[ https://issues.apache.org/jira/browse/CASSANDRA-6905?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14072090#comment-14072090 ]
Ala' Alkhaldi commented on CASSANDRA-6905: ------------------------------------------ Reading through the CommitLogReplayer.recover(file), I believe it replies all the mutations until the end of the file. I.e. It goes beyond the specified time stamp and filter each mutation in a separate thread. To confirm my theory, I add a test to the RecoveryManagerTest (see the file attached) and it worked successfully. WDYT [~brandon.williams] > commitlog archive replay should attempt to replay all mutations > --------------------------------------------------------------- > > Key: CASSANDRA-6905 > URL: https://issues.apache.org/jira/browse/CASSANDRA-6905 > Project: Cassandra > Issue Type: Improvement > Components: Core > Reporter: Brandon Williams > Assignee: Ala' Alkhaldi > Priority: Minor > Fix For: 2.0.10 > > Attachments: CASSANDRA-6905-test.txt > > > Currently when you do a point-in-time recovery using archived commitlogs, the > replay stops when the time is encountered, but since timestamps are supplied > by the client we can't guarantee the segment is ordered by timestamp, so some > mutations can be lost. Instead we could continue past the given timestamp, > and just filter out any mutations greater than it. -- This message was sent by Atlassian JIRA (v6.2#6252)