[
https://issues.apache.org/jira/browse/KAFKA-2225?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14577571#comment-14577571
]
Jun Rao commented on KAFKA-2225:
--------------------------------
A few notes on this.
1. We probably should only do this on the new java consumer.
2. We will need some kind of notion of zones in both the brokers and the
clients.
3. We probably need to fix ISR in TMR first. There is some discussion about
this in KAFKA-1367.
> Allow fetching from ISR
> -----------------------
>
> Key: KAFKA-2225
> URL: https://issues.apache.org/jira/browse/KAFKA-2225
> Project: Kafka
> Issue Type: Improvement
> Reporter: Theo Hultberg
> Assignee: Parth Brahmbhatt
>
> Currently clients are not allowed to fetch from replicas, even if they are in
> sync with the master. If I'm not missing anything significant it shouldn't be
> any difference fetching from the leader or an ISR, besides maybe some extra
> latency.
> For our use case it would be very beneficial to be able to fetch from
> replicas instead of just the leader. We run Kafka clusters that replicate
> across EC2 availability zones, and each byte sent between zones costs money.
> This bandwith usage costs us about the same as it costs to run the instances.
> If we could fetch from a replica in the same zone as the client we could
> avoid some of this cost.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
