Ok, so I'll try to do an explicitly kernal upgrade now. :-) --
Best Regards BaiYang [email protected] http://baiy.cn **** < END OF EMAIL > **** From: Willy Tarreau Date: 2015-11-20 18:55 To: Lukas Tribus CC: baiyang; haproxy Subject: Re: CPU 100% when waiting for the client timeout Hi Lukas, On Fri, Nov 20, 2015 at 11:34:50AM +0100, Lukas Tribus wrote: > >> # uname -a > >> Linux WD-G0-SRP1 3.2.0-29-generic #46-Ubuntu SMP Fri Jul 27 17:03:23 UTC > >> 2012 x86_64 x86_64 x86_64 GNU/Linux > > > > This one seems to have missed 3 years of bugfixes but anyway I don't see > > how any kernel bug could make haproxy fail, and if it did we'd have to > > find a workaround anyway (or a way to detect it). > > > FYI there are 2 epoll related fixes that have been backported to the ubuntu > 12.04 kernel, that the kernel above does *not* have: > > linux (3.2.0-64.96) precise; urgency=low > * parisc: fix epoll_pwait syscall on compat kernel > - LP: #1319885 > http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=ab3e55b119c9653b19ea4edffb86f04db867ac98 > > linux (3.2.0-38.59) precise-proposed; urgency=low > * epoll: prevent missed events on EPOLL_CTL_MOD > - LP: #1100376 > http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=128dd1759d96ad36c379240f8b9463e8acfd37a1 > > > I think the latter may be relevant? Yes looks like so. It may even explain the 900s timeout that ends up being reported as an error, if the event is not properly registered in the kernel and the error only pops up once the kernel declares the socket dead. It's just a guess of course, but I wouldn't rule this one out! Regards, Willy
