Coda File System

Re: Client/server inconsistency problem

From: Andreas J. Koenig <andreas.koenig_at_anima.de>
Date: 12 Jul 1999 14:57:40 +0200
>>>>> On Mon, 12 Jul 1999 08:31:25 -0400, Jan Harkes <jaharkes_at_cs.cmu.edu> said:

 > On Sat, Jul 10, 1999 at 06:11:15PM +0200, Andreas J. Koenig wrote:
 > ...
>> But I get the following error message:
>> 
>> ./usr/apache/lib/map_box/RCS/drawerOK.pm,v: No such file or directory
 > ... 
>> What kind of inconsistency am I facing here? How can I resolve the conflict?

 > Most likely an inode collision, 2 FIDs hash onto the same inode. When you
 > try to access the file, the kernel module should log some error about

 >     coda_cnode_make on initialized inode <xxx>, old <fid1> new <fid2>

 > We currently return ENOENT when that happens, I have been working on some
 > big changes that take the knowledge about fids out of the kernel, to avoid
 > this and some other problems during reintegration.

 > It is difficult to fix this. Normally I try to find which file/directory the
 > `old' fid represents, copy it to a new name, remove the old one and move the
 > copy back. This results in a new fid, and inode, for the old file and the
 > other one then becomes visible/accessible again.

I'm sorry, I had intended to wait till I have an answer, but then I
became too nervous and resolved the conflict with brute force. I
tarred the whole directory on the server, removed it and rebuilt it on
the client. That did work. In effect this strategy is pretty much the
same as yours, just more expensive.

Anyway, I'm glad to know that an occasional rsync for diagnosing
problems is called for. Thanks!

-- 
andreas
Received on 1999-07-12 08:56:50