[ https://issues.apache.org/jira/browse/CASSANDRA-3409?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Sylvain Lebresne updated CASSANDRA-3409: ---------------------------------------- Attachment: 3409.patch There really is two problems: * we reload the strategy every time we reload the CFS, i.e, for each upade_column_family, even if it's just changing gc_grace or something. This makes that bug much more problematic. * the only easy way I see make the reload of stategy safe is to grab the compaction lock. Patch attached that fix both (fixing the first problem makes grabing the lock not a big deal imho). > CFS reloading of the compaction strategy is done for every metadata update > and is not thread safe > ------------------------------------------------------------------------------------------------- > > Key: CASSANDRA-3409 > URL: https://issues.apache.org/jira/browse/CASSANDRA-3409 > Project: Cassandra > Issue Type: Bug > Components: Core > Affects Versions: 1.0.0 > Reporter: Sylvain Lebresne > Assignee: Sylvain Lebresne > Fix For: 1.0.1 > > Attachments: 3409.patch > > > The reloading of the compaction strategy done during CFS.reload is not thread > safe. In particular, this is a problem for leveled compactions. It could > leads to some sstable not being added to the manifest and also breaks the > 'only one leveledCompactionTask can run at any given time' assumption (which, > at least without CASSANDRA-3408 can likely leads to blocking compactions > completely). -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators: https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa For more information on JIRA, see: http://www.atlassian.com/software/jira