Addition of formatted logging possibly negatively effects build times and results.
Context
After merging the formatted logging changes (https://gitlab.com/BuildStream/benchmarks/merge_requests/51) the build stats for baserock binutils changed (https://gitlab.com/BuildStream/benchmarks/-/jobs/224488078) with a 10 second average degradation in total build time over previous days results (https://buildstream.gitlab.io/-/benchmarks/-/jobs/224488078/artifacts/pipeline_cache/results-2019-06-04-00:04:00.json). This is inconsistent with tests carried out with the 100 square test between logging and no logging (which showed no distinct change in performance), this might be specific to the baserock test but needs to be better understood.
If you are raising a bug, please summarize the bug and include:
- Expected behaviour - formatted logging should not have a substantial impact on measured "stopwatch" performance
- Current behaviour - degradation of 10s~12s in baserock binutils test.
- Run master branch on master test runner compare results to those prior to 03/06/19
Task Description
Describe the steps to take / implementation plan. If it's a bug, do you know how it may be tackled?
-
Confirm issue is specific to formatted logging introduction by reverting change and re-testing. -
Seek further advice and interim reversion in benchmarks/baserock bin utils test until buildstream provides some resolution.
Acceptance Criteria
Confirmation that issue is specifically with baserock build and that reversion of formatted logging changes in this tests removes the issue. Await resolution from buildstream.
average_time_master_Build_of_Baserock_stage1-binutils_for_x86_64_20190604-003028