[
https://issues.apache.org/jira/browse/KAFKA-3847?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15419261#comment-15419261
]
ASF GitHub Bot commented on KAFKA-3847:
---------------------------------------
GitHub user ewencp opened a pull request:
https://github.com/apache/kafka/pull/1727
KAFKA-3847: Use a separate producer per source task
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/ewencp/kafka kafka-3847-per-task-producers
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/1727.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 #1727
----
commit 98ec7f69bb5154a74a25c5e566e92bde1479796b
Author: Ewen Cheslack-Postava <m...@ewencp.org>
Date: 2016-08-12T18:22:04Z
KAFKA-3847: Use a separate producer per source task
----
> Connect tasks should not share a producer
> -----------------------------------------
>
> Key: KAFKA-3847
> URL: https://issues.apache.org/jira/browse/KAFKA-3847
> Project: Kafka
> Issue Type: Improvement
> Components: KafkaConnect
> Affects Versions: 0.10.0.0
> Reporter: Ewen Cheslack-Postava
> Assignee: Ewen Cheslack-Postava
> Priority: Critical
> Fix For: 0.10.1.0
>
>
> Currently the tasks share a producer. This is nice in terms of potentially
> coalescing requests to the same broker, keeping port usage reasonable,
> minimizing the # of connections to brokers (which is nice for brokers, not so
> important for connect itself). But it also means we unnecessarily tie tasks
> to each other in other ways -- e.g. when one needs to flush, it we
> effectively block it on other connector's data being produced and acked.
> Given that we allocate a consumer per sink, a lot of the arguments for
> sharing a producer effectively go away. We should decouple the tasks by using
> a separate producer for each task (or, at a minimum, for each connector's
> tasks).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
