[
https://issues.apache.org/jira/browse/HDFS-10301?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15411076#comment-15411076
]
Konstantin Shvachko commented on HDFS-10301:
--------------------------------------------
??the patch doesn't appear to close the race.??
It does. The problem is not that we release the lock, but that there is
block-report-related state in different places, particularly the BitSet in
{{DatanodeDescriptor}}, see e.g. [this
comment|https://issues.apache.org/jira/browse/HDFS-10301?focusedCommentId=15321613&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15321613]
under (1). The state can be reset by interleaving reports. So if we don't have
the state there is no race condition, because block reports are independent and
can be processed in any order.
The path does just that it removes the block-report-tracking state. [See
here|https://issues.apache.org/jira/browse/HDFS-10301?focusedCommentId=15259284&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-15259284]
under Approach. In earlier version of the patch Vinitha introduced storage
report as a separate RPC, which was opposed by Colin. The latest patch
incorporates the storage report with the RPC for the last storage report. But
the processing of all reports is still independent, therefore releasing the
lock is not a problem.
Just adding more details to Vinithas response.
??wouldn't something simple like this work???
I don't see how it will work. Not simple. The heartbeats can come at any time
between reports or between storages and update the reportId.
[~daryn], I think removing br-state substantially simplifies report processing
and makes reports independent (or idempotent), which is important by itself and
solves the problem of interleaving reports. The last patch solves the bug you
reported (thanks) and provides a unit test for it. As you see this jira was
under development for quite a while. Would be good to commit it soon. Do you
still stand behind your veto given the latest patch?
> BlockReport retransmissions may lead to storages falsely being declared
> zombie if storage report processing happens out of order
> --------------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-10301
> URL: https://issues.apache.org/jira/browse/HDFS-10301
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 2.6.1
> Reporter: Konstantin Shvachko
> Assignee: Vinitha Reddy Gankidi
> Priority: Critical
> Fix For: 2.7.4
>
> Attachments: HDFS-10301.002.patch, HDFS-10301.003.patch,
> HDFS-10301.004.patch, HDFS-10301.005.patch, HDFS-10301.006.patch,
> HDFS-10301.007.patch, HDFS-10301.008.patch, HDFS-10301.009.patch,
> HDFS-10301.01.patch, HDFS-10301.010.patch, HDFS-10301.011.patch,
> HDFS-10301.012.patch, HDFS-10301.013.patch, HDFS-10301.branch-2.7.patch,
> HDFS-10301.branch-2.patch, HDFS-10301.sample.patch, zombieStorageLogs.rtf
>
>
> When NameNode is busy a DataNode can timeout sending a block report. Then it
> sends the block report again. Then NameNode while process these two reports
> at the same time can interleave processing storages from different reports.
> This screws up the blockReportId field, which makes NameNode think that some
> storages are zombie. Replicas from zombie storages are immediately removed,
> causing missing blocks.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
