[OpenAFS] Unexpected behavior when renaming volumes

Bastian dea1306@melvex.xs4all.nl
Thu, 26 Apr 2007 23:48:44 +0200


Jeffrey Altman schreef:
> Bastian wrote:
>   
>> Hello All,
>>
>> I have a strange observation. If I rename a volume and I create another
>> volume using the old name, the two volumes have the same contents.
>>
>> Changes to one volume are reflected in both volumes immediately. Though
>> the volumes do not have the same ID, it looks like they are one and the
>> same. I repeated the action several times with the same effect. Deleting
>> and recreating both volumes solved the problem.
>>
>> Is this the intended behavior or is there something wrong?
>>
>> Bastian
>>
>> (More information: I am using OpenAFS 1.4.3 on Debian 4.0; all the vldb
>> servers and all the servers with copies of the volumes are up. After
>> each action I release the volumes)
>>     
>
> Clients find a volume by evaluating a mount point containing the volume
> name, looking up that name in the vldb to get its ID number, and then
> making all further references to it by number.  If you rename the volume
> clients that have already mapped the old name to the number will
> continue to access the same volume.
>
> When you create the new name and force the client to evaluate that name,
> it will receive the same volume ID it saw before.  Hence, both the old
> and new names will refer to the same volume on that client until the
> cache is flushed and the mount points are forced to be re-evaluated.
>
> In other words, renaming a volume does not cause the cache contents of
> the clients to be flushed automatically.
>
> Jeffrey Altman
> Secure Endpoints Inc.
>   
Thanks. That makes sense. I suppose I'd better flush the client cache a 
few times when I do rename actions like this.

Bastian