mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
I discovered this bug in trymerge when debugging https://github.com/pytorch/pytorch/pull/129013 in which Dr.CI reported no relevant failures while mergebot complained about some unrelated ROCm failures https://github.com/pytorch/pytorch/pull/129013#issuecomment-2183009217.
It turns out that mergebot took into account stale signals from older runs of the same workflow here. For example,
* https://github.com/pytorch/pytorch/actions/runs/9604985361 was the first run where it had a ROCm failure
* While https://github.com/pytorch/pytorch/actions/runs/9608926565 was the second attempt and it was all green
Notice that both runs came from the same push to commit [be69191](be69191f2d
) with [ciflow/rocm/129013](https://github.com/pytorch/pytorch/tree/ciflow/rocm/129013). So, we just need to check the signals from the newer run.
Note that Dr.CI handles this part correctly using the logic in https://github.com/pytorch/test-infra/blob/main/torchci/pages/api/drci/drci.ts#L1079-L1088. So, the fix in this PR is to bring the same logic to trymerge.
### Testing
`pytest -v test_trymerge.py`
Pull Request resolved: https://github.com/pytorch/pytorch/pull/129291
Approved by: https://github.com/ZainRizvi
664 KiB
664 KiB