[
https://issues.apache.org/jira/browse/HDFS-17121?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17747384#comment-17747384
]
ASF GitHub Bot commented on HDFS-17121:
---------------------------------------
LiuGuH commented on PR #5888:
URL: https://github.com/apache/hadoop/pull/5888#issuecomment-1651228974
> Good improvement. Not think carefully but my first feeling, IBR - FBR
mis-order could trigger some issues, such as miss some block report? One case,
a. generate full block report, b. received one replica named `r` and send IBR,
c. send full block report; And NameNode process FBF and IBR as the above order,
then NameNode will consider that this DataNode does not include replica `r`,
right?
>
> Thanks.
In the NamenodeRpcServer side，the ibr and fbr will both into
BlockManager.BlockReportProcessingThread queue . I think the order of ibr and
fbr cannot guarantee in the previous hadoop version.
blockreport will be processed with one after one. But the ibrs order is not
guaranteed.
> BPServiceActor to provide new thread to handle FBR
> --------------------------------------------------
>
> Key: HDFS-17121
> URL: https://issues.apache.org/jira/browse/HDFS-17121
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: hdfs
> Reporter: liuguanghua
> Priority: Minor
> Labels: pull-request-available
>
> After HDFS-16016 , it makes ibr in a thread to avoid heartbeat blocking with
> ibr when require readlock in Datanode.
> Now fbr should do as this. The reason is this:
> (1)heartbeat maybe block because of fbr with readlock in datanode
> (2)fbr will only may return FinalizeCommand
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
