[
https://issues.apache.org/jira/browse/TS-3436?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14541962#comment-14541962
]
Alan M. Carroll commented on TS-3436:
-------------------------------------
I spoke about this at the recent ATS summit and the consensus was this was fine
and preferable to having a separate timings structure per transaction. The main
issue was a desire for per plugin data of this type but that's a non-trivial
API change beyond the scope of this work.
> Add milestone data for API times
> --------------------------------
>
> Key: TS-3436
> URL: https://issues.apache.org/jira/browse/TS-3436
> Project: Traffic Server
> Issue Type: Improvement
> Components: Core
> Reporter: Alan M. Carroll
> Assignee: Alan M. Carroll
> Fix For: 6.0.0
>
>
> Add two new milestones that track API time, "active" and "total". These are
> reasonably equivalent to "wall clock time" and "cpu time" concepts for
> process. The total time measures how long, in real time, a transaction spends
> involved with API callouts in plugins. The active time tracks the amount of
> time spent in an API callout. When these two values are different this means
> time is being spent either in lock contention to run the plugin continuation
> or waiting for the plugin to re-enable the transaction. Originally I had just
> the "total" counter but I think the ability to detect contention and wait is
> very valuable.
> The values are a bit odd - they are the actual amount of time spent plus the
> start time of the state machine. I debated for a while about that with some
> other contributors and the concenssus was that overall the best option
> because it makes those milestones behave exactly like all the others (e.g.,
> you subtract sm_start to get the real value).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
