You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@lucene.apache.org by "Hoss Man (JIRA)" <ji...@apache.org> on 2019/01/25 03:06:00 UTC
[jira] [Commented] (SOLR-12313)
TestInjection#waitForInSyncWithLeader needs improvement.
[ https://issues.apache.org/jira/browse/SOLR-12313?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16751838#comment-16751838 ]
Hoss Man commented on SOLR-12313:
---------------------------------
Beyond the concerns mark previously raised regarding: documentation of purpose; how slow this can make tests; & wether it's correctly implemented – I've recently noticed two additional big problems:
* this method causes *guaranteed* failures in {{ForceLeaderTest.testReplicasInLIRNoLeader}} on branch_7x when the randomizer enables java {{asserts}} _and_ {{onlyLeaderIndexes}} randomizes to "true" (resulting in the use of tlog replicas)
** Note: forcing {{TestInjection.waitForReplicasInSync = null}} allows this test to sometimes pass, but it's still finicky due to the delay in search results being updated in tlog replicas
** Mark had previously (Nov2018) modified this test class to prevent tlog replicas, but aparently didn't notice on backporting that 7x had this additional test method (deprecated & removed in master) that used the {{onlyLeaderIndexes}} variable directly
* {{TestInjection.waitForInSyncWithLeader()}} does not obey the contract all test injection methods must obey
** notably it (by default) forces commits to tlog replicas to block until in sync with leader in non-test Solr instances if solr was started with JVM assertions
** this is an end user facing performance bottleneck, that can cause real world commits to time out!
Unless there are objections from someone who understands tlog replicas really well & is willing to step up and "fix" {{waitForInSyncWithLeader()}} (including mark's concerns above about how it's designed), I plan to do the following:
# create a sub-task to document & track the user affecting bug regarding TLOG+assertions / commits in CHANGES.txt
# commit a change making {{TestInjection.waitForInSyncWithLeader()}} a No-Op, citing both this issue and the new sub-task
** backporting down to 7x & resolving the sub-task as fixed in 7.7
# commit a change to ForceLeaderTest's initialization of {{onlyLeaderIndexes}} completely preventing the use of tlog replicas as mark intended in his Nov2018 commits
** backporting down to 7x
> TestInjection#waitForInSyncWithLeader needs improvement.
> --------------------------------------------------------
>
> Key: SOLR-12313
> URL: https://issues.apache.org/jira/browse/SOLR-12313
> Project: Solr
> Issue Type: Test
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Mark Miller
> Priority: Major
>
> This really should have some doc for why it would be used.
> I also think it causes BasicDistributedZkTest to take forever for sometimes and perhaps other tests?
> I think checking for uncommitted data is probably a race condition and should be removed.
> Checking index versions should follow the rules that replication does - if the slave is higher than the leader, it's in sync, being equal is not required. If it's expected for a test it should be a specific test that fails. This just introduces massive delays.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@lucene.apache.org
For additional commands, e-mail: dev-help@lucene.apache.org