[
https://issues.apache.org/jira/browse/HADOOP-14912?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16183268#comment-16183268
]
Daryn Sharp commented on HADOOP-14912:
--------------------------------------
In practice the impact isn't severe. The next call will dislodge one of the
"stuck" queued calls. Latency is affected to some degree, but is skewed
towards increased latency for lower priority buckets. Across multiple
production clusters, the number of lost permits is usually low single digit.
> FairCallQueue may defer servicing calls
> ---------------------------------------
>
> Key: HADOOP-14912
> URL: https://issues.apache.org/jira/browse/HADOOP-14912
> Project: Hadoop Common
> Issue Type: Bug
> Components: ipc
> Affects Versions: 2.8.2
> Reporter: Daryn Sharp
> Assignee: Daryn Sharp
>
> HADOOP-14033 switched a lock to a semaphore to allow concurrency for
> producers & consumers to the underlying queues. A race condition was created
> that may cause a consumer to acquire a permit but not extract an element,
> leaving the semaphore with fewer permits than queued elements. This causes a
> minimum number of calls to always be present in the call queue.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org
