Thanks, @tombeard - it works and appears safe. However, in my own (Centos) tests, CPU utilisation when workers are idle is noticeably higher than before.

Increasing the sleep duration helps - so if it is implemented this way then it should be exposed as a mod param with a sensible default.

Personally, it seems like a (mostly) unnecessary trade-off between CPU cycles and job-processing delay, for what appears to be a limited number of cases (FreeBSD only one reported to my knowledge). Mutex by default for the remainder seems to be both more efficient and 'real-time'.

If everyone else agrees the alternative approach is better all round, then I am happy to merge once the sleep duration has been made configurable. It would still be good to hear @miconda's thoughts on making this a compile-time decision, though.

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.