[
https://issues.apache.org/jira/browse/HDFS-6424?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14002446#comment-14002446
]
Jing Zhao commented on HDFS-6424:
---------------------------------
I think you're right, [~mingma]. We do not need to put unknown blocks in block
reports, and/or blocks for deleted files, into invalidateBlocks on SBN. The ANN
should be able to handle this, and if NN failover happens, the new ANN can
continue the block deletion work based on the future block reports.
The current patch looks good to me. Maybe we can add some a unit test for this?
Also let's see if other folks have some thoughts on this.
> blockReport doesn't need to invalidate blocks on SBN
> ----------------------------------------------------
>
> Key: HDFS-6424
> URL: https://issues.apache.org/jira/browse/HDFS-6424
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: Ming Ma
> Assignee: Ming Ma
> Attachments: HDFS-6424.patch
>
>
> After the fix in https://issues.apache.org/jira/browse/HDFS-6178,
> blockManager no longer compute pending replication work on SBN. As part of
> that, it also stop removing items from invalidateBlocks.
> blocks can still be added to invalidateBlocks on SBN as part of blockReport.
> As a result, the PendingDeletionBlocks metrics will keep going up on SBN.
> To fix that, we don't need to add blocks to invalidateBlocks during
> blockReport for SBN.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
