Simplify wait_time stats
This changes reporting of wait_time in the following ways:
* It stops reporting statsd counters for wait_time because a counter
for a time value does not make sense.
* It replaces zuul.node_type.LABEL.job.JOB.wait_time with
zuul.pipeline.PIPELINE.label.LABEL.wait_time. The interesting
thing about this metric is "how long do we wait for a given node
type?". That is not related to the job name, so we should not
include it in the metric. However, it is related to the pipeline
(because of pipeline priorities) so we should add it. This also
normalizes the terminology to label.
* It reports both variants of wait_time at the conclusion of a build.
Previously one was reported at the start. By doing that we could
have descrepancies of up to 1 hour in what should be identical data.
Change-Id: I921ca4c049903a68440a44dd6a088fa096d87693
1 file changed