Commit Graph

97 Commits

Author SHA1 Message Date
allegroai
69eb25db1f Fix running trains-agent from conda environment - conda.sh not found in first conda PATH match 2020-12-10 09:53:18 +02:00
allegroai
a41ea52f87 Add multiple packages support 2020-12-10 09:52:00 +02:00
allegroai
259113c989 Add PackageCollectorRequirement to allow multiple entries of the same package 2020-12-06 12:16:56 +02:00
allegroai
1afa3a3914 Add torchcsprng and torchtext to PyTorch resolving. Improve debug prints on auto cuda version resolving. 2020-12-06 12:15:12 +02:00
allegroai
b0c0f41f62 Allow zero context diffs (useful when blind patching repository) 2020-12-06 12:13:28 +02:00
allegroai
58eb5fbd5f Fix torch CUDA 11.1 support 2020-11-26 01:14:36 +02:00
allegroai
52f30b306a Fix git diff with empty line at the end of the git diff will cause corrupt diff apply message 2020-11-11 17:11:28 +02:00
allegroai
6df0f81ca0 Fix uid is None causes ValueError in str.startswith(). Fix str.split (should be on the filename itself, not the path). 2020-11-11 16:32:47 +02:00
allegroai
a61265effe Improve trying to find conda executable 2020-11-11 16:29:50 +02:00
allegroai
216b3e2179 Allow to specifying cudatoolkit version in "installed packages" when using Conda as package manager (trains issue #229) 2020-10-30 10:06:02 +02:00
allegroai
293a92f486 Improve k8s glue add --template-yaml 2020-10-23 01:28:22 +03:00
allegroai
6bad2b5352 Fix support non-ascii git diff 2020-10-23 01:27:59 +03:00
allegroai
5a510882b8 Ignore environment SSH_AUTH_SOCK. Only check if git_user/pass are configured, if they are not, leave the links as they are 2020-10-21 19:02:29 +03:00
allegroai
14655f19a0 Fix conda PYTHONPATH (point only to code, not to venv) 2020-10-15 23:26:58 +03:00
allegroai
8e6fce8d63 Add conda support for read-only pre-built environment (pass conda folder as docker_cmd on Task).
Fix conda restore prebuild tar.gz file, fix conda prefix by call conda-unpack from unzipped conda env.
2020-10-15 23:25:57 +03:00
allegroai
eb942cfedd Add agent.package_manager.conda_env_as_base_docker allowing "docker_cmd" to contain link to a full pre-packaged conda environment (conda-pack outputs a tar.gz). Use TRAINS_CONDA_ENV_PACKAGE to specify conda tar.gz file. 2020-10-15 23:23:46 +03:00
allegroai
a2156e73bf Fix conda pip freeze to be consistent with trains 0.16.3 2020-10-11 11:25:35 +03:00
allegroai
9fe77f3c28 Fix conda environment support for trains 0.16.3 full env. Add agent.package_manager.conda_full_env_update to allow conda to update back the requirements (default is false, to preserve previous behavior) 2020-10-11 11:24:52 +03:00
allegroai
6f078afafd Add Requirement.clone() 2020-10-11 11:21:49 +03:00
allegroai
15f4aa613e Suppress "\r" when reading a current chunk of a file. Add agent.suppress_carriage_return (default True) to support previous behavior. 2020-10-11 11:21:08 +03:00
allegroai
234d5fac2c When using force ssh protocol, only enforce on git_host if provided, otherwise apply everywhere 2020-10-05 18:26:21 +03:00
allegroai
6cbfb96ff8 Rename git_domain to git_host 2020-10-05 11:25:03 +03:00
allegroai
6e54e55c31 Add agent.force_git_ssh_port to control https to ssh link conversion for non standard ssh port 2020-10-04 19:42:44 +03:00
allegroai
3ff85b7b85 Replace back package version on conda and pip 2020-10-04 19:41:26 +03:00
allegroai
5640489f57 Replace torch version on pre-installed local file 2020-10-04 19:40:39 +03:00
allegroai
8135a6facf Add agent.git_domain setting for limiting git credential usage for a specific domain (env var TRAINS_AGENT_GIT_DOMAIN is also supported) 2020-10-04 19:40:04 +03:00
allegroai
b6ae4f211d Fix "package @ " should processed by us (pip will not test pre-installed version of the package compared with the link) 2020-10-04 19:38:33 +03:00
allegroai
a56f032ec4 Fix torch support to not change back the same link 2020-10-04 19:37:12 +03:00
allegroai
075736de20 Translate downloaded URL back to original link when new pip version is installed (otherwise we end up with file:///... links) 2020-10-04 19:36:14 +03:00
allegroai
d8543c892e When new pip version is installed, no need to install git packages twice (pip freeze will detect the correct git link version) 2020-10-04 19:35:26 +03:00
allegroai
c7a739fafa Add support for detecting new pip version (20+) supporting @ in requirements 2020-10-04 19:33:52 +03:00
allegroai
3bed0ef33c Add protection against bad file name parsing in git diff apply 2020-10-04 19:31:48 +03:00
allegroai
d419fa1e4f Update torch version after using system pre-installed version 2020-10-04 19:29:47 +03:00
allegroai
31a56c71bd Add preliminary agent uptime/downtime support 2020-09-29 19:34:51 +03:00
allegroai
28f47419b0 Fix incorrect check for spaces in current execution folder (only check in cache folders) 2020-09-15 20:26:02 +03:00
allegroai
6a24da2849 Add post_packages post_optional_packages to control packages installed after all the rest (e.g. horovod)
Rename CythonReq to PriorityPackageRequirement and HorovodReq to PostRequirement
2020-09-15 20:20:55 +03:00
allegroai
aaf8d802e7 Update documentation 2020-09-05 16:37:17 +03:00
allegroai
ca89a1e322 Fix pre-installed packages are ignored when installing a git package wheel. Reinstalling a git+http link is enough to make sure all requirements are met/installed (trains issue #196) 2020-09-05 16:37:17 +03:00
allegroai
4aacf9005e Fix GPU Windows monitoring support (Trains Issue #177) 2020-08-10 08:07:51 +03:00
allegroai
ce6831368f Fix GPU monitoring on Windows machines 2020-08-08 14:43:25 +03:00
allegroai
e4111c830b Fix GIT user/pass in requirements and support for '-e git+http' lines 2020-07-30 14:30:23 +03:00
allegroai
52c1772b04 Add requirement_parser into trains-agent instead as a dependency. Fix requirement_parser to support 'package @ git+http' lines 2020-07-30 14:29:37 +03:00
allegroai
17d2bf2a3e Change daemon --stop without any specific flag to terminate the agents by worker id lexicographic order 2020-07-11 01:43:54 +03:00
allegroai
94997f9c88 Add daemon --order-fairness for round-robin queue pulling
Add daemon --stop to terminate running agent (assume all the rest of the arguments are the same)
Clean up all log files on termination unless executed with --debug
2020-07-11 01:42:56 +03:00
allegroai
c6d998c4df Add terminate process and rmtree utilities 2020-07-11 01:40:50 +03:00
allegroai
1f53a06299 Add agent.force_git_ssh_protocol option to force all git links to ssh:// (issue #16)
Add git user/pass credentials for pip git packages (git+http and  git+ssh) (issue #22)
2020-06-18 01:55:14 +03:00
allegroai
54d9d77294 Allow services mode to re-register (docker can kill it and not exit gracefully) 2020-06-01 16:34:33 +03:00
allegroai
566427d550 Fix build failing due to missing session 2020-05-31 14:02:42 +03:00
allegroai
cc99077c92 Do not monitor GPU when running with --cpu-only 2020-05-31 14:01:14 +03:00
allegroai
22c5f043aa Fix detached mode to correctly use cache folder slots 2020-05-31 14:00:14 +03:00