[
https://issues.apache.org/jira/browse/RATIS-2434?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wei-Chiu Chuang reassigned RATIS-2434:
--------------------------------------
Assignee: Wei-Chiu Chuang
> Regression test for data race between SegmentedRaftLog.get() and
> LogSegment.append()
> ------------------------------------------------------------------------------------
>
> Key: RATIS-2434
> URL: https://issues.apache.org/jira/browse/RATIS-2434
> Project: Ratis
> Issue Type: Test
> Reporter: Wei-Chiu Chuang
> Assignee: Wei-Chiu Chuang
> Priority: Major
>
> RATIS-2129 has significant performance improvement due to the removal of
> RaftLog lock. There was a race condition bug between the order of entry cache
> and appending the entry to log segment record, which is fixed by RATIS-2282
>
> That is to say,
> [https://github.com/apache/ratis/blob/b73cf97c6f6349abb2e37a801f214629cb5245b3/ratis-server/src/main/java/org/apache/ratis/server/raftlog/segmented/LogSegment.java#L374-L376]
> {code:java}
> putEntryCache(record.getTermIndex(), entry, op);
> }
> records.append(record); {code}
> and
> [https://github.com/apache/ratis/blob/master/ratis-server/src/main/java/org/apache/ratis/server/raftlog/segmented/SegmentedRaftLog.java#L289-L293]
> {code:java}
> record = segment.getLogRecord(index);
> if (record == null) {
> return null;
> }
> final LogEntryProto entry =
> segment.getEntryFromCache(record.getTermIndex());
> {code}
> are not atomic. It is only correct if the append() puts into entry cache
> before appending to records.
>
> We should have a regression test to ensure no that order is not changed. (Add
> javadoc comments too)
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
