[
https://issues.apache.org/jira/browse/HDFS-16380?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
zy.jordan updated HDFS-16380:
-----------------------------
Description:
As HDFS-13522 introduced, user can access Observer NameNode by RBF. But it will
result in Extra requeue on Observer side, because RBF maintain an global
alignContext for each namespace. When a call go ANN, it update stateId in
alignContext instead of msync time arrived, and it will requeue on ONN side. In
some scenes (eg, Adhoc query), it is unnecessary.
Separate alignContext for different scenes, for example
AlignContext-A(msync:5s) for adhoc, AlignContext-B(msync:0s) for ETL. Maybe it
can reduce requeue time on ONN.
was:[HDFS-13522|https://issues.apache.org/jira/browse/HDFS-13522]
> Result in unnecessary requeue on Observer NameNode when access by RBF in some
> scenes
> ------------------------------------------------------------------------------------
>
> Key: HDFS-16380
> URL: https://issues.apache.org/jira/browse/HDFS-16380
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: rbf
> Affects Versions: 3.3.1
> Reporter: zy.jordan
> Priority: Major
> Fix For: 3.3.1
>
> Original Estimate: 40h
> Remaining Estimate: 40h
>
> As HDFS-13522 introduced, user can access Observer NameNode by RBF. But it
> will result in Extra requeue on Observer side, because RBF maintain an global
> alignContext for each namespace. When a call go ANN, it update stateId in
> alignContext instead of msync time arrived, and it will requeue on ONN side.
> In some scenes (eg, Adhoc query), it is unnecessary.
> Separate alignContext for different scenes, for example
> AlignContext-A(msync:5s) for adhoc, AlignContext-B(msync:0s) for ETL. Maybe
> it can reduce requeue time on ONN.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
