Skip to content

ci: point benchmark registry_file at tpu_sync/ after directory migration - #662

Closed
Sahil170595 wants to merge 1 commit into
google:mainfrom
Sahil170595:fix-benchmark-registry-path
Closed

ci: point benchmark registry_file at tpu_sync/ after directory migration#662
Sahil170595 wants to merge 1 commit into
google:mainfrom
Sahil170595:fix-benchmark-registry-path

Conversation

@Sahil170595

Copy link
Copy Markdown

Problem

Run Benchmarks has been failing on every main commit since 4e8ef56 ("Migrate third_party/tpu_raiden/tpu_raiden/benchmarks to third_party/tpu_raiden/tpu_sync/benchmarks").

That commit moved the benchmarks directory, but the two BAP workflows still pass the pre-migration path as registry_file. The reusable workflow resolves it against the repo checkout root, so the Validate Registry Security job fails before any benchmark is scheduled:

Error: Registry file not found:
/__w/tpu-sync/tpu-sync/user_repo/tpu_raiden/benchmarks/benchmark_registry.pbtxt

Timeline:

Last successful Run Benchmarks on main 675d06f5, 2026-08-10T17:19:34Z
Migration commit 4e8ef56, 2026-08-13
First failing run 4e8ef56d, 2026-08-13T22:59:54Z
Since then every main-branch run has failed

Fix

Update registry_file in both workflows to the migrated paths:

  • .github/workflows/run_benchmarks.ymltpu_sync/benchmarks/benchmark_registry.pbtxt
  • .github/workflows/update_d2h_h2d_record.ymltpu_sync/benchmarks/benchmark_registry_record.pbtxt

Both files exist at those paths on main, and the registries themselves already reference the migrated Bazel targets (//tpu_sync/benchmarks:h2d_d2h_benchmark_gating, //tpu_sync/benchmarks:h2h_cpp_gate), so no change is needed inside them.

Scope / what I could not verify

This clears the Validate Registry Security gate, which is where the run currently dies. I could not verify the downstream benchmark jobs end-to-end — those need the TPU runners (linux-x86-ct5lp-224-8tpu), which I don't expect to be provisioned for a fork PR. If anything further is stale post-migration, it would surface after this gate rather than being hidden behind it.

Note on #622

#622 ("Split benchmark workflows to publish metrics on postsubmit") carries the old tpu_raiden/benchmarks/... path into three new workflow files (presubmit_benchmarks.yml, postsubmit_benchmarks.yml, nightly_benchmarks.yml). Whichever of these lands second will need the same path update — flagging so it isn't reintroduced.

The benchmarks directory moved from tpu_raiden/benchmarks to
tpu_sync/benchmarks in 4e8ef56, but the two BAP workflows still pass the
old path as registry_file. The reusable workflow resolves that path
against the repo checkout root, so it fails before any benchmark runs:

  Error: Registry file not found:
  /__w/tpu-sync/tpu-sync/user_repo/tpu_raiden/benchmarks/benchmark_registry.pbtxt

Run Benchmarks has failed on every main-branch run since 4e8ef56; the
last success was 675d06f on 2026-08-10. Update both workflows to the
current paths. The registry files themselves already reference the
migrated Bazel targets (//tpu_sync/benchmarks:...), so no other change
is needed.
@google-cla

google-cla Bot commented Aug 15, 2026

Copy link
Copy Markdown

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

@Sahil170595

Copy link
Copy Markdown
Author

Superseded by 7dcefc2, which made the same change to both workflows. Closing.

For anyone finding this later: Run Benchmarks was failing on every main-branch commit from 4e8ef56 (the tpu_raiden/benchmarkstpu_sync/benchmarks migration) until 7dcefc2, because the reusable BAP workflow resolves registry_file against the checkout root and the path still pointed at the pre-migration location.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant