[Gluster-devel] Strange inconsistencies (No such file or directory)

[Kevan Benson]

Brian Taber wrote:
> Ok, I tried the debug once before, must have missed it.  I did see dome
> errors on the client side, but not the server side.  Here is the log
> excert:
>
> 2007-10-22 00:19:07 D [inode.c:351:__active_inode] fuse/inode: activating
> inode(16750137), lru=1/1024
> 2007-10-22 00:19:07 D [fuse-bridge.c:423:fuse_lookup] glusterfs-fuse:
> LOOKUP 16750137/1193026747.V19Iffa3a9M917884.anubis.mydomain.com
> (/mydomain.com/webmaster/new/1193026747.V19Iffa3a9M917884.anubis.mydomain.com)
> 2007-10-22 00:19:07 D [fuse-bridge.c:378:fuse_entry_cbk] glusterfs-fuse:
> ERR => -1 (2)
> 2007-10-22 00:19:07 D [inode.c:308:__destroy_inode] fuse/inode: destroy
> inode(0) [@0x7bbe4a0]
> 2007-10-22 00:19:07 D [inode.c:381:__passive_inode] fuse/inode:
> passivating inode(16750137), lru=2/1024
> 2007-10-22 00:19:07 D [inode.c:351:__active_inode] fuse/inode: activating
> inode(16753577), lru=1/1024
> 2007-10-22 00:19:07 D [inode.c:351:__active_inode] fuse/inode: activating
> inode(16750137), lru=0/1024
> 2007-10-22 00:19:07 D [fuse-bridge.c:1052:fuse_link] glusterfs-fuse: LINK
> /mydomain.com/webmaster/tmp/1193026747.P24588.anubis.mydomain.com
> /mydomain.com/webmaster/new/1193026747.V19Iffa3a9M917884.anubis.mydomain.com
> 2007-10-22 00:19:07 D [fuse-bridge.c:346:fuse_entry_cbk] glusterfs-fuse:
> ENTRY => 16753577
> 2007-10-22 00:19:07 D [inode.c:381:__passive_inode] fuse/inode:
> passivating inode(16753577), lru=1/1024
> 2007-10-22 00:19:07 D [fuse-bridge.c:917:fuse_unlink] glusterfs-fuse:
> UNLINK 16750136/1193026747.P24588.anubis.mydomain.com
>
> The client and server run fine without crashing, so I really couldn't
> backtrace it (or could I?)
>
> I tried to find the tla release you mentioned, I am not sure which one to
> use, do you have a direct link I could use?
>   

I don't see any errors there (the "D" after the time stamp denotes it's 
a debug level message.  It looks like a normal maildir operation to me.

Question:  In the order of operation you described for the original 
problem, where server 1 is running postfix and can't access the message 
after a while, what exactly is failing, what's causing the error?  I 
don't see how it could be postfix, as postfix's responsibility ends at 
the delivery of the message into the maildir store.  Is there a IMAP 
client running on this server?  I assume there's an IMAP client running 
on server 2, the webmail server.

It's normal in maildir folders for messages to be moved around from tmp/ 
to cur/ and new/, and in this case it's using hard links for atomic 
operations (as the LINK entry in fuse shows).  I can only think of 2 
things that might be happening to cause this (but that's just me), 
either glusterfs has a problem with hard links (which I'm not seeing on 
my end, but I doubt I'm replicating what your IMAP client does exactly), 
or one of the servers still thinks it will see a file in the new/ dir of 
a maildir root after it's been moved to cur/ (that is, it's been 
viewed), which could either be an application problem or a glusterfs 
caching issue (if any of the translators you use can cache directory 
data, which I'm fuzzy on).

-- 

-Kevan Benson
-A-1 Networks