[jira] [Commented] (CASSANDRA-12969) Index: index can significantly slow down boot
[ https://issues.apache.org/jira/browse/CASSANDRA-12969?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15721571#comment-15721571 ] Corentin Chary commented on CASSANDRA-12969: Hello, did it go well ? Another sensible optimization will be CASSANDRA-12962 > Index: index can significantly slow down boot > - > > Key: CASSANDRA-12969 > URL: https://issues.apache.org/jira/browse/CASSANDRA-12969 > Project: Cassandra > Issue Type: Improvement > Components: CQL >Reporter: Corentin Chary > Fix For: 3.x > > Attachments: 0004-index-do-not-re-insert-values-in-IndexInfo.patch > > > During startup, each existing index is opened and marked as built by adding > an entry in "IndexInfo" and forcing a flush. Because of that we end up > flushing one sstable per index. On systems on HDD this can take minutes for > nothing. > Thw following patch allows to avoid creating useless new sstables if the > index was already marked as built and will greatly reduce the startup time > (and improve availability during restarts). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
[jira] [Commented] (CASSANDRA-12969) Index: index can significantly slow down boot
[ https://issues.apache.org/jira/browse/CASSANDRA-12969?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=15708470#comment-15708470 ] Sam Tunnicliffe commented on CASSANDRA-12969: - Thanks, that is a sensible optimisation. I've pushed 3.X & trunk branches with the patch for CI. Assuming the test results look OK (which I'm sure they will), I'll commit once those are done. > Index: index can significantly slow down boot > - > > Key: CASSANDRA-12969 > URL: https://issues.apache.org/jira/browse/CASSANDRA-12969 > Project: Cassandra > Issue Type: Improvement > Components: CQL >Reporter: Corentin Chary > Fix For: 3.x > > Attachments: 0004-index-do-not-re-insert-values-in-IndexInfo.patch > > > During startup, each existing index is opened and marked as built by adding > an entry in "IndexInfo" and forcing a flush. Because of that we end up > flushing one sstable per index. On systems on HDD this can take minutes for > nothing. > Thw following patch allows to avoid creating useless new sstables if the > index was already marked as built and will greatly reduce the startup time > (and improve availability during restarts). -- This message was sent by Atlassian JIRA (v6.3.4#6332)
