Date: Sat, 12 Mar 2022 01:55:44 +0100 From: Michael Gmelin <grembo@freebsd.org> To: Kristof Provost <kp@freebsd.org> Cc: Johan Hendriks <joh.hendriks@gmail.com>, freebsd-net@freebsd.org, ">> \\\\\\\\Patrick M. Hausen\\\\" <hausen@punkt.de> Subject: Re: epair and vnet jail loose connection. Message-ID: <43AA6B37-6235-4787-A03F-B4C264C75A58@freebsd.org> In-Reply-To: <41ED1534-5E98-4D46-A562-811E80F82C5F@FreeBSD.org> References: <41ED1534-5E98-4D46-A562-811E80F82C5F@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
> On 12. Mar 2022, at 01:21, Kristof Provost <kp@freebsd.org> wrote: >=20 > =EF=BB=BFOn 11 Mar 2022, at 17:44, Johan Hendriks wrote: >>> On 09/03/2022 20:55, Johan Hendriks wrote: >>> The problem: >>> I have a FreeBSD 14 machine and a FreeBSD 13-stable machine, both runnin= g the same jails just to test the workings. >>>=20 >>> The jails that are running are a salt master, a haproxy jail, 2 webserv= ers, 2 varnish servers, 2 php jails one for php8.0 and one with 8.1. All the= jails are connected to bridge0 and all the jails use vnet. >>>=20 >>> I believe this worked on an older 14-HEAD machine, but i did not do a lo= t with it back then, and when i started testing again and after updating the= OS i noticed that one of the varnish jails lost it's network connection aft= er running for a few hours. I thought it was just something on HEAD so never= really looked at it. But later on when i start using the jails again and te= sting a test wordpress site i noticed that with a simple load test my haprox= y jail within one minute looses it's network connection. I see nothing in th= e logs, on the host and on the jail. >>> =46rom the jail i can not ping the other jails or the IP adres of the br= idge. I can however ping the jails own IP adres. =46rom the host i can also n= ot ping the haproxy jail IP adres. If i start a tcpdump on the epaira interf= ace from the haproxy jail i do see the packets arrive but not in the jail. >>>=20 >>> I used ZFS to send all the jails to a 13-STABLE machine and copied over t= he jail.conf file as well as the pf.conf file and i saw the same behavior. >>>=20 >>> Then i tried to use 13.0-RELEASE-p7 and on that machine i do not see thi= s happening. There i can stress test the machine for 10 minutes without a pr= oblem but on 14-HEAD and 13-STABLE within a minute the jail's network connec= tion fails and only a restart of the jail brings it back online to exhibit t= he same behavior if i start a simple load test which it should handle nicely= . >>>=20 >>> One of the jail hosts is running under VMWARE and the other is running u= nder Ubuntu with KVM. The 13.0-RELEASE-p7 jail host is running under Ubuntu w= ith KVM >>>=20 >>> Thank you for your time. >>> regards >>> Johan >>>=20 >> I did some bisecting and the latest commit that works on FreeBSD 13-Stabl= e is 009a56b2e >> Then the commit 2e0bee4c7 if_epair: implement fanout and above is showin= g the symptoms described above. >>=20 > Interestingly I cannot reproduce stalls in simple epair setups. > It would be useful if you could reduce the setup with the problem into a m= inimal configuration so we can figure out what other factors are involved. If there are clear instructions on how to reproduce, I=E2=80=99m happy to he= lp experimenting (I=E2=80=99m relying heavily on epair at this point). @Kristof: Did you try on bare metal or on vms? -m
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?43AA6B37-6235-4787-A03F-B4C264C75A58>