On Mon, Aug 05, 2013 at 07:38:59PM +0300, Timo Sirainen wrote:
The NFS workarounds code is doing some ugly stuff. I thought it would have, but looking at the code it doesn't seem so. But still easier to debug if you first see if the problem is with the NFS workarounds or the lib-index code. With lib-index you could also use lock_method=dotlock to see if that works better (although performance will be slightly worse also then).
I just tested mail_nfs_storage and mail_nfs_index both set to no : log messages are still here. They only stop when I stop dovecot.
Note :
The nfs mount is made through a private network (192.168.3.x). My simple fcntl unlock test, when the isilon mount is made through the same private network behave as dovecot (i.e. make the node log warning about non-existent ressource to unlock), but not when the mount is made through the "public" network (157.99.x.x). Even when dovecot is down and statd/lockd restarted.
I thought about some reverse name lookups but I don't think this is the problem.
Something very strange is happening, not matter what nfs workarounds dovecot is making. But it seems to involve this host and only this host...
-- Thomas Hummel | Institut Pasteur <hummel@pasteur.fr> | Groupe Exploitation et Infrastructure