Repository navigation
Conversation
These jobs are hand-rolled copies of test-infra's linux_job rather than
callers of it, so this maps the labels directly, per pytorch/pytorch's
.github/arc.yaml:
linux.4xlarge -> mt-l-x86iavx512-16-128
linux.12xlarge -> mt-l-x86iavx512-48-384
linux.g4dn.12xlarge.nvidia.gpu -> mt-l-x86iavx512-45-172-t4-4
Each of those jobs also opened with
sudo rm -rfv "${GITHUB_WORKSPACE}"
mkdir -p "${GITHUB_WORKSPACE}"
On OSDC the workspace is a bind mount, so removing the directory detaches it
and every later step fails with "getcwd: cannot access parent directories".
Clearing the contents keeps the inode, which is what test-infra's own
templates do after pytorch/test-infra#8789.
gpu-hvd-tests.yml is left alone: linux.8xlarge.nvidia.gpu has no entry in
arc.yaml, so there is no mapping to make, and guessing a GPU model is not the
same job.
Authored with Claude Code.
Each job gets a fresh pod, and GITHUB_WORKSPACE lives on an emptyDir volume created with it (verified: the workflow pod's 'work' volume is emptyDir, and runners are one job per pod). So the workspace is already empty at job start and there is nothing to clear. The step existed because the EC2 runners were reused across jobs and left stale content behind. gpu-hvd-tests.yml and mps-tests.yml keep theirs -- they are still on EC2. Authored with Claude Code.
linux.8xlarge.nvidia.gpu -> mt-l-x86iavx512-45-172-t4-4, the label gpu-tests already uses in this PR. No OSDC label carries exactly two GPUs, and this suite runs 'run_gpu_tests.sh 2 hvd', so the 4x T4 pod is the fit that keeps the test as-is: it asks for two of four. The GPU model changes M60 -> T4, which is unavoidable here. Drops the workspace cleanup for the same reason as the other jobs: the pod is ephemeral and GITHUB_WORKSPACE is an emptyDir, so it starts empty. Authored with Claude Code.
This branch has not been deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
These jobs are hand-rolled copies of test-infra's
linux_jobrather than callers of it, so this maps the labels directly, per pytorch/pytorch.github/arc.yaml:linux.4xlargemt-l-x86iavx512-16-128linux.12xlargemt-l-x86iavx512-48-384linux.g4dn.12xlarge.nvidia.gpumt-l-x86iavx512-45-172-t4-4Each job also opened with:
On OSDC the workspace is a bind mount, so removing the directory detaches it and every later step fails with
getcwd: cannot access parent directories. Clearing the contents keeps the inode — what test-infra's own templates do since pytorch/test-infra#8789.gpu-hvd-tests.ymlis deliberately left alone:linux.8xlarge.nvidia.gpuhas no entry inarc.yaml, so there is no mapping to make and guessing a GPU model would change what the job tests.Authored with Claude Code.