test results are reused when build gets OOM-killed #237
Comments
Interesting, this could be a template instance bug as well. |
We need to update this: ci-operator/pkg/steps/source.go Lines 247 to 251 in 0f8285d
|
I don't think it is that simple. Unless the original build is removed, we are going to be stuck in that loop.
|
Unless the OOM error is transient, probably yes. But OOM does not occur only because we didn't request enough memory. You can be OOM'd if you end up in a busy node with higher priority pods, or there may be an underlying issue in the node which seems to be the current case. EDIT
This is probably wrong, I think you are getting Evicted instead of OOM'd? In any case, |
|
It seems that this case applies for each phase. |
When a pod gets OOM-killed,
/retest
ing stops working and ci-operator reuses the OOM-killed test run's result. See https://openshift-gce-devel.appspot.com/pr/openshift_openshift-azure/984 (1725 to 1727). I had to get into the namespace and manually delete the build in order to restart the test successfully.Pod manifest:
The text was updated successfully, but these errors were encountered: