Updating ESSingleNodeTestCase to ensure that all free_context actions have been consumed before tearDown #110595
Conversation
pmpailis
added
>test
|
Pinging @elastic/es-search (Team:Search) |
test/framework/src/main/java/org/elasticsearch/test/ESSingleNodeTestCase.java
Outdated
Show resolved
Hide resolved
test/framework/src/main/java/org/elasticsearch/test/ESSingleNodeTestCase.java
Outdated
Show resolved
Hide resolved
| @@ -130,6 +132,8 @@ public void tearDown() throws Exception { | |||
| logger.trace("[{}#{}]: cleaning up after test", getTestClass().getSimpleName(), getTestName()); | |||
| awaitIndexShardCloseAsyncTasks(); | |||
| ensureNoInitializingShards(); | |||
| ensureAllFreeContextActionsAreConsumed(); | |||
There was a problem hiding this comment.
Why can't we await for the tasks to disappear? We should be able to wait for the tasks to be completed.
Is this causing more tests to fail other than the single test case that this was originally flagged for?
I would think adding something like this to the sub-class tear down:
logger.info("--> waiting for all ongoing tasks to complete within a reasonable time");
safeGet(clusterAdmin().prepareListTasks().setActions(FREE_CONTEXT_ACTION_NAME.name() + "*").setWaitForCompletion(true).execute());
would suffice.
There was a problem hiding this comment.
This seems to suffice indeed! Thanks for the great alternative @benwtrent ! Really wanted to avoid adding some much boilerplate code, but wasn't certain of the available feasible options for the mocked single node case, so I went with this.
Will remove all other changes in favor of this one-line :D
There was a problem hiding this comment.
@pmpailis sorry to harp on this, but does this really need to be in test/framework/src/main/java/org/elasticsearch/test/ESSingleNodeTestCase.java ? Why not only focus on the individual tests that have the issue? Are way more tests than just SearchProgressActionListenerIT failing?
There was a problem hiding this comment.
@pmpailis if no other tests than SearchProgressActionListenerIT are failing because of this, I would much prefer a tearDown() override, waiting there, and then calling super.tearDown()
There was a problem hiding this comment.
Unfortunately there were various tests failing for the same (seemingly) reason:
- [CI] ESSingleNodeTestCase fails during tear down #110409
- [CI] NativePrivilegeStoreCacheTests testGetPrivilegesUsesCache failing #110227 ( and a few more I think)
It seems that this is an issue for the whole suite of ESSingleNodeTestCase tests, hence why I opted for the class tearDown override.
There was a problem hiding this comment.
@pmpailis fabulous. I was not aware. Please make sure that this PR is mentioned in those various tests.
test/framework/src/main/java/org/elasticsearch/test/ESSingleNodeTestCase.java
Outdated
Show resolved
Hide resolved
…deTestCase.java Co-authored-by: Benjamin Trent <ben.w.trent@gmail.com>
… have been consumed before tearDown (elastic#110595)
Relates to #109830
Added a take to ensure that all
free_contextactions have been consumed by thegenericthreadpool before performing any checks on the state of open readers inESSingleNodeTestCase.