Closed ncdc closed 1 year ago
I can reproduce this locally. I've added extra logging to pkg/virtual/apiexport/controllers/apireconciler/apiexport_apireconciler_reconcile.go
to see what's going on. What I'm seeing is that https://github.com/kcp-dev/kcp/blob/61243123ebeee689dc2c71b4bb100fca7a3fbd5c/pkg/virtual/apiexport/controllers/apireconciler/apiexport_apireconciler_reconcile.go#L134
sometimes returns 0 exports, meaning kcp can't find the APIExport for sheriffs
in this case, so it never adds it to discovery. Not sure why yet.
By any chance, is this flake the same as what @jmprusi seemed to have tried to fix in PR https://github.com/kcp-dev/kcp/pull/2417/files#diff-8d6f5573691c6741b1615098ed0f51a30afa2e1e9a906f19d5cbfceaef754930 ?
@davidfestal yes, although that's not really a fix 😄
More data from a local test failure (sharded, 1 shard):
Note: the APIExport informer here is from the cache server
Oh I think when the APIExport that has sheriffs gets updated, we need to queue all other APIExports that have claims against its (sheriffs) identity. Testing this out.
Root cause:
reopening as I ran into this flake today: https://prow.ci.openshift.org/view/gs/origin-ci-test/pr-logs/pull/kcp-dev_kcp/2423/pull-ci-kcp-dev-kcp-main-e2e-sharded/1620860539901054976
Different root cause. Let's open a new one. /close
@ncdc: Closing this issue.
From https://prow.ci.openshift.org/view/gs/origin-ci-test/pr-logs/pull/kcp-dev_kcp/2707/pull-ci-kcp-dev-kcp-main-e2e-sharded/1620426348143054848 for PR #2707