[
https://issues.apache.org/jira/browse/HDFS-13846?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16593177#comment-16593177
]
Xiao Chen commented on HDFS-13846:
----------------------------------
Thanks Kitti for finding and fixing this, and Zsolt for reviewing!
This is a good find, and the patch looks pretty good overall. A few test
comment:
- We can take out the magic number (dataBlockNum) 2 here, make it a constant
and calculated from {{liveReplicas}}. {code}
when(blockInfo.getRealDataBlockNum()).thenReturn((short)2);
{code}
- The assertion part of {{testIncrementAndDecrementSafeBlockCount}} and
{{testIncrementAndDecrementStripedSafeBlockCount}} can be refactored into a
shared method.
> Safe blocks counter is not decremented correctly if the block is striped
> ------------------------------------------------------------------------
>
> Key: HDFS-13846
> URL: https://issues.apache.org/jira/browse/HDFS-13846
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs
> Affects Versions: 3.1.0
> Reporter: Kitti Nanasi
> Assignee: Kitti Nanasi
> Priority: Major
> Attachments: HDFS-13846.001.patch
>
>
> In BlockManagerSafeMode class, the "safe blocks" counter is incremented if
> the number of nodes containing the block equals to the number of data units
> specified by the erasure coding policy, which looks like this in the code:
> {code:java}
> final int safe = storedBlock.isStriped() ?
> ((BlockInfoStriped)storedBlock).getRealDataBlockNum() :
> safeReplication;
> if (storageNum == safe) {
> this.blockSafe++;
> {code}
> But when it is decremented the code does not check if the block is striped or
> not, just compares the number of nodes containing the block with 0
> (safeReplication - 1) if the block is complete, which is not correct.
> {code:java}
> if (storedBlock.isComplete() &&
> blockManager.countNodes(b).liveReplicas() == safeReplication - 1) {
> this.blockSafe--;
> assert blockSafe >= 0;
> checkSafeMode();
> }
> {code}
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
