[
https://issues.apache.org/jira/browse/STORM-2165?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15592615#comment-15592615
]
P. Taylor Goetz commented on STORM-2165:
----------------------------------------
Please see this for background: https://issues.apache.org/jira/browse/STORM-2169
If we use that approach for aggregating, then aggregating at the supervisor
would actually lead to more traffic to the metric store -- in addition to
forwarding the original metric, the supervisor would also be sending a value
for each level of aggregation.
I think if we have sane worker reporting intervals we can avoid too much of an
impact on the metrics store. Another future optimization would be to try to
stagger worker report times to avoid all workers reporting at once, but I'd
have to think more about how we would do that.
> Implement default worker metrics reporter and changes in Supervisor to read
> metrics
> -----------------------------------------------------------------------------------
>
> Key: STORM-2165
> URL: https://issues.apache.org/jira/browse/STORM-2165
> Project: Apache Storm
> Issue Type: Improvement
> Reporter: Alessandro Bellina
>
> A reporter that writes metrics to a known location such that Supervisor can
> read them and send them to the collection service (initially Nimbus).
> Also for this ticket to be complete, Supervisors should be able to read and
> send metrics to Nimbus.
> This doesn't involve aggregation, just individual worker metrics.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
