[
https://issues.apache.org/jira/browse/HDFS-7886?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14353400#comment-14353400
]
Plamen Jeliazkov commented on HDFS-7886:
----------------------------------------
Hi Yi,
I understand the problem a little better now.
Removing triggerBlockReports() doesn't fix the tests entirely; it causes the
AssertionError to happen less frequently but does not otherwise fix the tests.
The problem comes down to this assert error.
Prior to truncate, a replica, under normal operations, never changed state from
FINALIZED to RUR. I understand that it COULD happen, but as far as test cases
go I could not find one that tested such a case nor could I find a
specification of such a case in the design doc in HDFS-265, which Konstantin
referred me to.
What is happening is that in the following code from FsDataSetImpl, the RUR's
original replica points to a FINALIZED replica, because it is a truncate
recovery operation:
{code}
case RUR:
ReplicaUnderRecovery rur = (ReplicaUnderRecovery)b;
uc.get(rur.getVolume().getStorageID()).add(rur.getOriginalReplica());
break;
{code}
We should open up a new issue to discuss the change, but in short, due to
truncate, this assertion is no longer valid because a RUR can now point to a
FINALIZED original replica.
> TestFileTruncate#testTruncateWithDataNodesRestart runs timeout sometimes
> ------------------------------------------------------------------------
>
> Key: HDFS-7886
> URL: https://issues.apache.org/jira/browse/HDFS-7886
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: test
> Affects Versions: 2.7.0
> Reporter: Yi Liu
> Assignee: Plamen Jeliazkov
> Priority: Minor
> Attachments: HDFS-7886.patch
>
>
> https://builds.apache.org/job/PreCommit-HDFS-Build/9730//testReport/
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
