Coda File System

Re: SMB / Coda / venus hangs Additional Informations

From: Jan Harkes <jaharkes_at_cs.cmu.edu>
Date: Thu, 14 Aug 2003 09:39:05 -0400
On Thu, Aug 14, 2003 at 10:27:33AM +0200, Tom Weber wrote:
>   found some new facts to the problem posted yesterday on the list:
> 
>   each time one of the client crashes, it normally reports in codacon:
>     09:20:32 Checkpointing
>     09:20:32 to /usr/coda/spool/501/coda_.tar
>     09:20:32 and /usr/coda/spool/501/coda.cml
>     09:21:15 Reintegrate  pending tokens for uid = 501
>   where uid 501 is the httpd on the clients
> 
>   when the system hangs "cfs checkservers" reports :
>     Contacting servers .....
>     VIOCCKSERV: No such device or address
> 
>   after a venus -init the system is up running again

Ahh, unauthenticated users are never allowed to reintegrate. So as long
as your Coda client has a good connection write operations will succeed.
But as soon as the network, or the server, causes lost packets or slow
responses the client adapts and switches to writeback-caching
(write-disconnected).

It records operations in a local log (CML). Reintegrating the CML is
much more efficient because we can do up to 100 operations in a single
server-side transaction and in many cases operations can even be
optimized away (create/[store/]*remove -> nothing, or store[/store]* ->
last store, etc.)

However the reintegration log is associated with the local user-id and
is never sent to the servers for unauthenticated users. The client
doesn't really know the ACL (or how to interpret it), and if the user
wouldn't have permission to actually perform the update any attempt to
do so would automatically lead to a conflict. By giving the user a
chance to authenticate first we avoid conflicts that have to be repaired
by hand.

>   I also found sometimes this error near the time the system crashed:
> 
>     Aug 13 10:17:17 blade2 kernel: Coda: Bye bye.

Venus is shut down.

>     Aug 13 10:17:26 blade2 kernel: coda_read_super: device index: 0
>     Aug 13 10:17:26 blade2 kernel: coda_read_super: rootfid is (0xff000001.0x1.0x1)
>     Aug 13 10:17:26 blade2 kernel: coda_read_super: rootinode is 1367858057 dev 9

Venus is restarted. Nothing serious.

Jan
Received on 2003-08-14 09:41:55