[
https://issues.apache.org/jira/browse/KAFKA-7713?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16718002#comment-16718002
]
Jun Rao commented on KAFKA-7713:
--------------------------------
io-wait-ratio is calculated as the ratio of accumulated selector waiting time
over the length of the metric window. So as the metric window rolls, those
accumulated waiting time falling into the rolled window will cause the metric
to go above 1. Not sure what's the best way to address this. One way is to
treat the Rate/Meter metric special. When we record the value, we split the
value into the corresponding sample windows and bound the value by the window
size.
> producer io-wait-ratio > 1
> --------------------------
>
> Key: KAFKA-7713
> URL: https://issues.apache.org/jira/browse/KAFKA-7713
> Project: Kafka
> Issue Type: Bug
> Affects Versions: 2.1.0
> Reporter: dan norwood
> Priority: Major
>
> i am running a test on a streams application and gathering jmx measurements
> to determine what is causing some lag. using `kafka.tools.JmxTool` i was
> gathering the following attributes `'io-ratio', 'io-wait-ratio',
> 'select-rate', 'batch-size-avg', 'compression-rate-avg', 'record-size-avg',
> 'records-per-request-avg'` on my streams instances producers. i noticed that
> i was getting `io-wait-ratio > 1`, but according to docs it is "The fraction
> of time the I/O thread spent waiting."
>
> some example lines from jmxtool
> |StreamThread-8-producer:batch-size-avg|StreamThread-8-producer:compression-rate-avg|StreamThread-8-producer:io-ratio|*StreamThread-8-producer:io-wait-ratio*|StreamThread-8-producer:record-size-avg|StreamThread-8-producer:records-per-request-avg|StreamThread-8-producer:select-rate|
> |662.2613636|0.3425814926|1.01E-04|*1.178371974*|172.2045455|38.71666667|3.855527588|
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
