You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-issues@hadoop.apache.org by "Konstantin Shvachko (Jira)" <ji...@apache.org> on 2020/04/21 02:04:00 UTC
[jira] [Updated] (HDFS-15291) [SBN read] Implement
CyclicalBlockingQueue to avoid requing RPC calls on Observer.
[ https://issues.apache.org/jira/browse/HDFS-15291?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Konstantin Shvachko updated HDFS-15291:
---------------------------------------
Description:
RPC queue is currently based on {{LinkedBlockingQueue}}, which is FIFO.
For Observer we delay execution of a call if its lastSeenStateId is larger than the stateId of the Observer. The delay implemented as re-queuing the call to the end of the queue. Re-queue is not atomic. We can avoid moving elements in the queue by replacing {{LinkedBlockingQueue}} with a {{CyclicalBlockingQueue}}. So that instead of re-queuing we just move the head of the queue and the call automatically becomes the last.
was:
RPC queue is currently based on {{LinkedBlockingQueue}}, which is FIFO.
For Observer we delay execution of a call if its lastSeenStateId is larger than the stateId of the Observer. The delay implemented as re-queuing the call to the and of the queue. Re-queue is not atomic. We can avoid moving elements in the queue by replacing {{LinkedBlockingQueue}} with a {{CyclicalBlockingQueue}}. So that instead of re-queuing we just move the head of the queue and the call automatically becomes the last.
> [SBN read] Implement CyclicalBlockingQueue to avoid requing RPC calls on Observer.
> ----------------------------------------------------------------------------------
>
> Key: HDFS-15291
> URL: https://issues.apache.org/jira/browse/HDFS-15291
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: namenode
> Reporter: Konstantin Shvachko
> Priority: Major
>
> RPC queue is currently based on {{LinkedBlockingQueue}}, which is FIFO.
> For Observer we delay execution of a call if its lastSeenStateId is larger than the stateId of the Observer. The delay implemented as re-queuing the call to the end of the queue. Re-queue is not atomic. We can avoid moving elements in the queue by replacing {{LinkedBlockingQueue}} with a {{CyclicalBlockingQueue}}. So that instead of re-queuing we just move the head of the queue and the call automatically becomes the last.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org