Date: Tue, 5 Dec 2006 12:30:27 +0200 From: "Ari Suutari" <ari.suutari@syncrontech.com> To: "Gleb Smirnoff" <glebius@FreeBSD.org> Cc: net@FreeBSD.org Subject: Re: kern/104377: [carp] [patch] CARP interface doesn't go up on VmWare Message-ID: <100a01c71858$64b1efc0$6602a8c0@sad.syncrontech.com> References: <200612041538.kB4FcQjk073583@freefall.freebsd.org> <0eaa01c71835$f12055f0$6602a8c0@sad.syncrontech.com> <20061205100407.GB32700@cell.sick.ru>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi,
> On Tue, Dec 05, 2006 at 08:23:50AM +0200, Ari Suutari wrote:
> A> Hi,
> A>
> A> OK, what would then be the right fix to get things working
> A> under vmware ? I run a bunch of servers here and some of
> A> them are redundant pairs. We have some pressure to virtualize
> A> those servers, but we cannot do it as the carp does not work.
> A>
> A> I don't really get it why the link state is so important here, as
> A> to my understanding carp works in similar as vrrp, using
> A> heartbeats ? Also, the current state of matters is more confusing,
> A> since you can get the carp interface up by issuing another
> A> "ifconfig up" (people suggested this to me, but I cannot accept
> A> that a system providing redundancy requires this kind of kludgery)
> A>
> A> I can accept that my solution is not the correct one, but
> A> it is a little hard to accept turning it down without
> A> giving any suggestion how to really fix things.
>
> When one created a redundant routers, he enables CARP an all
> interfaces of the router. Imagine, that one interface goes down,
> but CARP doesn't notice that and keeps claiming to be the master
> on the other interfaces. Traffic comes to it, and it sends it to
> downed interface.
Doesn't the other machine notice this from the absense of hearbeat
on that interface ? I thought that this could combined with
net.inet.carp.preempt sysctl to force carp to fail over the
other interfaces in this case also. If this doesn't work then you
are right; I really didn't test this under vmware (I should
have tested it, of course).
>
> If interface, that went down, had reported its state then CARP
> would had noticed that and would had lowered its priority,
> gave up mastering, and became backup. This will be redundant.
Is it impossible to add link state reporting to lnc driver ?
I think this would be the perfect fix and acceptable by
everyone.
Ari S.
>
> A> ----- Original Message -----
> A> From: "Gleb Smirnoff" <glebius@FreeBSD.org>
> A> To: <ari.suutari@syncrontech.com>; <glebius@FreeBSD.org>;
> A> <freebsd-bugs@FreeBSD.org>
> A> Sent: Monday, December 04, 2006 5:38 PM
> A> Subject: Re: kern/104377: [carp] [patch] CARP interface doesn't go up on
> A> VmWare
> A>
> A>
> A> >Synopsis: [carp] [patch] CARP interface doesn't go up on VmWare
> A> >
> A> >State-Changed-From-To: open->closed
> A> >State-Changed-By: glebius
> A> >State-Changed-When: Mon Dec 4 15:34:00 UTC 2006
> A> >State-Changed-Why:
> A> >I am sorry, but I am not going to commit this patch. Let me explain.
> A> >CARP is not going to work properly on interfaces that do not report
> A> >its state being changed. The proposed patch will change CARP behavior
> A> >to be pretending to work when interface doesn't support reporting
> A> >its link state. I think it is better to refuse to work earlier, then
> A> >pretend to be working but don't provide any redundancy. The proposed
> A> >patch is going to confuse people.
> A> >
> A> >http://www.freebsd.org/cgi/query-pr.cgi?pr=104377
>
> --
> Totus tuus, Glebius.
> GLEBIUS-RIPN GLEB-RIPE
>
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?100a01c71858$64b1efc0$6602a8c0>