kafka-jira mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Dong Lin (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (KAFKA-5871) bound the throttle time in byte rate quota
Date Tue, 17 Oct 2017 01:49:00 GMT

    [ https://issues.apache.org/jira/browse/KAFKA-5871?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16206901#comment-16206901

Dong Lin commented on KAFKA-5871:

[~junrao] I am not sure we can do this. The issue with having a hard bound on the throttle
time is that, if the parallelism between producers and broker is high, which can happen if
inflight request is high or if there are many producers of the same user (e.g. with MapReduce),
then the throughput used by this client-id can be unlimited. This weakens the semantics of
quota if we can not bound the throughput used by a given user.

> bound the throttle time in byte rate quota
> ------------------------------------------
>                 Key: KAFKA-5871
>                 URL: https://issues.apache.org/jira/browse/KAFKA-5871
>             Project: Kafka
>          Issue Type: Improvement
>          Components: core
>    Affects Versions:
>            Reporter: Jun Rao
> Currently, if a user sets a very small byte rate quota, the calculated throttled time
can be higher than the request timeout in the client, which will cause the client request
to time out and be retried.
> On the other hand, for the request time quota, we bound the throttled time with the metric
window size. This prevents client timeout and seems to be better. We probably want to implement
the same thing in the byte rate quota.

This message was sent by Atlassian JIRA

View raw message