Closed exodist closed 3 years ago
no, this was not the issue
Reopening, but the issue is different than initially expected.
If a resource becomes available after a test run has started, but was not unavailable due to another process using it (DB init takes longer that stage init) then the IPC will block until at least 1 process exited before realizing it has resources to fill more job slots.
In a run at $work only 7 tests were running after a timeout despite -j8 and 8 total resources. I suspect the logic that restores resources to the pool does not work right with timeout kills.