[
https://issues.apache.org/jira/browse/TEZ-4213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17172619#comment-17172619
]
Ashutosh Chauhan commented on TEZ-4213:
---------------------------------------
bq. class LimitedSizeQueue<E> extends LinkedBlockingQueue<E> {
can
https://docs.oracle.com/javase/8/docs/api/java/util/concurrent/ArrayBlockingQueue.html
be used instead here?
bq. int queue_size = thread_count << 1;
could be int queue_size = thread_count << 2; since adding to queue would be
much faster, having larger queue is better.
bq. rawExecutor = new ThreadPoolExecutor(thread_count, thread_count,
lets use config for max thread count. For min we can use : min(4, thread_count)
since better to start with lower thread count, in case they are not needed.
> Bound appContext executor capacity using a configurable property
> ----------------------------------------------------------------
>
> Key: TEZ-4213
> URL: https://issues.apache.org/jira/browse/TEZ-4213
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Panagiotis Garefalakis
> Assignee: Panagiotis Garefalakis
> Priority: Major
> Attachments: TEZ-4213.01.patch, TEZ-4213.02.patch, TEZ-4213.03.patch,
> TEZ-4213.04.patch, TEZ-4213.05.patch
>
> Time Spent: 1h 50m
> Remaining Estimate: 0h
>
> After TEZ-4170 was merged, appContext executor pool is also used by the
> RootInputInitializerManager to speed up SplitGeneration.
> However, this executor pool currently has not capacity limit
> https://github.com/apache/tez/blob/master/tez-dag/src/main/java/org/apache/tez/dag/app/DAGAppMaster.java#L624
> The problem the occurs when generating splits for larger inputs (thousands or
> more) is that it can could result to
> {color:red}java.lang.OutOfMemoryError{color}
> that is also reproducible with a test case.
> https://github.com/apache/tez/blob/master/tez-dag/src/main/java/org/apache/tez/dag/app/dag/RootInputInitializerManager.java#L130
> To avoid such errors, I propose to limit the capacity of this pool to a
> configurable value that can be for example the number of physical cores by
> default.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
