[ https://issues.apache.org/jira/browse/HBASE-15973?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15317489#comment-15317489 ]
Enis Soztutar commented on HBASE-15973: --------------------------------------- Instead of looking at time of oldest edit, we should track time of first insertion to the memstore after initialization, or flush start. > PeriodicMemstoreFlusher is causing excessive flushes when back-in-time > inserts are happening > -------------------------------------------------------------------------------------------- > > Key: HBASE-15973 > URL: https://issues.apache.org/jira/browse/HBASE-15973 > Project: HBase > Issue Type: Bug > Reporter: Enis Soztutar > Fix For: 2.0.0, 1.3.0, 1.4.0 > > > In a production cluster, we have noticed a case where flushes were happening > for 200-400KB in sizes. Turns out the periodic memstore flusher is force > flushing because cells with older timestamps (in this case days old) were > being inserted. > We have periodic memstore flusher with 1 hour defaulted, so in a case where > replication is lagging, or phoenix secondary index rebuild or the user doing > back-in-time inserts with cell timestamps older than 1 hour, we will flush > extremely frequently. -- This message was sent by Atlassian JIRA (v6.3.4#6332)