[
https://issues.apache.org/jira/browse/KAFKA-1251?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13942454#comment-13942454
]
Jay Kreps commented on KAFKA-1251:
----------------------------------
Guozhong,
1. No idea. I did a comparison between a Coda Hale Histogram and my metrics
(both stand-alone and in the new producer) to make sure I hadn't blown
something. I haven't done any experiments with the old producer.
2. I agree the selector metrics are a little unintuitive. How about this:
"select-calls-per-second" => "Number of times the I/O layer checked for new I/O
to perform per second",
"io-wait-time-avg-ns" => "The average length of time the I/O thread speant
waiting for a socket ready for reads or writes in ns"
"io-wait-percentage" => "The fraction of time the I/O thread spent waiting."
"io-time-avg-ns" => "The average length of time for I/O per select call in
nanoseconds."
"io-percentage" => "The fraction of time the I/O thread spent doing I/O"
"batch-size-avg" => "The average number of bytes sent per partition
per-request." (calling it per batch doesn't help because what is a batch)
select-calls-per-second is a bit of an implementation detail, but it is
important to understand how many iterations the selector is actually doing to
interpret some of the other metrics.
I actually think waiting-threads and ready-partitions work fine without the
-count suffix. As in the phase "there are 5 waiting threads" or "there are 12
ready partitions".
> Add metrics to the producer
> ---------------------------
>
> Key: KAFKA-1251
> URL: https://issues.apache.org/jira/browse/KAFKA-1251
> Project: Kafka
> Issue Type: Sub-task
> Components: producer
> Reporter: Jay Kreps
> Assignee: Jay Kreps
> Attachments: KAFKA-1251.patch, KAFKA-1251_2014-03-19_10:19:27.patch,
> KAFKA-1251_2014-03-19_10:29:05.patch, KAFKA-1251_2014-03-19_17:30:32.patch
>
>
> Currently there are no metrics.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
