Date: Fri, 4 Dec 2009 01:13:23 -0800 (PST) From: alexpalias-bsdstable@yahoo.com To: FreeBSD Stable <freebsd-stable@freebsd.org>, Zaphod Beeblebrox <zbeeble@gmail.com> Subject: Re: Quggaa locking hard. Message-ID: <776232.27938.qm@web56408.mail.re3.yahoo.com> In-Reply-To: <5f67a8c40912031946v1ab3efua5e43bd7d767361a@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
I have also seen this with a recent version of FreeBSD 8 (I know 8.0-BETA2 = didn't have this problem, also I have an 8.0-RC1 without problems, but I t= hink RC3 did have it, and I'm sure -RELEASE has it).=0A=0AA few more detail= s: =0A=0AIt happened both on amd64 and i386. I couldn't debug amd64 (it w= as a live server and we couldn't afford it), but on i386 flowcleaner was us= ing a LOT of CPU. =0A=0AIt seemed to happen after booting, when quagga was= importing global routing tables (~300k routes) from 2 BGP sessions. At le= ast one of the sessions seemed to finish importing routes, but the kernel r= outing table seemed to be growing very slowly. =0A=0ADoing "netstat -nr | = wc -l" took way longer than usual (20-30 seconds versus 9 seconds now), and= it only reported about 100k routes. Doing it again after a minute or so s= howed the number of routes grew by around 10k.=0A=0ADuring this time, both = quagga and zebra were very slow to respond to a new telnet session opened t= o them.=0A=0AAs a workaround, I did sysctl net.inet.flowtable.enable=3D0. = This didn't ease the load on the CPU, but having it in /etc/sysctl.conf and= rebooting did help (quagga started up normally and all routes are where th= ey should be).=0A=0AHope this helps=0A Alex=0A=0A--- On Fri, 12/4/09, Za= phod Beeblebrox <zbeeble@gmail.com> wrote:=0A=0A> From: Zaphod Beeblebrox <= zbeeble@gmail.com>=0A> Subject: Quggaa locking hard.=0A> To: "FreeBSD Stabl= e" <freebsd-stable@freebsd.org>=0A> Date: Friday, December 4, 2009, 5:46 AM= =0A> I'm still investigating this, but my=0A> quagga is locking hard on Fre= eBSD 8.0=0A> and not locking hard on 7.2.=A0 It seems (at this early=0A> po= int in the=0A> investigation) that both bgpd and zebra are wedging and=0A> = zebra is listed as=0A> being in the "RUN" state.=0A> =0A> curiously, the lo= ad is also 4.0 (exactly the number of=0A> cores in the machine)=0A> even th= ough the machine also reads 100% idle.=0A>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?776232.27938.qm>