[
https://issues.apache.org/jira/browse/HDFS-15730?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17252600#comment-17252600
]
Jinglun commented on HDFS-15730:
--------------------------------
Hi [~ayushtkn], thanks your comments !
{quote}In case we delete the excess index say b0 and we loose the only
remaining b0 as well, we might loose a chance of reconstruction as well?
{quote}
It won't happen. The method BlockManager#processExtraRedundancyBlock()
construct an array nonExcess. We can see the corrupt nodes are excluded from
the array.
{code:java}
if (!isExcess(cur, block)) {
if (cur.isInService()) {
// exclude corrupt replicas
if (corruptNodes == null || !corruptNodes.contains(cur)) { // Corrupt
storage is excluded here !
nonExcess.add(storage);
}
}
}
{code}
Then it calls BlockManager#chooseExcessRedundancies(). All the excess storage
are chosen from the array nonExcess. Supposing we have 2 b0. One is corrupt.
The corrupt one will be excluded from nonExcess, so the left b0 won't be excess
anymore and won't be deleted in BlockManager#chooseExcessRedundancies().
> Erasure Coding: Fix unit test bug of
> TestAddOverReplicatedStripedBlocks.testProcessOverReplicatedAndCorruptStripedBlock.
> ------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-15730
> URL: https://issues.apache.org/jira/browse/HDFS-15730
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Jinglun
> Assignee: Jinglun
> Priority: Minor
> Attachments: HDFS-15730.001.patch
>
>
> I'm working on ec replication and find a bug of the test case:
> TestAddOverReplicatedStripedBlocks#testProcessOverReplicatedAndCorruptStripedBlock.
> The test case added 2 redundant block then check the block indices. It wrote
> 'the redundant internal blocks will not be deleted before the corrupted block
> gets reconstructed.'
> But actually the redundant block could be deleted when there is corrupted
> block. The reason the test could pass is it runs very fast and checks the
> block indices before the redundant block is deleted and reported to the
> NameNode.
> The patch is both a fix and an explanation of the bug.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
