Skip to content
This repository was archived by the owner on Sep 30, 2026. It is now read-only.

Move the Linux CI jobs to OSDC runners - #1577

Draft
huydhn wants to merge 1 commit into
mainfrom
osdc/ci-runners
Draft

huydhn wants to merge 1 commit into
mainfrom
osdc/ci-runners

Conversation

@huydhn

@huydhn huydhn commented Sep 25, 2026

Copy link
Copy Markdown
Contributor
job before after
linux_cpu_kineto, linux_cpu_pytorch pr-test linux.12xlarge mt-l-x86iavx512-48-384
linux_cuda_kineto, linux_cuda_pytorch pr-test linux.g5.4xlarge.nvidia.gpu mt-l-x86aavx2-11-41-a10g
linux_rocm_kineto build linux.4xlarge mt-l-x86iavx512-16-128
linux_rocm_pytorch build linux.12xlarge mt-l-x86iavx512-48-384

The GPU pod sits on the same g5.4xlarge the label resolved to, so the GPU is unchanged and only the Kubernetes overhead comes off.

Only the jobs whose runner actually moves go to linux_job_v3. OSDC needs v3 because it rejects a job with no container, and v3 supplies pytorch/almalinux-builder for the arch.

The ROCm test jobs and both XPU jobs stay on v2 with their existing runners. v3 pulls the job image from the private ECR and neither runner can authenticate to it. The ROCm workflows are mixed — their x86 build job moves, their test job does not.

Authored with Claude Code.

Only the jobs whose runner moves go to linux_job_v3, which OSDC needs because
it rejects a job with no container. The ROCm test and XPU jobs stay on v2 with
their existing runners: v3 pulls the job image from the private ECR and those
runners cannot authenticate to it.
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant