[
https://issues.apache.org/jira/browse/KAFKA-2950?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15043568#comment-15043568
]
ASF GitHub Bot commented on KAFKA-2950:
---------------------------------------
GitHub user jkreps opened a pull request:
https://github.com/apache/kafka/pull/632
KAFKA-2950: Fix performance regression in the producer
Removes all the System.currentTimeMillis calls to help with performance on
small messages.
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/jkreps/kafka producer-perf-regression
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/632.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 #632
----
commit 69d94701d4de3283f3783826edb9321dc6c800ec
Author: Jay Kreps <jay.kr...@gmail.com>
Date: 2015-12-05T23:08:04Z
Fix performance regression in the producer for small messages due to too
many System.currentTimeMillis() calls.
----
> Performance regression in producer
> ----------------------------------
>
> Key: KAFKA-2950
> URL: https://issues.apache.org/jira/browse/KAFKA-2950
> Project: Kafka
> Issue Type: Bug
> Reporter: Jay Kreps
>
> For small messages the producer has gotten slower since the 0.8 release. E.g.
> for a single thread on linux sending 100 byte messages the decrease seems to
> be about 30%. The root cause seems to be that the new timeout we added for
> max.block.ms ends up doing about 4 more system calls to check the ellapsed
> time.
> The reason for these calls is to detect slow serializers or partitioners. But
> I think this is not worth the performance hit. I think we can say the config
> is only for blocking due to metadata or memory allocation.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
