Currently there are a number of tests in the CppSuite that validate statistic values at the end of the test run, an example is the hs_cleanup test which checks the number of pages removed by a checkpoint to test that that functionality is working and that it is removing roughly the expected number of pages. However this hasn't translated well as often when it fails the ranges are modified and a potential regression could be overlooked.
Since the completion of PM-1707 it has become significantly easier to create new performance plots or graphs in evergreen / atlas. I'd like to leverage this new functionality using the CppSuite. If we can plot graphs of the end of run statistics then we can add them to the performance plots page and have them validated by the triage baron.
If this is feasible I think there are three pieces of work here:
- Update the cppsuite to output interesting stats from each test.
- Update the perf plot pipeline to be able to consume them.
- Figure out how to plot stats that may be in very different ranges, e.g. a test that outputs a stat between 0-100 and another stat between 500000 and 600000.
jeremy.thorp do you think this is a valid use case for the new perf plot pipeline? If so I think we can split this into two tickets and get the work scheduled.