[ https://issues.apache.org/jira/browse/CASSANDRA-2427?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13048838#comment-13048838 ]
Jonathan Ellis commented on CASSANDRA-2427: ------------------------------------------- That's a good stab at it. I'd like to make a couple changes: - the executor mbean should really be only for the executor. Let's create a CommitLogMBean where we expose this and the executor stats. (IExecutorMBean can go away.) - if we only check for flush-needed when we rotate the commitlog, "denormalizing" segment total size is unnecessary -- let's just go through the segments collection and add up length() on demand. Sorry for suggesting otherwise to start with! > Heuristic or hard cap to prevent fragmented commit logs from bringing down > the server > ------------------------------------------------------------------------------------- > > Key: CASSANDRA-2427 > URL: https://issues.apache.org/jira/browse/CASSANDRA-2427 > Project: Cassandra > Issue Type: Improvement > Reporter: Benjamin Coverston > Assignee: Patricio Echague > Labels: commitlog, hardening > Fix For: 1.0 > > Attachments: CASSANDRA-2427-2-trunk-addNewSetting.txt, > CASSANDRA-2427-Jconsole-snapshot.tiff, CASSANDRA-2427-trunk.txt > > > Widely divergent write rates on column families can cause the commit log > segments to fragment. In some cases we have seen the commit log partition > overrun. > One solution here would be to create a heuristic for segment fragmentation to > trigger a flush (commit log segments/memtable) or simply track the free disk > space and force a global flush when the disk gets to 80% capacity. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira