[
https://issues.apache.org/jira/browse/SOLR-11685?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Varun Thacker updated SOLR-11685:
---------------------------------
Attachment: solr_master_7574.log
solr_master_8983.log
Attaching solr_master*.log which was from some manual testing.
I did some manual testing where I started the cloud example which creates a 2X2
collection
Then I had a script running which indexes constantly . I then simulated a GC
pause on node2 ( 7574 ) by running {{kill -STOP <node2_pid> && sleep 60 && kill
-CONT <node2_pid>}} .
We hit two types of race conditions here:
- node2 thinks it's the leader and rejects a request from node1 ( the actual
leader at the time )
- node1 forwards a request when it was a leader . node2 becomes the leader
before the update reaches and rejects the request.
This is very easy to reproduce when you index constantly and then force a
zookeeper connection timeout on the node which has the leader.
> CollectionsAPIDistributedZkTest.testCollectionsAPI fails regularly with
> leader mismatch
> ---------------------------------------------------------------------------------------
>
> Key: SOLR-11685
> URL: https://issues.apache.org/jira/browse/SOLR-11685
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Reporter: Varun Thacker
> Attachments: jenkins_7x_257.log, jenkins_master_7045.log,
> solr_master_7574.log, solr_master_8983.log
>
>
> I've been noticing lots of failures on Jenkins where the document add get's
> rejected because of leader conflict and throws an error like
> {code}
> ClusterState says we are the leader
> (https://127.0.0.1:38715/solr/awhollynewcollection_0_shard2_replica_n2), but
> locally we don't think so. Request came from null
> {code}
> Scanning Jenkins logs I see that these failures have increased since Sept
> 28th and has been failing daily.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
