Skip to content

Angehu/perf gate poc - #3

Draft
angehu-nv wants to merge 2 commits into
NVIDIA-Omniverse:developfrom
angehu-nv:angehu/perf-gate-poc
Draft

Angehu/perf gate poc#3
angehu-nv wants to merge 2 commits into
NVIDIA-Omniverse:developfrom
angehu-nv:angehu/perf-gate-poc

Conversation

@angehu-nv

Copy link
Copy Markdown

No description provided.

@angehu-nv
angehu-nv force-pushed the angehu/perf-gate-poc branch from 67af377 to df15280 Compare June 15, 2026 23:56
Neil4561 pushed a commit that referenced this pull request Jun 16, 2026
Replace the tools/perf_regression_gate tree wholesale with the latest
revision from PR #3 (angehu/perf-gate-poc, "hardening fixes"), then
re-layer the runner/ops wiring needed to actually run on the NVIDIA
RTX-6000 fleet:

- Public Isaac Sim image override (fork cannot pull nvcr.io/nvidian/*),
  NGC_API_KEY env, bench_matrix output, full checkout for the matrix step.
- World-writable jit-cache + artifact bind mounts (chmod 0777) and
  WARP/CUDA cache env/mounts to fix the wp.init() PermissionError that the
  lazy task loader masked as a spurious mdp AttributeError.
- Rolling JIT cache key so the warm cache refills each run instead of
  freezing empty, plus an opt-in type=gha image layer cache (gha-cache-scope)
  for fleets where ECR cannot be resolved.
- Trim triggers to the runnable push path (pull-request/<N> + protected
  branches + manual dispatch); pull_request/merge_group never get fleet
  runners.
- Rename baseline branch angehu/perf-baselines -> perf-baselines.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant