 40
+
+stats {
+    max_metrics_concurrency: 4
+}