[
https://issues.apache.org/jira/browse/QPID-7024?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15126258#comment-15126258
]
Keith Wall commented on QPID-7024:
----------------------------------
Ideally thread pool should sized to be number of peers in the group + 1, where
the additional one exists to cater the task that coordinates discovering the
node. Perhaps the {{RemoteNodeStateLearner}} can take the responsibility to
increase the ThreadPoolExecutor.corePoolSize as nodes are added to the group.
Also I noticed one task (transferMaster) uses the _groupChangeExecutor. This
looked curious at first sight, but on reflection does not seem bad.
Internally, it uses ReplicationGroupAdmin to connect to any node in the group
and request the master transfer. If this were to use the _environmentExecutor
then the request to transfer the master would queue behind the restart of the
node local. This queuing seems unnecessary.
> BDB HA - Group change thread-pool sized to availableProcessors +1
> -----------------------------------------------------------------
>
> Key: QPID-7024
> URL: https://issues.apache.org/jira/browse/QPID-7024
> Project: Qpid
> Issue Type: Bug
> Components: Java Broker
> Reporter: Keith Wall
> Priority: Minor
>
> The BDB HA implementation internally uses a thread pool to process group
> change events. This thread pool is sized to
> {{Runtime.getRuntime().availableProcessors() + 1}}. On a machine with many
> cores, this will create a very large number of threads.
> There is probably no reason why this thread pool needs to be quite so large.
> This problem does not cause a functional impact, but will consume more
> resources (threads + stack) than required.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]