-
Story
-
Resolution: Done
-
Undefined
-
None
-
None
-
None
-
Product / Portfolio Work
-
False
-
-
False
-
None
-
None
-
None
-
None
-
None
Again related to OCPBUGS-50510, a change went into the api that makes the apiserver more correctly reject requests with etcdserver timeout rather than allowing clients to retry, because the old approach was unsafe. This triggered regressions as we now see occasional failures that would have succeeded before. The patch must stay in.
This is impacting the pass rate of some pod sandbox tests. In TRT-2061 however we may slow down monitortest init, which seems to be when these usually happen.
We'll need to juggle carefully with regards to our ability to see if the problem went away. This perhaps should not merge until the slowdown happens and we have time to see it working. Then we can breakout the test which will give us a clear insight into how often this precise problem is happening in CI. The test will then be held to a 95% pass rate in comp readiness as a new test.
Should just be a new regex here: https://github.com/openshift/origin/blob/0253b2bc9f8e306da7a04477474cc03d56562760/pkg/monitortests/network/legacynetworkmonitortests/networking.go#L53
- is related to
-
OCPBUGS-50510 etcd timeouts causing failed pod sandbox creation writing network status
-
- Closed
-