[
https://issues.apache.org/jira/browse/CASSANDRA-16767?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17370160#comment-17370160
]
Scott Carey commented on CASSANDRA-16767:
-----------------------------------------
Pull request here: [https://github.com/apache/cassandra/pull/1085]
I wanted to do this partial garbagecollect without first cancelling all ongoing
compaction on the table, or locking the whole table. But that would require a
bigger overhaul. This re-uses the same logic as the existing garbagecollect
which cancells existing gcs then locks all existing sstables at the start.
The unit test covers the new code and changes, and I also manually tested
'nodetool garbagecollect --user-defined <list of tables>' by replacing the
cassandra jar with my snapshot on an existing cassandra install. As expected,
the SSTables I listed were garbagecollected and new SSTables were created.
> Add user defined SSTables option for nodetool garbagecollect
> ------------------------------------------------------------
>
> Key: CASSANDRA-16767
> URL: https://issues.apache.org/jira/browse/CASSANDRA-16767
> Project: Cassandra
> Issue Type: Improvement
> Reporter: Scott Carey
> Assignee: Scott Carey
> Priority: Normal
>
> nodetool garbagecollect does not yet have an option to supply a user defined
> list of SSTables to process.
> This is unfortunate, because there are many cases where an operator would
> know which subset of tables are in need of a garbagecollect. Perhaps with
> STCS, one would choose to garbagecollect only the largest file.
> With a large LCS table, it is typical that the highest levels have the most
> overwritten data, and it may take a very long time to run a full
> garbagecollect or full compaction, but a relatively short time to process a
> smaller subset of SSTables.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
