[
https://issues.apache.org/jira/browse/CASSANDRA-14349?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16420955#comment-16420955
]
Jay Zhuang commented on CASSANDRA-14349:
----------------------------------------
Hi [~shichao.an], nice finding. It would be great to have a dTest for that,
just restart the node a few time and check if there's any orphaned commitlog in
{{cdc_raw}} directory. Any non-active commitlog that doesn't have idx file
should be considered orphaned.
cc. [~JoshuaMcKenzie]
> Untracked CDC segment files are not deleted after replay
> --------------------------------------------------------
>
> Key: CASSANDRA-14349
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14349
> Project: Cassandra
> Issue Type: Bug
> Components: Local Write-Read Paths
> Reporter: Shichao An
> Assignee: Shichao An
> Priority: Minor
>
> When CDC is enabled, a hard link to each commit log file will be created in
> cdc_raw directory. Those commit logs with CDC mutations will also have cdc
> index files created along with the hard links; these are intended for the
> consumer to handle and clean them up.
> However, if we don't produce any CDC traffic, those hard links in cdc_raw
> will be never cleaned up (because hard links will still be created, without
> the index files), whereas the real original commit logs are correctly deleted
> after replay during process startup. This will results in many untracked hard
> links in cdc_raw if we restart the cassandra process many times. I am able to
> use CCM to reproduce it in trunk version which has the CASSANDRA-12148
> changes.
> This seems a bug in handleReplayedSegment of the commit log segment manager
> which neglects to take care of CDC commit logs. I will attach a patch here.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
