Created attachment 3353 [details] build perf old vs new The metrics that are being saved in globalres.log file by oe-build-perf-test are not being consistent, giving big differences between 2 executions, see attached log.
Do you have logs from the tests? Seems that the tests are (mostly) failing because the elapsed time is only few minutes although it should be about an hour.
Seems like the executions were successfully completed and I could not find the logs, if the execution is completed logs are saved? where are placed?
Created attachment 3366 [details] Logs attached
Created attachment 3367 [details] ubuntu logs
Logs attached for ubuntu a fedora machines
Seems that you attached the same log twice(?) I find it difficult believing that the attached log is from any of the results in the .csv file. There is no correlation between the time logs (results.log.N) and the time values in the csv.
Created attachment 3370 [details] logs_ubuntu_fedora
(In reply to comment #6) > Seems that you attached the same log twice(?) I find it difficult believing > that the attached log is from any of the results in the .csv file. There is > no correlation between the time logs (results.log.N) and the time values in > the csv. Sorry I made a mistake uploading the correct logs, I already attached the correct ones
Now I understood the issue. The problem was that a badly broken version of the function writing globalres.log ended up in my first patchset. End result being that all correct values were there but they were basically in random order (because of usage of dict() instead of OrderedDict). The problem should now be fixed in latest master (after 3bdb7b2e512b2f160360e95ed5b2be3871ec0b4b). Please test and re-open this bug if you still see some issues.
Metrics are consistent for 2.3 M3 rc2. poky commit: 415b72ffcbd26e5f3664370d8b2a9b8105fb6342