GitHub user mridulm opened a pull request:
https://github.com/apache/spark/pull/5852
[SPARK-6166] [Spark Core] Limit number of concurrent outbound connections.
The overall number of requests in the cluster comes down with this (though
in theory, you can still have one node getting swamped). In our experiments
this has allowed jobs to progress where they previously die due to workers
repeatedly going down when number of requests go up.
Typically happens when you have large number of nodes, and data per node is
not too high (in one of the many jobs in the application) - drastic lowering of
spark.reducer.maxSizeInFlightMb results is very poor performance overall.
Note: I have not included any testcases - suggestions on how to do the same
in current spark test framework would be great - I have not looked at spark
test suites in a while. Simulating this is slightly 'interesting'
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/mridulm/spark limit_outbound_connections
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/5852.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #5852
----
commit 32168fc7a14cc341220ebf7548f568acc7e774d0
Author: Mridul Muralidharan <[email protected]>
Date: 2015-05-01T22:32:02Z
Limit number of outbound connections. We end up with a few nodes getting
swamped at times.
----
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
