Re: Re-init lockd and kstatd in runtime.

[NeilBrown <neilb@suse.de>]

[-- Attachment #1: Type: text/plain, Size: 6452 bytes --]

On Thu, 2 Feb 2012 04:50:06 +0000 "Vlasov, Vladimir" <vvlasov@mera.ru> wrote:

> Well, nobody deal with HA Linux clusters?

I certainly don't.

And like I said before - you are a SUSE customer and you are using a SUSE
product so your best best it to ask SUSE for help.

The in-kernel 'statd' which you are fighting with is a SUSE-specific
innovation (which I thankfully got rid of in later products) so no-one here
is going to be likely to help you with it.

NeilBrown


> 
> Best regards
> Vladimir Vlasov
> -----Original Message-----
> From: linux-nfs-owner@vger.kernel.org [mailto:linux-nfs-owner@vger.kernel.org] On Behalf Of Vlasov, Vladimir
> Sent: Friday, January 27, 2012 12:32 PM
> To: NeilBrown
> Cc: linux-nfs@vger.kernel.org
> Subject: RE: Re-init lockd and kstatd in runtime.
> 
> Finally I have found notes on HA NFS on SLES9 - http://www.gatsby.ucl.ac.uk/~johnp/compute/ha-nfs-sles9.html
> There are more info on this topic, for instance http://linux-ha.org/DRBD/NFS Suddenly I realized that previous designer was following these guidelines :-)
> 
> I have found that all documents suggest to have both primary and standby hosts configured as NFS server only and start/stop server processes when role is changed.
> This way /var/lib/nfs, even relocating on DRBD partition, is always available on NFS server startup.
> So, it is OK to have /var/lib/nfs as link to the folder on DRBD partition.
> 
> What I have different is that my system breaks rule of acting as NFS servers only: standby host acts as NFS client until it becomes primary, then it should act as server.
> 
> 
> I'll summarize info here:
> "server" is DNS record, resolved into IP of the primary host.
> 
> HOST1, primary:
> /dev/drbd is mounted as /storage
> /storage is exported
> server:/storage is mounted as /some/path /var/lib/nfs is link to the /some/path/var/lib/nfs and is available
> 
> HOST2:
> /dev/drbd is not mounted, /storage is empty server:/storage is mounted as /some/path /var/lib/nfs is link to the /some/path/var/lib/nfs and is empty
> 
> So, due to this /some/path mount existing on both hosts I have problem with initialization of the kstatd on the standby host, where /var/lib/nfs looks like empty folder.
> I have requirement for having /some/path on both hosts, so I can't fallback to guidelines advices.
> I need to do NFS client became NFS server somehow. There should be some runtime re-initialization. This is what I'm looking for.
> 
> Could someone comment on this?
> Is there any way to do so?
> Or any better solution?
> 
> By the way, browsing sources I have found that one can set nsm_use_kstatd to 1 on lockd exit, then initialization will be successful on next mount or export attempts.
> I think someone can find it useful so created patch. Unfortunately I don't have neither time nor enough code knowledge.
> In my case this proposal has drawback - lockd exits when all mounts are unmounted only, so I have to umount /some/path first and then mount it back.
> It can be painful as soon as some files can be opened by local user and umount will complain with it's usual "device is busy".
> 
> Also I was thinking about having /var/lin/nfs locally, separate for each host, so kstatd will be initialized always, but how do I share vital info about clients then?
> I have found it to be dead-end.
> 
> 
> Best regards
> Vladimir Vlasov
> 
> -----Original Message-----
> From: Vlasov, Vladimir
> Sent: Thursday, January 26, 2012 4:14 PM
> To: 'NeilBrown'
> Cc: linux-nfs@vger.kernel.org
> Subject: RE: Re-init lockd and kstatd in runtime.
> 
> Yes, it is SLES10 SP3 and I'm a customer.
> I'm not sure that Novell can help me, I expect the same proposal from them - "just make sure that /var/lib/nfs is available on startup".
> I'm not satisfied with this because I have some kind of high availability system built with two Linux hosts.
> 
> Hosts share DRBD partition and one acting as primary exports, for instance /some/path.
> As soon as primary host fails, another one becomes primary and exports /some/path.
> There are more hosts on the network, they mount:
> server:/some/path
> where "server" is resolved to the IP of current primary host which currently exports path.
> 
> So, in order to keep and overtake data about clients, /var/lib/nfs is replaced with link and points to /some/path/var/lib/nfs.
> If You don't familiar with DRBD then the most important thing You should know is that it is available for R/W on primary host only.
> So, on primary host DRBD is R/W, /var/lib/nfs is available and kstatd starts.
> On another host DRBD is NOT R/W, /var/lib/nfs is empry and kstatd doesn't start.
> This is why I'm looking for how do I re-init it at any time.
> 
> I was looking for guidelines of using NFS in clusters but didn't found anything useful.
> 
> Best regards
> Vladimir Vlasov
> 
> -----Original Message-----
> From: NeilBrown [mailto:neilb@suse.de]
> Sent: Thursday, January 26, 2012 3:41 PM
> To: Vlasov, Vladimir
> Cc: linux-nfs@vger.kernel.org
> Subject: Re: Re-init lockd and kstatd in runtime.
> 
> On Thu, 26 Jan 2012 11:15:02 +0000 "Vlasov, Vladimir" <vvlasov@mera.ru> wrote:
> 
> > Hi.
> > 
> > I'm wondering if there is any way to re-init lockd and kstatd after startup?
> > 
> > Look, in svc.c static variable nsm_use_kstatd is set to 1.
> > Then, if nsm_kernel_statd_init() returns error, nsm_use_kstatd is set to 0.
> > And we end up without kstatd.
> > 
> > Lockd will exit as soon as last NFS mount is unmounted and on next mount attempt lockd() in svc.c will be called again, but nsm_use_kstatd is 0 already and kstatd init will be skipped.
> > 
> > So, looks like if /var/lib/nfs/state is missed on host startup, there is no way to recover and tell kernel init kstatd later on demand. Or?
> 
> This would be SLES10 - correct?
> 
> I suggest you simply make sure /var/lib/nfs/state is not missing at startup.
> 
> The issue sounds vaguely familiar ... maybe it is fixed in SLES10-SP4...
> 
> What release are you running?  Are you a SLES customer?  If you are, you would be better off logging a support request on the SUSE bugzilla (bugzilla.novell.com).
> 
> NeilBrown
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at  http://vger.kernel.org/majordomo-info.html


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 828 bytes --]