✨ feat(Ray): Enhance Ray

[binw666]

Description

This PR:

• Add task_runner.py to support specifying resources, py_modules, and pip.

Fixes # (issue)

Type of change

• New feature (non-breaking change which adds functionality)

Checklist

• I have run the pre-commit checks with ./isaaclab.sh --format
• I have made corresponding changes to the documentation
• My changes generate no new warnings
• I have added tests that prove my fix is effective or that my feature works
• I have updated the changelog and the corresponding version in the extension's config/extension.toml file
• I have added my name to the CONTRIBUTORS.md or my name already exists there

[binw666]

@ozhanozen @garylvov I have added task_runner.py as an additional method for submitting jobs. This method supports specifying resources, py_modules, and pip dependencies. If you have time, could you please help review the code? Thank you!

[garylvov]

Hi I will review by the end of this weekend please ping me if no activity within 1 week

[garylvov]

Nit: can you show a real example in a config file as well if possible? Maybe will make this example a bit more clear

[binw666]

In the most recent commit, I have updated the comments and included an official example from IsaacLab for multi-GPU training. This may help clarify the usage.

[garylvov]

Similar to https://github.com/binw666/IsaacLab/blob/752be19bade88c1f1e2a06a1ca6519baafbba216/scripts/reinforcement_learning/ray/submit_job.py#L79 seems like duplicate logic that could be unified

[binw666]

This does resemble the submit_job function in submit_job.py, but it serves a different purpose. Here, the function calls remote to execute, whereas submit_job submits tasks via client.submit_job. Do you have any better suggestions?

[garylvov]

Hi @binw666 thank you for the PR please let me know if you have any questions about my comments I'm happy to help. Once you address these issues just rerequest me for the review or @ me and I'll check it out again.

Thanks!

[binw666]

@garylvov Thank you very much for your review. I have updated the code and documentation to address the above issues.

[ozhanozen]

@binw666, I don't have a HW currently that I can use to test the functionality of the PR, but I try to give feedback on other stuff as much as possible.

I noticed that some of the highlighted line numbers of the file ray.rst are outdated (some before your changes). Would it be possible for you to go over the file and update them as the last step once everything else is done?

[ozhanozen]

@garylvov @binw666

I’m not sure if this PR is the right place to bring this up or if it should be addressed in a separate one, but based on my experience, I believe the current Ray integration is missing a key feature: support for a custom CLIReporter.

Right now, the integration defaults to the standard CLIReporter, which often displays metrics that aren’t particularly relevant. Similar to how we allow users to specify a --cfg_class (e.g., CartpoleTheiaJobCfgE), I suggest we let them optionally provide a custom CLIReporter class (that goes to the RunConfig as an argument). If none is provided, we can fall back to the default.

What do you think about this?

[garylvov]

I think a custom CLIReporter could be good. As for including it in this PR, I am personally ok with it, but it's up to @binw666 . I could see why @binw666 may want to limit scope here, in which case we can follow up with another PR after this one is merged.

[garylvov]

That being said, it's probably quickest to include that functionality in this PR in terms of getting things merged into main