[CI] Multiple tests failing with "some shards are still open" error

[mark-vieira]

This has failed 4 times already today:

:modules:reindex:test » org.elasticsearch.index.reindex.ReindexFailureTests » classMethod (0.007s)
Some shards are still open after the threadpool terminated. Something is leaking index readers or store references.
java.lang.IllegalStateException: Some shards are still open after the threadpool terminated. Something is leaking index readers or store references.Open stacktrace

Here's are some build scans. Seems to be isolated to 7.x for now:

https://gradle-enterprise.elastic.co/s/gcnoaw5pepki2/tests/5nfchac5yggnq-rjw76l2zc7kn2
https://gradle-enterprise.elastic.co/s/bqddtyap54zai/tests/5nfchac5yggnq-rjw76l2zc7kn2
https://gradle-enterprise.elastic.co/s/crutz4g7pzhay/tests/5nfchac5yggnq-rjw76l2zc7kn2
https://gradle-enterprise.elastic.co/s/zkfh4z2lj67g2/tests/5nfchac5yggnq-rjw76l2zc7kn2

[elasticmachine]

Pinging @elastic/es-distributed (:Distributed/Reindex)

[mark-vieira]

Got another one so I've gone ahead and muted this test with 16afbf9.

https://gradle-enterprise.elastic.co/s/mdkgsffws5mly/tests/failed

[mark-vieira]

Interestingly, I also just saw ShardSizeTermsIT fail with the same error:

https://gradle-enterprise.elastic.co/s/c5ihsym47tw2y/tests/kyv2y2z3r4v7m-tvapgmw3k3fym

[mark-vieira]

Ok, so this seems to be a more generic error that's happening in master and 7.x and looks to have started on February 6th. From what I see so far this has happened in RecoveryWhileUnderLoadIT, ReindexFailureTests, SearchRestCancellationIT, ShardSizeTermsIT and AsyncSearchActionTests. So this is likely some more underlying issue. Here are a number of example build scans:

https://gradle-enterprise.elastic.co/s/llocok2qap4c2/tests/kyv2y2z3r4v7m-n5sx3bdgtt26o
https://gradle-enterprise.elastic.co/s/fbwpfift4mjsw/tests/5nfchac5yggnq-rjw76l2zc7kn2
https://gradle-enterprise.elastic.co/s/b2cthh57ikcbi/tests/xnglhbfnjd7ae-iwi4iupewsj6m
https://gradle-enterprise.elastic.co/s/c5ihsym47tw2y/tests/kyv2y2z3r4v7m-tvapgmw3k3fym
https://gradle-enterprise.elastic.co/s/jsstsgsksw5je/tests/xnglhbfnjd7ae-iwi4iupewsj6m

I'm assuming this falls under @elastic/es-distributed just because I see the word "shards" a lot, but please educate me if this is a wrong assumption.

Given this looks to be a very recent issue, and is happening with some regularity, I think we should give this some priority. Given the number of tests this seems to be cropping up in, muting isn't going to be an effective strategy here.

[dnhatn]

These failures might relate to #46091.

[DaveCTurner]

I suspect #51708, on the grounds that on my machine 1dcf1df passed 140 iterations of ./gradlew :modules:reindex:test --tests org.elasticsearch.index.reindex.ReindexFailureTests.testResponseOnSearchFailure but eb69c6f fails with Some shards are still open about one run in ten. I'm running stress -c 16 -m 8 -d 2 too, not sure if that's helpful to reproduce or not.

@dnhatn would you like to investigate further, or should we ask the search team to take a look?

[dnhatn]

@DaveCTurner Good find. I will take a look. Thank you for looking :).

[dnhatn]

Yes, these failures are indeed caused by #51708. I am working on the fix.

[elasticmachine]

Pinging @elastic/es-search (:Search/Search)

[dnhatn]

I've opened #52099.