[
https://issues.apache.org/jira/browse/HBASE-26868?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17509559#comment-17509559
]
Xiaolin Ha commented on HBASE-26868:
------------------------------------
Thanks, [~bbeaudreault] . The original purpose I introduced read/write requests
size to metrics is to design request size cost functions for the
StochasticLoadBalancer. And I used the same counter as the request count in our
company.
If the worry here is the number overflow, we can reset the counter to 0 when it
is negative, just like when the region moved or RS aborted the request count is
reset. I think the key point here is that the accumulated counter result is not
as important as the rate that the counter changes(for balancer and QPS).
While the counter is used in the balancer cost functions, so the rate
calculator should handle the decrease of the counters, e.g.
[100,200,0,100,200], as describe in HBASE-26872.
> Introduce read/write request size metrics
> -----------------------------------------
>
> Key: HBASE-26868
> URL: https://issues.apache.org/jira/browse/HBASE-26868
> Project: HBase
> Issue Type: Sub-task
> Components: regionserver
> Affects Versions: 2.4.11
> Reporter: Xiaolin Ha
> Assignee: Xiaolin Ha
> Priority: Major
> Fix For: 2.5.0, 2.6.0, 3.0.0-alpha-3
>
>
> Calculating the read and write size(unit is bytes) in the region, and
> aggregated in table and regionserver levels.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
