On Thu, Apr 26, 2018 at 10:58:07AM +0300, Slawa Olhovchenkov wrote: > > > 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. > > > > That's exactly what happens. > > No, as I see. After start load on this sockets I am see CPU use on CPU's > 0-7 too. All CPU rise load simultaneous.
It should not be the case, but now that I'm thinking about it, it would in fact be the same boot race as the issue on the DNS : the FDs are registered before forking the threads, and then we have to update the relevant pollers in each thread. But the information used to know what state the FD is in to know if we need to perform an update is shared, so depending on the boot order, the state can be updated on one poller preventing the other ones from getting it. That's what we're working on. Now I fear that we'll have to perform deeper changes to 1.8 than what we initially expected :-/ > May be I am miss some in config? No, your config is simple and OK. Willy
