[
https://issues.apache.org/jira/browse/MAPREDUCE-6473?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15092785#comment-15092785
]
Eric Payne commented on MAPREDUCE-6473:
---------------------------------------
[~kshukla], Thanks for documenting this problem and providing the patch for it.
It looks like the following could happen in {{Cluster.java}}:
- thread 1 enters {{initProviderList}} and begins to process the entries in
{{frameworkLoader}}
- thread 2 enters {{initProviderList}}, sees that {{providerList}} is null, and
exits. Then, in {{initialize}}, no processing for {{providerList}} would happen
and it would throw an IOE when {{clientProtocolProvider}} is null. Am I missing
something?
> Job submission can take a long time during Cluster initialization
> -----------------------------------------------------------------
>
> Key: MAPREDUCE-6473
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-6473
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Components: performance
> Reporter: Kuhu Shukla
> Assignee: Kuhu Shukla
> Attachments: 99%ile.png, MAPREDUCE-6473-PerfTest.txt,
> MAPREDUCE-6473-v1.patch, MAPREDUCE-6473-v2.patch, MAPREDUCE-6473-v3.patch,
> MAPREDUCE-6473-v4.patch, MAPREDUCE-6473-v5.patch, MAPREDUCE-6473-v6.patch,
> avgtime.png
>
>
> During initialization in Cluster.java, the framework provider classes are
> loaded inside a sync block which can considerably increase job submission
> time when the number of submissions are high. The motive is to reduce time
> spent in this sync block safely to improve performance.
> {noformat}
> synchronized (frameworkLoader) {
> for (ClientProtocolProvider provider : frameworkLoader) {
> LOG.debug("Trying ClientProtocolProvider : "
> + provider.getClass().getName());
> {noformat}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
