You are viewing a plain text version of this content. The canonical link for it is here.
Posted to jira@kafka.apache.org by "Ashish Surana (JIRA)" <ji...@apache.org> on 2018/09/07 10:30:00 UTC

[jira] [Comment Edited] (KAFKA-7149) Reduce assignment data size to improve kafka streams scalability

    [ https://issues.apache.org/jira/browse/KAFKA-7149?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16606957#comment-16606957 ] 

Ashish Surana edited comment on KAFKA-7149 at 9/7/18 10:29 AM:
---------------------------------------------------------------

[~guozhang] As you suggested, below changes can reduce the assignment data size especially when a topic-group has multiple input topics as topic-partitions are repeated in assignment-info in that case:

1. Replace num.partitions with num.tasks

2. Replace [topic-partitions] with [task-ids]

This will also require some changes to identify the corresponding host while doing lookup for a given key in store.

Along with these encoding changes, compression is definitely required for further size reduction. [~mjsax] suggestion to handle this is what I was also thinking and makes sense to support backward/forward compatibility.


was (Author: asurana):
[~guozhang] As you suggested, below changes can reduce the assignment data size especially when a topic-group has multiple input topics as topic-partitions are repeated in assignment-info in that case:

1. Replace num.partitions with num.tasks

2. Replace [topic-partitions] with [task-ids]

This will also require some changes to identify the corresponding host while doing lookup for a given key in store.

Along with these encoding changes, compression is definitely required for further size reduction. [~mjsax] suggestion to handle this is what I was also thinking and makes sense to support backward compatibility.

> Reduce assignment data size to improve kafka streams scalability
> ----------------------------------------------------------------
>
>                 Key: KAFKA-7149
>                 URL: https://issues.apache.org/jira/browse/KAFKA-7149
>             Project: Kafka
>          Issue Type: Improvement
>          Components: streams
>            Reporter: Ashish Surana
>            Assignee: Ashish Surana
>            Priority: Major
>
> We observed that when we have high number of partitions, instances or stream-threads, assignment-data size grows too fast and we start getting below RecordTooLargeException at kafka-broker.
> Workaround of this issue is commented at: https://issues.apache.org/jira/browse/KAFKA-6976
> Still it limits the scalability of kafka streams as moving around 100MBs of assignment data for each rebalancing affects performance & reliability (timeout exceptions starts appearing) as well. Also this limits kafka streams scale even with high max.message.bytes setting as data size increases pretty quickly with number of partitions, instances or stream-threads.
>  
> Solution:
> To address this issue in our cluster, we are sending the compressed assignment-data. We saw assignment-data size reduced by 8X-10X. This improved the kafka streams scalability drastically for us and we could now run it with more than 8,000 partitions.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)