On Tue, Apr 18, 2017 at 05:18:13PM +0300, Slawa Olhovchenkov wrote: > > Now the remaining questions are : > > - how is it possible to have such a large timeout in the timer queue > > - why doesn't the situation fix by itself on the next millisecond when > > it's not supposed anymore to be possible to have such a large timeout ; > > May be tv_update_date(delta_ms, status) for negative status do nothing.
Yes it does, it reinitializes the offset to zero so that now_ms is the exact copy of the output of gettimeofday(). It's normally used at initialization (I've checked this as well). > > - why doesn't it affect other pollers > > What you mean? Only kqueue on freebsd 11 reports an issue for now. It could be due to an API difference or anything but for now that's an element to keep in mind. > Also, logging only timeout.tv_sec/timeout.tv_nsec don't help you: I am > already see this negative values (timeout.tv_sec is about -10700 at > 10:38 UTC, timeout.tv_nsec very large negative and changed, don't > record, sorry). Olivier just told me he managed to reproduce so we'll save some production now :-) Cheers, Willy
