Repository navigation
Tests: Bound resource_monitor's charges by what a wrong charge shows - #417
Conversation
develop failed on macOS on the 42.1.0 version bump: the teller's 20 ms burn was charged 102.7 ms against a ceiling of 70. A pause of the runner lands on whichever script is running, so a fixed ceiling near the burn measures the runner. The ceilings now sit where a wrong charge would be. The bank's handler burns 400 ms, so a teller charged for it shows 420 or more and is held under 200. The bank's fixed ceiling is replaced by the two charges together not exceeding the frame they ran in, which is what charging time twice would break on any machine. The slow threshold moves to 200 ms, 170 from every burn: a pause of 80 would have crossed the old one as well.
|
No actionable comments were generated in the recent review. WalkthroughThe resource monitor integration test increases the bank audit handler duration and slow threshold. It measures the frame duration when resources finish and checks the bank and teller charges against updated bounds. ChangesResource monitor timing test
Estimated code review effort: 2 (Simple) | ~10 minutes Change: Other Suggested reviewers: Merge Risk:
|
What
resource_monitor: charges each resource for its own script when one calls anotherfailed on macOS ondevelopat the 42.1.0 version bump (a commit that changes onlyVERSION): the teller's 20 ms burn was charged 102.7 ms against a ceiling of 70. The same commit passed on macOS in the release run.A pause of the runner lands on whichever script is running, so a fixed ceiling close to the burn measures the runner, not the attribution. Same kind of fix as #412.
< 300) is replaced byteller + bank <= the frame's wall time: charging time twice breaks that on any machine, and a runner pause cannot.tellerStats->slowReports == 0.The test takes about 200 ms longer.
Tests
resource_monitorrun three times locally on Windows: 14/14 each time. Test-only change.Summary by CodeRabbit