[
https://issues.apache.org/jira/browse/CASSANDRA-20577?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
David Capwell updated CASSANDRA-20577:
--------------------------------------
Bug Category: Parent values: Degradation(12984)Level 1 values: Resource
Management(12995)
Complexity: Low Hanging Fruit
Discovered By: Fuzz Test
Fix Version/s: 5.x
Severity: Normal
Status: Open (was: Triage Needed)
> zero copy streaming allocates direct memory that isn't used, but does help to
> fragment the memory space
> -------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-20577
> URL: https://issues.apache.org/jira/browse/CASSANDRA-20577
> Project: Apache Cassandra
> Issue Type: Bug
> Components: Consistency/Streaming
> Reporter: David Capwell
> Assignee: David Capwell
> Priority: Normal
> Fix For: 5.x
>
>
> This is forked from the patch to CASSANDRA-20567 as it isn’t related to SAI,
> so better to isolate.
> When Zero Copy Streaming creates its per-file writers, it allocates direct
> memory per file; but this allocation is never used. This causes issues as
> the direct memory managed by Cassandra gets fragmented, so we run out of
> space and OOM
> {code}
> INFO [node3_Stream-Deserializer-/127.0.0.1:7012-cf906983] 2025-04-21
> 21:55:51,613 ERROR 21:55:51 [Stream 13815658-1dd2-11b2-98f1-3dfc14dd20f3]
> Error while reading sstable from stream for table = ks1.tbl
> java.lang.OutOfMemoryError: Cannot reserve 2097152 bytes of direct buffer
> memory (allocated: 1055235964, limit: 1056309248)
> {code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
