Hi Otto, On 22-02-2023 08:25, Otto Kekäläinen wrote:
A fresh run today passed, proving the Debian autopkgtest host hardware/kernel/overload theory is likely the cause.
I don't think a pass proves anything in that respect. I've seen so many tests being flaky because they assume things that aren't always true (like being the only thing running, or having race conditions that get exposed on certain hardware or setups).
Regarding "overload"; in my opinion a database server like MariaDB should be resilient against a heavily loaded server. No data corruption should occur under load. I'm wondering what it means if a loaded system could cause your tests to fail. Could it be that the *test* is sensitive to a loaded system (e.g. race conditions) and that a failure isn't representing the actual MariaDB server?
Paul
OpenPGP_signature
Description: OpenPGP digital signature
