Repository navigation

Commit 7986b06

authored

[release/2.13] Bump binary build timeout 280 -> 400 minutes (#188551)

The manywheel CUDA 12.9 build (manywheel-cuda-cu129-build) on the v2.13.0 release candidates is being cancelled by the 280-minute job timeout. It is not hung and not a build error: the build progresses steadily and was at "[707/712] Install the project" on the final wheel when killed, i.e. essentially complete after ~285 minutes. Root cause: the first wheel does a full from-scratch CUDA compile of the whole arch list (7.5;8.0;8.6;9.0;10.0;12.0+PTX), which now includes the Blackwell sm_100/sm_120 arches added for 12.9. That first build takes ~3h (flash_attention CUDA TUs dominate); the remaining ~5 Python wheels reuse the objects at ~18 min each, totaling just over the 280-minute cap. Release binary builds do not use sccache, so the first build is always from scratch. Rather than trim the arch list (Blackwell support is intended for 12.9), give the builds 2 more hours of headroom by raising the shared common.timeout_minutes from 280 to 400. This also bumps the other Linux manywheel builds (cpu/cu126/cu130/cu132) and the macOS-arm64 wheel build, which share the same value; the change only raises a ceiling and has no effect on builds that finish sooner. Windows binary builds use a separate value (timeout_minutes_windows_binary=360) and are unchanged. Test Plan: Regenerated workflows and confirmed the diff is limited to the timeout bump: ``` python3 .github/scripts/generate_ci_workflows.py git diff .github/workflows/ # only "timeout-minutes: 280" -> "400" ``` Authored with assistance from Claude.

1 parent 0bdbc26 commit 7986b06Copy full SHA for 7986b06

3 files changed

.github
- templates
  - common.yml.j2
- workflows
  - generated-linux-binary-manywheel-nightly.yml
  - generated-macos-arm64-binary-wheel-nightly.yml

`‎.github/templates/common.yml.j2‎`

-Original file line number
+Diff line change
 {%- set upload_artifact_action = "actions/upload-artifact@v4.4.0" -%}
 {%- set download_artifact_action = "actions/download-artifact@v4.1.7" -%}
 -{%- set timeout_minutes = 280 -%}
 +{%- set timeout_minutes = 400 -%}
 {%- set timeout_minutes_windows_binary = 360 -%}
 {%- macro concurrency(build_environment) -%}

`‎.github/workflows/generated-linux-binary-manywheel-nightly.yml‎`

Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

`‎.github/workflows/generated-macos-arm64-binary-wheel-nightly.yml‎`

Some generated files are not rendered by default. Learn more about customizing how changed files appear on GitHub.

Comments

(0)