Coda File System

Re: write failure issues

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Wed, 19 May 2004 00:35:03 -0400
On Tue, May 18, 2004 at 06:25:47PM -0400, Greg Troxel wrote:
> venus can refuse to go connected due to conflicts, and it seems
> various other issues.  It may well be logging why (or you can turn up
> debugging logs), but I think it would be helpful for cfs lv to report
> the reason why venus is not in connected mode.  You are not the first
> person to have issues here.

Correct. Well, actually it is clear why venus isn't in connected mode
here. There are CML entries that haven't been reintegrated yet. The
problem is ofcourse why these are not being reintegrated even after all
the attempts to tell venus to do so with cfs writereconnect and
forcereintegrate.

And one of the possible reasons can be that we failed to get an
exclusive lock on the volume, which means that we can't even safely
update an error message in the volume to indicate that we failed to
lock. Another reintegration thread could be trying to reintegrate in the
background, or we might be working on server-server resolution, or a
mutating RPC call is simply taking a very long time, all of these cases
prevent us from exclusive locking the volume and makes things seem to be
stuck.

Yes, we could add messages for some of the trivial problems, like a
reintegration conflict, or when the first entry in the CML hasn't aged
enough to be considered for reintegration. But I have a suspicion that
the volume lock is the problem here as there aren't too many possibly
errors that can occur in the VIOC_STATUSWB ioctl code with returns the
"Writeback is enabled/disabled" message and most of them seem to be
related to either getting the pointer to the volume or while locking
the volume.

Jan
Received on 2004-05-19 00:36:09