[
https://issues.apache.org/jira/browse/CASSANDRA-19942?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17883598#comment-17883598
]
Dmitry Konstantinov commented on CASSANDRA-19942:
-------------------------------------------------
A possible fix for the issue: https://github.com/apache/cassandra/pull/3559
An alternative option can be to fill buffer with null values not based on count
but fully, using buffer.length - it looks more reliable but more expensive...
[~benedict] if it is possible could you please take a look as a reviewer?
> Pooled BTree.FastBuilder keeps references to data after reset
> -------------------------------------------------------------
>
> Key: CASSANDRA-19942
> URL: https://issues.apache.org/jira/browse/CASSANDRA-19942
> Project: Cassandra
> Issue Type: Bug
> Components: Local/Other
> Reporter: Dmitry Konstantinov
> Assignee: Dmitry Konstantinov
> Priority: Normal
> Fix For: 4.0.x, 4.1.x, 5.0.x, 5.x
>
> Attachments: image-2024-09-22-13-51-02-751.png
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> Hi I am analyzing of argecolumn_test.TestLargeColumn test failure
> (CASSANDRA-19702) and have noticed that sometimes when I run the test locally
> it fails with OOM. The test sends an insert with a large column value.
> In the created heap dump I have found that there is a kind of memory leaking
> for the sent column values though thread-local variable used for re-usable
> Pooled BTree.FastBuilder:
> !image-2024-09-22-13-51-02-751.png!
> In normal cases it is probably not critical because the objects are not that
> huge but it is still a leak.
> reset() logic in BTree.FastBuilder does not clean references to such values
> because during other operations like build() "count" value can be reset to 0.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
