[ https://issues.apache.org/jira/browse/CASSANDRA-3179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13102777#comment-13102777 ]
Jonathan Ellis commented on CASSANDRA-3179: ------------------------------------------- The problem is that our whole "zero copy" mmap'd read path is broken by "unmap sstables as soon as they are no longer referenced." Specifically, we release references during reads in CollationController or getRangeSlice as soon as we have generated CF objects from the sstables, but we will continue to reference the buffer contents later on when we send the results back to the user or the coordinator (or, as in the 7031 log, just use the results internally). > JVM segfaults > ------------- > > Key: CASSANDRA-3179 > URL: https://issues.apache.org/jira/browse/CASSANDRA-3179 > Project: Cassandra > Issue Type: Bug > Components: Core > Affects Versions: 1.0.0 > Environment: java version "1.6.0_26" > Java(TM) SE Runtime Environment (build 1.6.0_26-b03) > Java HotSpot(TM) 64-Bit Server VM (build 20.1-b02, mixed mode) > Reporter: Brandon Williams > Assignee: Sylvain Lebresne > Fix For: 1.0.0 > > Attachments: hs_err_pid12074.log, hs_err_pid28971.log, > hs_err_pid7031.log > > > Both with and without compressed OOPs enabled. Seems to mostly happen during > compaction+reads. I'll attach some hs_err files shortly. -- This message is automatically generated by JIRA. For more information on JIRA, see: http://www.atlassian.com/software/jira