Repository navigation
Conversation
Signed-off-by: Ma, Guokai <guokai.ma@intel.com>
Signed-off-by: Ma, Guokai <guokai.ma@intel.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 52d7f4d793
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
|
|
||
|
|
||
| def _driver_version() -> str: | ||
| result = subprocess.run(["nvidia-smi", "--query-gpu=driver_version,name", "--format=csv,noheader"], |
There was a problem hiding this comment.
When this file is collected by the required CPU workflow (.github/workflows/cpu-torch-latest.yml:252-257 runs all of unit/), nvidia-smi is not installed, but this call executes unconditionally and raises FileNotFoundError; the CUDA availability guard in the second test never protects this test. Gate the probe on an NVIDIA environment or move it into the NVIDIA-only sandbox controller.
Useful? React with 👍 / 👎.
| def test_report_host_driver_version(): | ||
| print(f"DS_CI_DRIVER_PROBE: {_driver_version()}") |
There was a problem hiding this comment.
In the Modal path, ci/torch_latest.py:739-748 invokes pytest with --verbose but not -s; pytest --help documents -s as the shortcut for --capture=no and says captured output is shown for failed tests. Because this test normally passes, the printed driver version is suppressed and therefore does not become visible in the CI log as intended. Emit it as an unconditional controller command before pytest or explicitly disable capture for the probe.
Useful? React with 👍 / 👎.
| def test_report_host_driver_version(): | ||
| print(f"DS_CI_DRIVER_PROBE: {_driver_version()}") |
There was a problem hiding this comment.
After this commit is merged, normal merge-queue runs use subset selection and execute exactly the impacted test files (.github/workflows/TEST_SELECTION.md:176-188). The selector only includes this standalone test when the test itself changes (ci/tests_fetcher.py:543-567), so unrelated future changes—the runs for which a fleet driver regression must be diagnosed—will not execute the probe at all. Put the report in the sandbox controller's unconditional pre-pytest commands rather than relying on test selection.
Useful? React with 👍 / 👎.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
1 participant
Diagnostic probe for #8810 / merge-queue hang wave: prints host driver version (nvidia-smi) inside the Modal sandbox so infra regressions (suspected fleet driver update to 610.57.04) are visible in CI logs. Also adds a small CUDA context health test.