[Dovecot] Unlock non existent locks

[Thomas Hummel]

On Fri, Aug 02, 2013 at 03:38:47PM +0300, Timo Sirainen wrote:

> Since you have only one Dovecot accessing the NFS, you don't need either
> mail_nfs_storage=yes or mail_nfs_index=yes. My guess is that by setting
> those to "no", you'll also solve this:
> 
> > 2013-08-02T14:12:29+02:00 <0.5> XXXX-10(id10) /boot/kernel.amd64/kernel: [lkf_delegate.c:2752](pid 46390="kt: dwt3")(tid=101282) dev_local_lkf_unlock(): no lock entry present to unlock for resource: 1:19d5:fdbe ;client: 0xa51cc3f444107

Thanks, I'll try that but that wouldn't be a good solution for when I'd want to
use more dovecot servers anyway.

My tests using a simple fcntl() C program (unlock a non-locked file) is giving
me a hard time figuring out what conditions cause the file server to log this
message :

On some Isilon node/destination IP combinations, I'd have the message, on other
I wouldn't.

Still, it seems I cannot reproduce the problem on any other hosts (mounting
this server) than the one running the dovecot server. Oddly enough, my fcntl
test causes this message even with dovecot stopped, statd and lockd restarted
and the filesystem un-and-re-mounted while the same setup (up to date via
FreeBSD update, i.e. same base, same nfs client) would not make the server log
the message.

Anyway, the problem seems harmless. But is it legit that dovecot try to unlock
non (or no more) locked files as it seems ?

Thanks for your help.

-- 
Thomas Hummel 	    | Institut Pasteur
<hummel at pasteur.fr> | Groupe Exploitation et Infrastructure