2020-12-22 23:33:17 +00:00
< div align = "center" >
2022-01-27 09:48:19 +00:00
< a href = "https://app.clear.ml" > < img src = "https://github.com/allegroai/clearml/blob/master/docs/clearml-logo.svg?raw=true" width = "250px" > < / a >
2020-12-22 23:33:17 +00:00
2020-12-22 23:34:03 +00:00
## **`clearml-session` </br> CLI for launching JupyterLab / VSCode on a remote machine**
2020-12-22 23:33:17 +00:00
[](https://img.shields.io/github/license/allegroai/clearml-session.svg)
[](https://img.shields.io/pypi/pyversions/clearml-session.svg)
[](https://img.shields.io/pypi/v/clearml-session.svg)
[](https://pypi.python.org/pypi/clearml-session/)
[](https://join.slack.com/t/allegroai-trains/shared_invite/zt-c0t13pty-aVUZZW1TSSSg2vyIGVPBhg)
< / div >
2020-12-22 19:32:02 +00:00
**`clearml-session`** is a utility for launching detachable remote interactive sessions (MacOS, Windows, Linux)
### tl;dr
CLI to launch remote sessions for JupyterLab / VSCode-server / SSH, inside any docker image!
### What does it do?
Starting a clearml (ob)session from your local machine triggers the following:
- ClearML allocates a remote instance (GPU) from your dedicated pool
- On the allocated instance it will spin **jupyter-lab** + **vscode server** + **SSH** access for
interactive usage (i.e., development)
2022-01-24 08:36:02 +00:00
- ClearML will start monitoring machine performance, allowing DevOps to detect stale instances and spin them down
> ℹ ️
2020-12-22 19:32:02 +00:00
### Use-cases for remote interactive sessions:
1. Development requires resources not available on the current developer's machines
2. Team resource sharing (e.g. how to dynamically assign GPUs to developers)
3. Spin a copy of a previously executed experiment for remote debugging purposes (:open_mouth:!)
4. Scale-out development to multiple clouds, assign development machines on AWS/GCP/Azure in a seamless way
## Prerequisites:
* **An SSH client installed on your machine** - To verify open your terminal and execute `ssh` , if you did not receive an error, we are good to go.
* At least one `clearml-agent` running on a remote host. See installation [details ](https://github.com/allegroai/clearml-agent ).
Supported OS: MacOS, Windows, Linux
## Secure & Stable
**clearml-session** creates a single, secure, and encrypted connection to the remote machine over SSH.
SSH credentials are automatically generated by the CLI and contain fully random 32 bytes password.
All http connections are tunneled over the SSH connection,
allowing users to add additional services on the remote machine (!)
Furthermore, all tunneled connections have a special stable network layer allowing you to refresh the underlying SSH
connection without breaking any network sockets!
This means that if the network connection is unstable, you can refresh
the base SSH network tunnel, without breaking JupyterLab/VSCode-server or your own SSH connection
(e.h. debugging over SSH with PyCharm)
---
## How to use: Interactive Session
1. run `clearml-session`
2. select the requested queue (resource)
3. wait until a machine is up and ready
4. click on the link to the remote JupyterLab/VSCode OR connect with the provided SSH details
**Notice! You can also**: Select a **docker image** to execute in, install required **python packages** , run **bash script** ,
pass **git credentials** , etc.
See below for full CLI options.
## Frequently Asked Questions:
2022-01-24 08:36:02 +00:00
#### How Does ClearML enable this?
2020-12-22 19:32:02 +00:00
The `clearml-session` creates a new interactive `Task` in the system (default project: DevOps).
This `Task` is responsible for setting the SSH and JupyterLab/VSCode on the host machine.
The local `clearml-session` awaits for the interactive Task to finish with the initial setup, then
it connects via SSH to the host machine (see "safe and stable" above), and tunnels
both SSH and JupyterLab over the SSH connection.
The end results is a local link which you can use to access the JupyterLab/VSCode on the remote machine, over a **secure and encrypted** connection!
#### How can this be used to scale up/out development resources?
**Clearml** has a cloud autoscaler, so you can easily and automatically spin machines for development!
There is also a default docker image to use when initiating a task.
This means that using **clearml-session**s
with the autoscaler enabled, allows for turn-key secure development environment inside a docker of your choosing.
2022-07-11 08:18:19 +00:00
Learn more about it [here ](https://clear.ml/docs/latest/docs/guides/services/aws_autoscaler )
2020-12-22 19:32:02 +00:00
#### Does this fit Work From Home situations?
**YES**. Install `clearml-agent` on target machines inside the organization, connect over your company VPN
and use `clearml-session` to gain access to a dedicated on-prem machine with the docker of your choosing
(with out-of-the-box support for any internal docker artifactory).
2022-07-11 08:18:19 +00:00
Learn more about how to utilize your office workstations and on-prem machines [here ](https://clear.ml/docs/latest/docs/clearml_agent )
2020-12-22 19:32:02 +00:00
## Tutorials
### Getting started
2022-07-11 08:18:19 +00:00
Requirements `clearml` python package installed and configured (see detailed [instructions ](https://clear.ml/docs/latest/docs/getting_started/ds/ds_first_steps ))
2020-12-22 19:32:02 +00:00
``` bash
pip install clearml-session
2022-07-11 05:58:15 +00:00
clearml-session --docker nvcr.io/nvidia/pytorch:20.11-py3 --git-credentials
2020-12-22 19:32:02 +00:00
```
Wait for the machine to spin up:
Expected CLI output would look something like:
``` console
Creating new session
New session created [id=3d38e738c5ff458a9ec465e77e19da23]
Waiting for remote machine allocation [id=3d38e738c5ff458a9ec465e77e19da23]
.Status [queued]
....Status [in_progress]
Remote machine allocated
Setting remote environment [Task id=3d38e738c5ff458a9ec465e77e19da23]
Setup process details: https://app.community.clear.ml/projects/64ae77968db24b27abf86a501667c330/experiments/3d38e738c5ff458a9ec465e77e19da23/output/log
Waiting for environment setup to complete [usually about 20-30 seconds]
..............
Remote machine is ready
Setting up connection to remote session
Starting SSH tunnel
Warning: Permanently added '[192.168.0.17]:10022' (ECDSA) to the list of known hosts.
root@192.168.0.17's password: f7bae03235ff2a62b6bfbc6ab9479f9e28640a068b1208b63f60cb097b3a1784
Interactive session is running:
SSH: ssh root@localhost -p 8022 [password: f7bae03235ff2a62b6bfbc6ab9479f9e28640a068b1208b63f60cb097b3a1784]
Jupyter Lab URL: http://localhost:8878/?token=df52806d36ad30738117937507b213ac14ed638b8c336a7e
VSCode server available at http://localhost:8898/
Connection is up and running
Enter "r" (or "reconnect") to reconnect the session (for example after suspend)
Ctrl-C (or "quit") to abort (remote session remains active)
or "Shutdown" to shutdown remote interactive session
```
Click on the JupyterLab link (http://localhost:8878/?token=xyz)
Open your terminal, clone your code & start working :)
### Leaving a session and reconnecting from the same machine
On the `clearml-session` CLI terminal, enter 'quit' or press Ctrl-C
It will close the CLI but leaves the remote session running
When you want to reconnect to it, execute:
``` bash
clearml-session
```
Then press "Y" (or enter) to reconnect to the already running session
``` console
clearml-session - launch interactive session
Checking previous session
Connect to active session id=3d38e738c5ff458a9ec465e77e19da23 [Y]/n?
```
### Shutting down a remote session
On the `clearml-session` CLI terminal, enter 'shutdown' (case-insensitive)
It will shut down the remote session, free the resource and close the CLI
``` console
Enter "r" (or "reconnect") to reconnect the session (for example after suspend)
Ctrl-C (or "quit") to abort (remote session rema
Yes of course, current SSO supports Google/GitHub/BitBucket/... + SAML/LDAP (Usually with user
ins active)
or "Shutdown" to shutdown remote interactive session
shutdown
Shutting down interactive session
Interactive session ended
Leaving interactive session
```
### Connecting to a running interactive session from a different machine
Continue working on an interactive session from **any** machine.
In the `clearml` web UI, go to DevOps project, and find your interactive session.
Click on the ID button next to the Task name, and copy the unique ID.
``` bash
clearml-session --attach < session_id_here >
```
Click on the JupyterLab/VSCode link, or connect directly to the SSH session
### Debug a previously executed experiment
If you have a previously executed experiment in the system,
you can create an exact copy of the experiment and debug it on the remote interactive session.
`clearml-session` will replicate the exact remote environment, add JupyterLab/VSCode/SSH and allow you interactively
execute and debug the experiment, on the allocated remote machine.
In the `clearml` web UI, find the experiment (Task) you wish to debug.
Click on the ID button next to the Task name, and copy the unique ID.
``` bash
2021-12-12 21:22:51 +00:00
clearml-session --debugging-session < experiment_id_here >
2020-12-22 19:32:02 +00:00
```
Click on the JupyterLab/VSCode link, or connect directly to the SSH session
## CLI options
``` bash
clearml-session --help
```
``` console
clearml-session - CLI for launching JupyterLab / VSCode on a remote machine
usage: clearml-session [-h] [--version] [--attach [ATTACH]]
2022-01-01 21:20:21 +00:00
[--debugging-session DEBUGGING_SESSION] [--queue QUEUE]
2021-12-12 21:22:51 +00:00
[--docker DOCKER] [--docker-args DOCKER_ARGS]
[--public-ip [true/false]]
[--remote-ssh-port REMOTE_SSH_PORT]
2020-12-22 19:32:02 +00:00
[--vscode-server [true/false]]
2021-12-12 21:22:51 +00:00
[--vscode-version VSCODE_VERSION]
2021-02-21 16:22:01 +00:00
[--jupyter-lab [true/false]]
2020-12-22 19:32:02 +00:00
[--git-credentials [true/false]]
[--user-folder USER_FOLDER]
[--packages [PACKAGES [PACKAGES ...]]]
[--requirements REQUIREMENTS]
[--init-script [INIT_SCRIPT]]
[--config-file CONFIG_FILE]
[--remote-gateway [REMOTE_GATEWAY]]
[--base-task-id BASE_TASK_ID] [--project PROJECT]
2022-01-01 21:20:21 +00:00
[--keepalive [true/false]]
2020-12-22 19:32:02 +00:00
[--queue-excluded-tag [QUEUE_EXCLUDED_TAG [QUEUE_EXCLUDED_TAG ...]]]
[--queue-include-tag [QUEUE_INCLUDE_TAG [QUEUE_INCLUDE_TAG ...]]]
[--skip-docker-network] [--password PASSWORD]
2022-01-01 21:20:21 +00:00
[--username USERNAME] [--verbose]
2020-12-22 19:32:02 +00:00
clearml-session - CLI for launching JupyterLab / VSCode on a remote machine
optional arguments:
-h, --help show this help message and exit
--version Display the clearml-session utility version
--attach [ATTACH] Attach to running interactive session (default:
previous session)
2022-01-01 21:20:21 +00:00
--debugging-session DEBUGGING_SESSION
2020-12-22 19:32:02 +00:00
Pass existing Task id (experiment), create a copy of
the experiment on a remote machine, and launch
jupyter/ssh for interactive access. Example
2021-12-12 21:22:51 +00:00
--debugging-session < task_id >
2020-12-22 19:32:02 +00:00
--queue QUEUE Select the queue to launch the interactive session on
(default: previously used queue)
--docker DOCKER Select the docker image to use in the interactive
session on (default: previously used docker image or
`nvidia/cuda:10.1-runtime-ubuntu18.04` )
2021-12-12 21:22:51 +00:00
--docker-args DOCKER_ARGS
Add additional arguments for the docker image to use
in the interactive session on (default: previously
used docker-args)
2020-12-22 19:32:02 +00:00
--public-ip [true/false]
If True register the public IP of the remote machine.
Set if running on the cloud. Default: false (use for
local / on-premises)
2021-12-12 21:22:51 +00:00
--remote-ssh-port REMOTE_SSH_PORT
Set the remote ssh server port, running on the agent`s
machine. (default: 10022)
2020-12-22 19:32:02 +00:00
--vscode-server [true/false]
2021-02-21 16:22:01 +00:00
Install vscode server (code-server) on interactive
2020-12-22 19:32:02 +00:00
session (default: true)
2021-12-12 21:22:51 +00:00
--vscode-version VSCODE_VERSION
Set vscode server (code-server) version, as well as
vscode python extension version < vscode:python-ext >
(example: "3.7.4:2020.10.332292344")
2021-02-21 16:22:01 +00:00
--jupyter-lab [true/false]
Install Jupyter-Lab on interactive session (default:
true)
2020-12-22 19:32:02 +00:00
--git-credentials [true/false]
If true, local .git-credentials file is sent to the
interactive session. (default: false)
--user-folder USER_FOLDER
Advanced: Set the remote base folder (default: ~/)
--packages [PACKAGES [PACKAGES ...]]
Additional packages to add, supports version numbers
(default: previously added packages). examples:
--packages torch==1.7 tqdm
--requirements REQUIREMENTS
Specify requirements.txt file to install when setting
the interactive session. Requirements file is read and
stored in `packages` section as default for the next
sessions. Can be overridden by calling `--packages`
--init-script [INIT_SCRIPT]
Specify BASH init script file to be executed when
setting the interactive session. Script content is
read and stored as default script for the next
sessions. To clear the init-script do not pass a file
--config-file CONFIG_FILE
Advanced: Change the configuration file used to store
2021-02-21 16:22:01 +00:00
the previous state (default: ~/.clearml_session.json)
2020-12-22 19:32:02 +00:00
--remote-gateway [REMOTE_GATEWAY]
2021-12-12 21:22:51 +00:00
Advanced: Specify gateway ip/address:port to be passed
to interactive session (for use with k8s ingestion /
ELB)
2020-12-22 19:32:02 +00:00
--base-task-id BASE_TASK_ID
Advanced: Set the base task ID for the interactive
session. (default: previously used Task). Use `none`
for the default interactive session
--project PROJECT Advanced: Set the project name for the interactive
session Task
2022-01-01 21:20:21 +00:00
--keepalive [true/false]
Advanced: If set, enables the transparent proxy always
keeping the sockets alive. Default: False, do not use
transparent socket for mitigating connection drops.
2020-12-22 19:32:02 +00:00
--queue-excluded-tag [QUEUE_EXCLUDED_TAG [QUEUE_EXCLUDED_TAG ...]]
Advanced: Excluded queues with this specific tag from
the selection
--queue-include-tag [QUEUE_INCLUDE_TAG [QUEUE_INCLUDE_TAG ...]]
Advanced: Only include queues with this specific tag
from the selection
--skip-docker-network
Advanced: If set, `--network host` is **not** passed
to docker (assumes k8s network ingestion) (default:
false)
--password PASSWORD Advanced: Select ssh password for the interactive
2021-02-21 16:22:01 +00:00
session (default: `randomly-generated` or previously
used one)
--username USERNAME Advanced: Select ssh username for the interactive
session (default: `root` or previously used one)
2021-12-12 21:22:51 +00:00
--verbose Advanced: If set, print verbose progress information,
e.g. the remote machine setup process log
2020-12-22 19:32:02 +00:00
Notice! all arguments are stored as new defaults for the next session
```
