[
https://issues.apache.org/jira/browse/HDFS-12084?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16179080#comment-16179080
]
Kihwal Lee commented on HDFS-12084:
-----------------------------------
I was looking at the 2.8 patch.
- Explicit calls to abandonBlock() : unprotectedRemoveBlock() will decrement it.
- Delete before commit : not covered?
- Delete after commit : this jira
- Abandoned. Lease recovery and commitBlockSynchronization() will take care of
it.
- pipeline updates (i.e. abandoned replicas): not covered?
It feels that we need to improve deletion path to take care of both before and
after commit cases. We need to think a bit more about pipeline updates.
> Scheduled Count will not decrement when file is deleted before all IBR's
> received
> ---------------------------------------------------------------------------------
>
> Key: HDFS-12084
> URL: https://issues.apache.org/jira/browse/HDFS-12084
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Reporter: Brahma Reddy Battula
> Assignee: Brahma Reddy Battula
> Attachments: HDFS-12084-001.patch, HDFS-12084-002.patch,
> HDFS-12084-003.patch, HDFS-12084-branch-2.patch
>
>
> When small files creation && deletion happens so frequently and DN's did not
> report blocks to NN before deletion, then scheduled count will keep on
> increment and which will not deleted as blocks are deleted.
> *Note*: Every 20 mins,this can be rolled, but with in 20 mins, count can be
> more as so many operations.
> when batchIBR enabled with committed allowed=1 this will be observed more.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
