+++ This bug was initially created as a clone of Bug #1177533 +++ Description of problem: The event loop is calculating timeout in seconds, but poll timeout should be milliseconds. This causes poll to wait up to 30 milliseconds instead of 30 seconds where there are no events to handle. Profiling a system running one vm for 5 mintues show _process_events is dominating the profile, taking 55% of cpu time. Version-Release number of selected component (if applicable): 55df4a7f1c6cfefdc70b439d0d5a3f735f6bbd4b How reproducible: Always Steps to Reproduce: 1. Run vdsm with very little load and profiling enabled Actual results: _process_events consuming more then 50% cpu time, called millions of times per 5 minutes Expected results: _process_events should consume almost no cpu time --- Additional comment from Nir Soffer on 2014-12-28 11:54:11 EST --- Fixed in master, needs trivial backport.
ok, vdsm-4.16.8.1-5.el6ev.x86_64 no high cpu found and _process_events was not present between highest 20 listed calls from profiling stats.