On Wed, 14 Dec 2011 16:10:39 -0500 "J. Bruce Fields" <bfields@xxxxxxxxxxxx> wrote: > On Wed, Dec 14, 2011 at 11:28:22AM -0500, Steve Dickson wrote: > > > > > > On 12/14/2011 11:00 AM, Jeff Layton wrote: > > > On Wed, 14 Dec 2011 10:56:46 -0500 > > > Steve Dickson <SteveD@xxxxxxxxxx> wrote: > > > > > >> > > >> > > >> On 12/14/2011 10:37 AM, Jeff Layton wrote: > > >>> On Wed, 14 Dec 2011 10:29:49 -0500 > > >>> Steve Dickson <SteveD@xxxxxxxxxx> wrote: > > >>> > > >>>> > > >>>> > > >>>> On 12/14/2011 10:19 AM, Jeff Layton wrote: > > >>>>> On Wed, 14 Dec 2011 10:09:04 -0500 > > >>>>> Steve Dickson <SteveD@xxxxxxxxxx> wrote: > > >>>>> > > >>>>>> > > >>>>>> > > >>>>>> On 12/14/2011 08:57 AM, Jeff Layton wrote: > > >>>>>>> Generally, we want this daemon started before nfsd starts. Deal with the > > >>>>>>> situation where the pipe hasn't shown up yet. > > >>>>>> This can be done with your systemd start up script. Plus I'm not sure its > > >>>>>> a good idea to steal cpu cycles waiting for an event that may never happen... > > >>>>>> > > >>>>> > > >>>>> Presumably you just wouldn't start the daemon if you have no intent to > > >>>>> use it. > > >>>>> > > >>>>> It does sleep 1s between each check, so the time is fairly minimal, > > >>>>> but I'm definitely open to doing this differently. What may be > > >>>>> reasonable is adding code to the daemon to check and see if the > > >>>>> v4recoverydir is present. If it is, then just exit. Otherwise, wait for > > >>>>> the pipe to show up. > > >>>> Why just let the systemd scrips worry about the order of when to start > > >>>> things up... To be honest, that is one thing systemd does do fairly well. > > >>>> > > >>> > > >>> Because not everyone uses systemd, and we have to deal with the > > >>> "legacy" case too for the transition phase. > > >>> > > >>> It's generally preferable not to start up nfsd until everything it > > >>> needs is up. If we do what you suggest, then we're basically mandating > > >>> that this daemon can't start until nfsd is up and running. > > >> Order has ways been a part of how and when things are started which > > >> have always been handled by initscripts. That's their job, to start > > >> things in the correct order. > > >> > > >> I understand you want to make the daemon bullet proof... but starting > > >> things up in the wrong order is an error... IMHO... > > >> > > >>> > > >>> Could you give some details on how you think this ought to work? > > >>> > > >> I would think a error message stating unable to open whatever and then > > >> say something like please make sure the nfs server is up and running, > > >> would work... It seems to me this is a pretty common way of handling > > >> this type of situation.... although I can not come up with a > > >> explicit example, atm. > > >> > > > > > > That's doable simply by dropping this patch. I think it'll make this > > > more fragile, but if that's the consensus, I'll go along with it. > > > > > If that is the only fragile part then I you are in very good shape! ;-) > > > > Thanks again! > > Please patch section 3.1 of nfs-utils/README while you're at it. > > The start-up order *is* rather complicated, though, and easy to get > wrong. And we'd rather not make nfsd wait unnecessarily for this dameon > to start up.... > > Why not use directory notifications like idmapd and gssd always have? > I suppose I could do that. I'll take a look at what they do. Thanks, -- Jeff Layton <jlayton@xxxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
