On Thu, Apr 26, 2018 at 09:25:59AM +0200, Willy Tarreau wrote: > On Thu, Apr 26, 2018 at 10:21:27AM +0300, Slawa Olhovchenkov wrote: > > > > Pollers distinct from frontend? > > > > Can I bind pollers to CPU? > > > > > > Each thread has its own poller. Since you map threads to CPUs you indeed > > > have one poller per CPU. > > > > Each pooler pool all sockets or only sockets from binded frontends? > > All sockets. All FDs in fact. This is normal, it's an event loop, it needs > to be notified of *any* event (fd activity, signal).
I am mean in case of dedicated listen socket pooler also can be dedicated, for load planing. For example: frontend tcp1 bind x.x.x.206:443 bind-process 1/9-1/16 mode tcp threads 1-8 don't need any events from this socket at all. This is also reduce communication w/ kernel, rise locality of data. I mean locality accpeted socket only to one pooler will be good too. > > > Please try this patch. It works for me. I finally managed to reproduce > > > the issue even with epoll(), it's just that it's much harder to see it, > > > but after trying multiple times eventually you see it as well. Under > > > poll() however the issue occasionally happens and disappears by itself. > > > > Like work for me too, thank > > Great, thank you. You can use it for now to fix your production, it will > likely be the one we'll use in 1.8 in the end. We're still working on > addressing the root cause in 1.9 first. 10x