[ 
https://issues.apache.org/jira/browse/HADOOP-1903?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

stack updated HADOOP-1903:
--------------------------

    Attachment: 1903.patch

If flush is interrupted, force a restart of the HRS so hlog is replayed -- 
currently this only happens on restarts -- so memcache is refilled with what 
was in the aborted/dropped snapshot.  Otherwise, opportunity for HRS to 
misrepresent content of its stores.

HADOOP-1903 Possible data loss if Exception happens between snapshot and flush
to disk.

M  src/contrib/hbase/src/java/org/apache/hadoop/hbase/HRegionServer.java
    javadoc additions.
    (Flusher.chore): Restart HRS if we get a DroppedSnapshotException.
M src/contrib/hbase/src/java/org/apache/hadoop/hbase/HLog.java
    (abort): Renamed abortCacheFlush
    (cleanup): Added.
M  src/contrib/hbase/src/java/org/apache/hadoop/hbase/HMaster.java
    Add TODO to comment.
M  src/contrib/hbase/src/java/org/apache/hadoop/hbase/HMemcache.java
    Add comment on state.
M  src/contrib/hbase/src/java/org/apache/hadoop/hbase/HRegion.java
    Add throws javadoc to a couple of methods. Add comments to
    catch of IOException in cache flush.

> [hbase] Possible data loss if Exception happens between snapshot and flush to 
> disk.
> -----------------------------------------------------------------------------------
>
>                 Key: HADOOP-1903
>                 URL: https://issues.apache.org/jira/browse/HADOOP-1903
>             Project: Hadoop
>          Issue Type: Bug
>            Reporter: stack
>            Assignee: stack
>            Priority: Minor
>             Fix For: 0.15.0
>
>         Attachments: 1903.patch
>
>
> There exists a little window during which we can lose data.  During a 
> memcache flush, we make an inmemory copy, a 'snapshot'.  The memcache is then 
> zeroed and off we go again taking updates.  Meantime, in background we are 
> supposed to flush the snapshot to disk.  If this process is interrupted -- 
> e.g. the HDFS is yanked from under us or if an OOME occurs in this thread -- 
> then the content of the snapshot is lost.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.

Reply via email to