[
https://issues.apache.org/jira/browse/HBASE-9397?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13755450#comment-13755450
]
Matteo Bertozzi commented on HBASE-9397:
----------------------------------------
looks good, with a fix to
{code}
- if (isTakingSnapshot(snapshotTable)) {
- SnapshotSentinel handler = this.snapshotHandlers.get(snapshotTable);
+ if (isTakingSnapshot(snapshot)) {
+ SnapshotSentinel handler = restoreHandlers.get(snapshotTable);
{code}
instead of restoreHandlers.get() should be snapshotHandlers.get() I think you
copy pasted from the code below that is checking if restore is in progress.
{code}
+ synchronized boolean isTakingSnapshot(final SnapshotDescription snapshot) {
...
+ while (it.hasNext()) {
...
+ if (snapshot.getName().equals(sentinel.getSnapshot().getName()) &&
!sentinel.isFinished())
...
+ return isTakingSnapshot(snapshotTable);
{code}
nit(I'm good even without this change): I think the table should be checked
first, I assume is more likely to have a job taking a snapshot on the same
table (e.g. snapshot-table-000N) than having two jobs taking a snapshot with
the same name (e.g. snapshot-000N)
> Snapshots with the same name are allowed to proceed concurrently
> ----------------------------------------------------------------
>
> Key: HBASE-9397
> URL: https://issues.apache.org/jira/browse/HBASE-9397
> Project: HBase
> Issue Type: Bug
> Components: snapshots
> Affects Versions: 0.95.2, 0.94.11
> Reporter: Jerry He
> Assignee: Jerry He
> Fix For: 0.94.12, 0.96.0
>
> Attachments: HBASE-9397-0.94.patch, HBASE-9397-trunk.patch
>
>
> Snapshots with the same name (but on different tables) are allowed to proceed
> concurrently.
> This seems to be loop hole created by allowing multiple snapshots (on
> different tables) to run concurrently.
> There are two checks in SnapshotManager, but fail to catch this particular
> case.
> In isSnapshotCompleted(), we only check the completed snapshot directory.
> In isTakingSnapshot(), we only check for the same table name.
> The end result is the concurrently running snapshots with the same name are
> overlapping and messing up each other. For example, cleaning up the other's
> snapshot working directory in .hbase-snapshot/.tmp/snapshot-name.
> {code}
> 2013-08-29 18:25:13,443 ERROR
> org.apache.hadoop.hbase.master.snapshot.TakeSnapshotHandler: Failed taking
> snapshot { ss=mysnapshot table=TestTable type=FLUSH } due to
> exception:Couldn't read snapshot info
> from:hdfs://hdtest009:9000/hbase/.hbase-snapshot/.tmp/mysnapshot/.snapshotinfo
> org.apache.hadoop.hbase.snapshot.CorruptedSnapshotException: Couldn't read
> snapshot info
> from:hdfs://hdtest009:9000/hbase/.hbase-snapshot/.tmp/mysnapshot/.snapshotinfo
> at
> org.apache.hadoop.hbase.snapshot.SnapshotDescriptionUtils.readSnapshotInfo(SnapshotDescriptionUtils.java:321)
> at
> org.apache.hadoop.hbase.master.snapshot.MasterSnapshotVerifier.verifySnapshotDescription(MasterSnapshotVerifier.java:123)
> {code}
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
