This is related to #2720 but is triggering frequently on playback; not just recording, on AMD CPUs. (Specifically threadrippers.)
Bumping the skid from 10000 to 30000 worked for Nika; but made it extremely slow.
The issue is (I'm told) that on these CPUs the PMU interrupt is fired with a variable (and often too high) latency; causing us to overshoot the bounds.
Fiddling with kernel.perf_event_max_sample_rate as described in #2720 might be a solution, but personally I never saw those error messages in my failures.
This is related to #2720 but is triggering frequently on playback; not just recording, on AMD CPUs. (Specifically threadrippers.)
Bumping the skid from 10000 to 30000 worked for Nika; but made it extremely slow.
The issue is (I'm told) that on these CPUs the PMU interrupt is fired with a variable (and often too high) latency; causing us to overshoot the bounds.