[ https://issues.apache.org/jira/browse/CASSANDRA-12358?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15413774#comment-15413774 ]
Aleksey Yeschenko commented on CASSANDRA-12358: ----------------------------------------------- Committed as [4878852fe4aae3516c21fdeafac5c5746a93c31f|https://github.com/apache/cassandra/commit/4878852fe4aae3516c21fdeafac5c5746a93c31f] to trunk, thanks. > Slow PostFlush execution due to 2i flushing can cause near OOM to OOM > --------------------------------------------------------------------- > > Key: CASSANDRA-12358 > URL: https://issues.apache.org/jira/browse/CASSANDRA-12358 > Project: Cassandra > Issue Type: Bug > Components: Core > Reporter: Ariel Weisberg > Assignee: Ariel Weisberg > Fix For: 3.10 > > > 2i can be slow to flush for a variety of reasons. Potentially slower than the > rate at which Memtables can ingest and flush data. If this occurs the heap > fills up with Memtables that are waiting for PostFlush to run. > This occurs because reclaiming the memory is done before PostFlush runs. > I will post a branch that has the reclaim memory task run after PostFlush has > completed. As far as I can tell this is safe and correct since the memory is > committed up until that point. > It's not clear to me if PostFlush has to bind the Memtables or not. I suspect > it does, but I'm not sure if that is a route I should go down. -- This message was sent by Atlassian JIRA (v6.3.4#6332)