cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Paulo Motta (JIRA)" <>
Subject [jira] [Commented] (CASSANDRA-11739) Cache key references might cause OOM on incremental repair
Date Tue, 17 May 2016 21:38:12 GMT


Paulo Motta commented on CASSANDRA-11739:

Attaching patch that does not keep {{SSTableReader}} references for the duration of repair,
but sstable filenames instead, allowing obsoleted {{SSTableReaders}} to be garbage collected.
At anticompaction time, we iterate over all readers and reference only those that were added
to the parent repair session and are still active. Added unit test on {{ActiveRepairServiceTest}}
to verify that active repaired {{SSTableReaders}} are being returned.

this will help keep inc. repair memory usage low before CASSANDRA-8858 (which should be added
to 3.0+).


commit info: merge conflicts on 2.2 and 3.0. 3.0 merges cleanly upwards.

> Cache key references might cause OOM on incremental repair
> ----------------------------------------------------------
>                 Key: CASSANDRA-11739
>                 URL:
>             Project: Cassandra
>          Issue Type: Bug
>            Reporter: Paulo Motta
>            Assignee: Paulo Motta
>         Attachments: heapdump.png
> We keep {{SSTableReader}} references for the duration of the repair to anti-compact later,
and their tidier keep references to cache keys to be invalidated which are only cleaned up
by GC after repair is finished. These cache keys can accumulate while repair is being executed
leading to OOM for large tables/keyspaces.
> Heap dump attached.

This message was sent by Atlassian JIRA

View raw message