[
https://issues.apache.org/jira/browse/KAFKA-3979?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15385865#comment-15385865
]
ASF GitHub Bot commented on KAFKA-3979:
---------------------------------------
GitHub user nepal opened a pull request:
https://github.com/apache/kafka/pull/1642
[KAFKA-3979] Optimize memory used by replication process by using ada…
…ptive fetch message size
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/nepal/kafka adaptive-replica-fetch-size
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/1642.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 #1642
----
commit 300be1adb0089df4990c0bc2f8aaf0a50556fc8b
Author: Andrey L. Neporada <anepor...@yandex-team.ru>
Date: 2016-07-20T13:37:15Z
[KAFKA-3979] Optimize memory used by replication process by using adaptive
fetch message size
----
> Optimize memory used by replication process by using adaptive fetch message
> size
> --------------------------------------------------------------------------------
>
> Key: KAFKA-3979
> URL: https://issues.apache.org/jira/browse/KAFKA-3979
> Project: Kafka
> Issue Type: Improvement
> Components: replication
> Affects Versions: 0.10.0.0
> Reporter: Andrey Neporada
>
> Current replication process fetches messages in replica.fetch.max.bytes-sized
> chunks.
> Since replica.fetch.max.bytes should be bigger than max.message.bytes for
> replication to work, one can face big memory consumption for replication
> process, especially for installations with big number of partitions.
> Proposed solution is to try to fetch messages in smaller chunks (say
> replica.fetch.base.bytes).
> If we encounter message bigger than current fetch chunk, we increase chunk
> (f.e. twofold) and retry. After replicating this bigger message, we shrunk
> fetch chunk size back until it reaches replica.fetch.base.bytes
> replica.fetch.base.bytes should be chosen big enough not to affect throughput
> and to be bigger than most of messages.
> However, it can be much less than replica.fetch.max.bytes.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
