[
https://issues.apache.org/jira/browse/HDFS-15099?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17010284#comment-17010284
]
Konstantin Shvachko edited comment on HDFS-15099 at 1/8/20 2:41 AM:
--------------------------------------------------------------------
Currently {{getBlockLocations()}} trying to update aTime on Observer only logs
a warning {{StandbyException - "Operation category WRITE is not supported in
state observer"}}, but doesn't redirect to Active.
We see when the read load is high this correlates with increase in RPC queue
time on Observer, which slows down the cluster in extreme cases.
was (Author: shv):
Currently {{getBlockLocations()}} on Observer only logs a warning
{{StandbyException - "Operation category WRITE is not supported in state
observer"}}, but doesn't redirect to Active.
We see when the read load is high this correlates with increase in RPC queue
time on Observer, which slows down the cluster in extreme cases.
> [SBN Read] getBlockLocations() should throw ObserverRetryOnActiveException on
> an attempt to change aTime on ObserverNode
> ------------------------------------------------------------------------------------------------------------------------
>
> Key: HDFS-15099
> URL: https://issues.apache.org/jira/browse/HDFS-15099
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: namenode
> Affects Versions: 2.10.0
> Reporter: Konstantin Shvachko
> Priority: Major
>
> The precision of updating an INode's aTime while executing
> {{getBlockLocations()}} is 1 hour by default. Updates cannot be handled by
> ObserverNode, so the call should be redirected to Active NameNode. In order
> to redirect to active the ObserverNode should through
> {{ObserverRetryOnActiveException}}.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
