Jun Rao commented on KAFKA-4178:

[~benstopford], thanks for the patch. I like your idea of consolidating the 
Rate to use just FixedSubWindowPolicy, i.e., if the window size is less than a 
full window size, we adjust it to the size of just 1 full window (instead of # 
samples * full window size). This addresses (1) the NaN issue where the 
recording and the measuring are done within the same ms since the adjusted 
window size will never be 0; (2) the issue where the rate can be super high 
when the measured window before adjustment is very small (which can lead to too 
long of a delay for throttled clients); (3) the issue of spiking with 
replication throttling since the spiking can only happen on the first window 
instead of # samples * full window size.

[~jjkoshy], does that sound good to you too?

> Replication Throttling: Consolidate Rate Classes
> ------------------------------------------------
>                 Key: KAFKA-4178
>                 URL: https://issues.apache.org/jira/browse/KAFKA-4178
>             Project: Kafka
>          Issue Type: Improvement
>          Components: replication
>    Affects Versions:
>            Reporter: Ben Stopford
> Replication throttling is using a different implementation of Rate to client 
> throttling (Rate & SimpleRate). These should be consolidated so both use the 
> same approach. 

This message was sent by Atlassian JIRA

Reply via email to