From: Mike Snitzer Subject: Re: RFC for multipath queue_if_no_path timeout. Date: Thu, 17 Oct 2013 17:13:38 -0400 Message-ID: <20131017211338.GB30993@redhat.com> References: <1380215696.25252.36.camel@bobble.lax.corp.google.com> <20130926172422.GA31328@agk-dp.fab.redhat.com> <1380216716.25252.39.camel@bobble.lax.corp.google.com> <20130926173814.GB31328@agk-dp.fab.redhat.com> <1380217633.25252.46.camel@bobble.lax.corp.google.com> <20130926232241.GC31328@agk-dp.fab.redhat.com> <20130926234957.GA3658@redhat.com> <1382036590.1980.32.camel@bobble.lax.corp.google.com> <20131017191511.GA30452@redhat.com> <1382042759.1980.40.camel@bobble.lax.corp.google.com> Reply-To: device-mapper development Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline In-Reply-To: <1382042759.1980.40.camel@bobble.lax.corp.google.com> List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: dm-devel-bounces@redhat.com Errors-To: dm-devel-bounces@redhat.com To: Frank Mayhar Cc: dm-devel List-Id: dm-devel.ids On Thu, Oct 17 2013 at 4:45pm -0400, Frank Mayhar wrote: > On Thu, 2013-10-17 at 15:15 -0400, Mike Snitzer wrote: > > Shouldn't you wait for the application to shutdown before ripping the > > network out? Seems odd to just throw away queued IO. > > In general, yes, and we do that, but there's still the occasional > shutdown that doesn't go through those extra steps. > > > A proper shutdown sequence really should avoid this problem in general, > > the multipath daemon would only be shutdown once all mpath devices are > > deactivated. > > > > Then, if you still want to gracefully handle the case where there is no > > network (and hence no paths) on shutdown the multipathd would still be > > around to transition to a table that doesn't have queue_if_no_path. > > Of course, this still depends on the daemon surviving everything else > circumstances can throw at it. > > In our particular case, changing the shutdown sequence is far more > difficult than using the kernel mechanism I've already provided. This > would also simplify things for other folks who use queue_if_no_path. Cannot say that argument wins me over but I will say that if you intend to take the approach to have the kernel have a timeout; please pursue the approach Hannes offered: https://patchwork.kernel.org/patch/2953231/ It is much cleaner and if it works for your needs we can see about getting a tested version upstream.