[
https://issues.apache.org/jira/browse/HBASE-2729?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12881104#action_12881104
]
HBase Review Board commented on HBASE-2729:
-------------------------------------------
Message from: "Todd Lipcon" <[email protected]>
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
http://review.hbase.org/r/216/
-----------------------------------------------------------
Review request for hbase, stack and Ryan Rawson.
Summary
-------
Fixes bugs where an exception in the middle of flushing a file leaves a
half-written StoreFile in the region dir, preventing that region from
recovering, or, in the case of transient errors, causing silent loss of half a
file's worth of data.
I also got rid of the compaction dir here, and am just using one region-wide
tmp dir. Is there some reason this is a bad idea?
This addresses bug HBASE-2729.
http://issues.apache.org/jira/browse/HBASE-2729
Diffs
-----
src/main/java/org/apache/hadoop/hbase/regionserver/HRegion.java 1794df8
src/main/java/org/apache/hadoop/hbase/regionserver/Store.java 04b7522
src/test/java/org/apache/hadoop/hbase/regionserver/TestCompaction.java
9e5ca46
src/test/java/org/apache/hadoop/hbase/regionserver/TestStore.java a65e947
Diff: http://review.hbase.org/r/216/diff
Testing
-------
Ran TestCompaction and TestStore. Will start a cluster test running before I go
to bed.
Thanks,
Todd
> flushCache should write to a tmp directory and then move into the store
> directory
> ---------------------------------------------------------------------------------
>
> Key: HBASE-2729
> URL: https://issues.apache.org/jira/browse/HBASE-2729
> Project: HBase
> Issue Type: Bug
> Components: regionserver
> Affects Versions: 0.21.0
> Reporter: Todd Lipcon
> Assignee: stack
> Priority: Blocker
>
> Currently it appears that internalFlushCache writes directly to the target
> spot of the flushed data. The finally() block appends the metadata and closes
> the file as if nothing bad went wrong in case of an exception. This is really
> bad, since it means that an IOE in the middle of flushing cache could easily
> write a valid looking file with only half the data, which would then prevent
> us from recovering those edits during log replay.
> Instead, it should flush to a tmp location and move it into the region dir
> only after it's successfully written.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
