[jboss-jira] [JBoss JIRA] (WFLY-6321) Create tool to monitor clustering thread pool usage

[Richard Achmatowicz (JIRA)]

    [ https://issues.jboss.org/browse/WFLY-6321?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13172238#comment-13172238 ] 

Richard Achmatowicz commented on WFLY-6321:
-------------------------------------------

The thread pools of interest in JGroups:
- created in org.jgroups.protocols.TP
- thread pools:
-- regular
-- OOB
-- internal

The thread pools of interest in Infinispan:
- majority created in org.infinispan.factories.NamedExecutorsFactory
- thread pools:
-- transport
-- remote
-- notification
-- persistence
-- expiration
-- replicationQueue
-- stateTransferExecutor
-- totalOrderExecutor
-- async
-- timeout

There are two approaches to monitoring thread pools used here:
- dump thread pool statistics every X calls to the Executor
- dump thread pool statistics every X seconds 

For each thread pool, we present these statistics:
- pool: the number of threads currently available in the pool
- activePool: the number of active threads (i.e. threads which are executing some task)
- queuedTasks: the number of queued tasks waiting for an available thread
- completedTasks: the number of tasks completed by some thread in the thread pool

Rejected execution exceptions are also monitored; these occur when the thread pool is "overwhelmed".


> Create tool to monitor clustering thread pool usage
> ---------------------------------------------------
>
>                 Key: WFLY-6321
>                 URL: https://issues.jboss.org/browse/WFLY-6321
>             Project: WildFly
>          Issue Type: Task
>          Components: Clustering
>            Reporter: Richard Achmatowicz
>            Assignee: Richard Achmatowicz
>            Priority: Minor
>
> Scheduled executors and thread pools are used widely in JGroups and Infinispan for asynchronously executing tasks.  When thread pools are not adequately sized to the load they are subjected to, this can lead to performance problems. 
> It would be helpful if we could see thread pool usage as a function of elapse time during performance runs, in order to diagnose potential thread pool issues.
> This task will provide a Byteman-based tool to monitor threadpool usage and allow a report to be attached to SmartFrog test jobs.



--
This message was sent by Atlassian JIRA
(v6.4.11#64026)