On Thu Oct 12 15:32:20 2023 +0000, Francois Gouget wrote:
Do we know why we are getting these timing failures? Getting 150 ms instead of 125 ms I can understand: just a scheduling delay because the system is busy. But how can we get 31 ms, less than the expected 50 ms minimum? What is the reasoning behind the 50 and 100 ms thresholds?
No very clever reasoning, just some values which were supposedly different enough, as well as compatible with the performance default read-ahead which is apparently of ~70ms (difference between QUEUE and ATTIME delivery), and not too large to avoid slowing down the test.
Fwiw I only see spurious failures with unexpected high times (~265ms) for QUEUE delivery mode in the recent runs, maybe it's a genuine bug rather than timing failure.