janhoy opened a new pull request, #903:
URL: https://github.com/apache/solr-operator/pull/903

   Fixes #901
   
   Fixes the three flaky e2e specs reported in #901. Only test code changes.
   
   ### Ingress › Can adapt to changing needs
   
   As described in the issue: the headless service is only deleted after the 
pod update, so line 134 now uses `eventuallyExpectNoService` instead of the 
`Consistently`-based `expectNoService`. The earlier check at the start of the 
spec keeps `expectNoService`, since the headless service must never exist there.
   
   ### Storage › Persistent Data - Expansion › Fully Expands
   
   The CI artifacts show that right after the rolling restart, the other Solr 
node fails to connect to the restarted pod for a minute or more ("Connect timed 
out"). Every such request takes Solr's full 60s connect timeout (`QTime` ≈ 
60000), so the 5s `Within` of `queryCollection` only ever allowed a single 
attempt. The spec now uses a new `queryCollectionAfterRestart` helper that 
allows several attempts within 3 minutes. The likely cause is stale DNS for the 
restarted pod's headless hostname, but the logs don't show the dialled IP, so 
that is not confirmed.
   
   The older failure at line 129 (job 128914789734) predates #895, which 
already fixed it.
   
   ### Backups › Local Directory - Single
   
   Of the three linked jobs, two were not flakes: 129101306235 and 129101299268 
ran on the `readonly-rootfs` and `restricted-pss-defaults` PR branches, where 
the `volume-prep` init container crash-looped with "Backup repository directory 
… is not writable".
   
   The remaining one (132634657662) is a real flake. The previous spec in the 
same namespace (Storage) ended in the same second this spec created its `foo` 
SolrCloud, and `cleanupTest` did not wait for pods to be deleted. The new 
StatefulSet's create of `foo-solrcloud-1` hit `AlreadyExists` on the 
still-terminating old pod. The StatefulSet controller records no event for 
that, retries with exponential backoff, and is not woken when the orphaned pod 
disappears, so the cloud stayed at 1 of 2 pods until the 240s timeout. 
`cleanupTest` now waits up to 3 minutes for all pods in the namespace to be 
gone.
   
   ### Testing
   
   `make e2e-tests TEST_LABELS=pr TEST_PARALLELISM=2 SOLR_IMAGE=solr:9.11.0 
KUBERNETES_VERSION=v1.33.7` passed locally (11/11). One green run does not 
prove intermittent failures are fixed; CI runs will tell.
   
   🤖 Done 100% by Claude Code Opus
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to