[ 
https://issues.apache.org/jira/browse/KAFKA-2722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14985687#comment-14985687
 ] 

ASF GitHub Bot commented on KAFKA-2722:
---------------------------------------

GitHub user becketqin reopened a pull request:

    https://github.com/apache/kafka/pull/402

    KAFKA-2722: Improve ISR change propagation.

    The patch has two changes:
    1. fixed a bug in controller that it sends UpdateMetadataRequest of all the 
partitions in the cluster.
    2. Uses the following rules to propagate ISR change: 1) if there are ISR 
changes pending propagation and the last ISR change is more than five seconds 
ago, propagate the changes. 2) if there is ISR change at T in the recent five 
seconds, delay the propagation until T + 5s. 3) if the last propagation is more 
than 1 min ago, ignore rule No.2 and propagate ISR change if there are changes 
pending propagation.
    
    This algorithm avoids a fixed configuration of ISR propagation interval as 
we discussed about in KIP-29.

You can merge this pull request into a Git repository by running:

    $ git pull https://github.com/becketqin/kafka KAFKA-2722

Alternatively you can review and apply these changes as the patch at:

    https://github.com/apache/kafka/pull/402.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 #402
    
----
commit 13892856d806183536657f0c3ea2aa63b1f1c4f2
Author: Jiangjie Qin <becket....@gmail.com>
Date:   2015-11-02T01:26:27Z

    Improve ISR change propagation.

----


> Improve ISR change propagation
> ------------------------------
>
>                 Key: KAFKA-2722
>                 URL: https://issues.apache.org/jira/browse/KAFKA-2722
>             Project: Kafka
>          Issue Type: Bug
>            Reporter: Jiangjie Qin
>            Assignee: Jiangjie Qin
>            Priority: Blocker
>             Fix For: 0.9.0.0
>
>
> Currently the ISR change propagation interval is hard coded to 5 seconds, 
> this might still create a lot of ISR change propagation for a large cluster 
> in cases such as rolling bounce. The patch uses a dynamic propagation 
> interval and fixed a performance bug in IsrChangeNotificationListener on 
> controller.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to