Make trace-pipeline job run before pipeline_and_job_duration job
What does this MR do and why?
Describe in detail what your merge request does and why.
If the trace-pipeline
and pipeline_and_job_duration
jobs run at the
same time, the pipeline_and_job_duration
job might finish first which
results in the traces generated by trace-pipeline
containing the pipeline_and_job_duration
job as well.
Since the pipeline_and_job_duration
job is executed many hours after (because it is triggered by a manual job) the rest of the pipeline, this can skew the
trace of the pipeline. Here is an example trace that is skewed by the metrics:pipeline_and_job_duration
job: https://observe.gitlab.com/v1/jaeger/1112072/trace/0ff0bca252ad336472acfe15b0b6f866 (scroll right to the bottom).
By running the trace job first, we avoid this problem.
The duration metrics recorded by the pipeline_and_job_duration
job can
also be skewed because of the same problem. To solve this, the
trace-pipeline
job has been explicitly ignored when recording pipeline
durations.
Author Check-list
-
Has documentation been updated?