[
https://issues.apache.org/jira/browse/CASSANDRA-20550?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Capwell updated CASSANDRA-20550:
--------------------------------------
Description: SAIUtil only checks the status for the indexes on node1, which
knows that node2/node3 have completed the indexing... but doesn't confirm that
node2 knows about node3 or that node3 knows about node2... this causes a flaky
race condition where we start a query on node2, node2 then sees that node3 is
still building the index (it isn't, this is based off gossip), so node2 rejects
the query! (was: In running
org.apache.cassandra.distributed.test.cql3.MultiNodeTableWalkWithoutReadRepairTest
in a repeat job (trying to do 100k seeds without error) I see that from time
to time selects fail in cases like)
> AST fuzz tests can be flakey in multi node clusters due to ephemeral read
> errors, should retry in these cases
> -------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-20550
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20550
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Test/fuzz
> Reporter: David Capwell
> Assignee: David Capwell
> Priority: Normal
> Fix For: 5.x
>
> Time Spent: 2h 10m
> Remaining Estimate: 0h
>
> SAIUtil only checks the status for the indexes on node1, which knows that
> node2/node3 have completed the indexing... but doesn't confirm that node2
> knows about node3 or that node3 knows about node2... this causes a flaky race
> condition where we start a query on node2, node2 then sees that node3 is
> still building the index (it isn't, this is based off gossip), so node2
> rejects the query!
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
