[
https://issues.apache.org/jira/browse/CASSANDRA-21055?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18042915#comment-18042915
]
Dmitry Konstantinov commented on CASSANDRA-21055:
-------------------------------------------------
[~dcapwell] thank you for finding the root cause, it was actually quite tricky
> When updating a multi cell collection element, if the update is rejected then
> the shared Row.Builder is not freed causing all future mutations to be
> rejected
> -------------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-21055
> URL: https://issues.apache.org/jira/browse/CASSANDRA-21055
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Consistency/Coordination
> Reporter: David Capwell
> Assignee: David Capwell
> Priority: Normal
> Fix For: 6.x
>
> Time Spent: 4h 10m
> Remaining Estimate: 0h
>
> BTreeRow.pooledUnsortedBuilder() is used to lower memory costs for requests,
> and need to call builder.build() to trigger cleanup of resources; but this
> does not happen if the mutation is rejected!
> {code}
> UPDATE tbl SET col[0] = 42 WHERE pk = 0;
> {code}
> The query above is able to hit this when the pk doesn’t exist, it leaks the
> BTreeRow.Builder and all mutations to this thread will now reject as
> org.apache.cassandra.db.rows.BTreeRow.Builder#newRow will detect this
> happened and will trigger a assert.
> This was found in CASSANDRA-20828, but forking off into its own issue for
> better isolation.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
