ci: Multi-tenancy for tests and garbage collection

[ko3n1g]

What does this PR do ?

More robust CI pipeline by creating isolated directories for job-outputs.

Collection: [Note which collection this PR will affect]

Changelog

• Add specific line by line info of high level changes in this PR.

Usage

• You can potentially add a usage example below

# Add a code snippet demonstrating how to use this 

GitHub Actions CI

The Jenkins CI system has been replaced by GitHub Actions self-hosted runners.

The GitHub Actions CI will run automatically when the "Run CICD" label is added to the PR.
To re-run CI remove and add the label again.
To run CI on an untrusted fork, a NeMo user with write access must first click "Approve and run".

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you add or update any necessary documentation?
• Does the PR affect components that are optional to install? (Ex: Numba, Pynini, Apex etc)
  • Reviewer: Does the PR have correct import guards for all optional libraries?

PR Type:

• New Feature
• Bugfix
• Documentation

If you haven't finished some of the above items you can still open "Draft" PR.

Who can review?

Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.

Additional Information

• Related to # (issue)

[ko3n1g]

Hey @pablo-garay , this is an proposal how we could make our CI pipeline a bit more stable. We should/could isolate folders that are being written to, and garbage-collect them by if: always() steps.

The only caveat I can see is that templating our tests might become a little bit more tricky. But I don't think it would be a blocker. See #9177 for reference

[pablo-garay]

You can just make it be:

        mkdir -p /home/TestData/nlp/megatron_gpt/starcoder-ci-hf/${{ github.run_id }};
        python scripts/checkpoint_converters/convert_starcoder_hf_to_nemo.py \
        --input_name_or_path /home/TestData/nlp/megatron_gpt/starcoder-ci-hf \
        --output_path /home/TestData/nlp/megatron_gpt/starcoder-ci-hf/${{ github.run_id }}
        rm -rf /home/TestData/nlp/megatron_gpt/starcoder-ci-hf/megatron_starcoder_tp1_pp1.nemo; 
        rm -rf /home/TestData/nlp/megatron_gpt/starcoder-ci-hf/${{ github.run_id }}/

[pablo-garay]

No need for another Cleanup step. Make it all in just one script

[ko3n1g]

Yes that would help us to keep a more unified test case template, but we would need a different (maybe scheduled?) workflow that cleans up directories that weren't cleaned up due to workflow cancellation.

not a big issue, what do you prefer?

[pablo-garay]

we would need a different (maybe scheduled?) workflow that cleans up directories that weren't cleaned up due to workflow cancellation.

Ah good point. Let's do as you have it here then. Scheduled workflow will also have same issue given other workflow can be running that test & it will be synchonization mess.

Leave it as it is now then :) and please kindly do this analogous change for all the other tests prefixed "L2_Community_LLM_Checkpoints_tests_"

I.e. all the tests that are affected by this:
rm -rvf /mnt/datadrive/TestData/nlp/megatron_llama/llama3-ci-hf/model_weights
rm -rvf /mnt/datadrive/TestData/nlp/megatron_llama/model_weights
rm -rvf /mnt/datadrive/TestData/nlp/megatron_llama/llama3-ci-hf/model_weights
rm -rvf /mnt/datadrive/TestData/nlp/megatron_gpt/falcon-ci-hf/model_weights
rm -rvf /mnt/datadrive/TestData/nlp/megatron_gpt/starcoder-ci-hf/model_weights

[pablo-garay]

Fantastic proposal

We need this for all tests whose name start with prefix: L2_Community_LLM_Checkpoints_tests_
Could you please kindly make the proposed change & add for the other tests? :)

[pablo-garay]

We just need to do for all the tests that I mentioned to you offline.

[pablo-garay]

Not sure why there are two cleanups
Will merge & can do follow up PR to sort this out if/as needed