Coda File System

Re: Creating a replicated volume

From: Jeremy Bowen <jeremyb_at_iserve.co.nz>
Date: Tue, 13 Sep 2005 09:26:57 +1200
Hi Ivan

On Mon, 12 Sep 2005 9:06 pm, Ivan Popov wrote:
> Volumes are connected into the file name space by so called "mount point"
> objects. You create a mount point by
>
>  cfs mkm /coda/realm/path
>
> which connect a volume named "/path" (if such one exists) at
> /coda/realm/path
>
> To keep the things simple you should synchronize
> volume names and the corresponding mount point path names.
> Fortunately, it is the default for "cfs mkm".
>
> Well, nothing prevents you from deliberately connect an arbitrary volume to
> an arbitrary place in the file tree by doing
>
>  cfs mkm /coda/realm/path anyvolumename
>
> but I would strongly advice you against that - this possibility is useful
> only in rare cases for troubleshooting or alike, or when you paths become
> too long (volume names are currently limited to 32 chars).
>
> A volume named "/test" becomes visible as soon as you
> create the corresponding mountpoint with
>
>  cfs mkm /coda/mrblue.localdomain/test
>
> and becomes invisible if you happen to remove the mount point:
>
>  cfs rmm /coda/mrblue.localdomain/test
>
> > I'm getting horribly confused here. What does the <volname> argument to
> > createvol_rep relate to ? How do I specify this volume on the client
> > system ?
>
> In the underlying implementation - due to AFS tradition,
> the volume names and file names namespaces are totally unrelated,
> so a server needs a hint when it should do a lookup for a certain volume.
>
> We can not yet break that tradition and make the one-to-one mapping
> implicit - as there are some cases where we still have to connect volumes
> arbitrarily. Possibly we could avoid those corner cases later by careful
> (re)design, but it would need some extra effort.
>
> At the very least we would have to implement arbitrary length volume names
> and make a decision about where and how backup volumes shall appear
> in the file name space.
> Volume management would have to be done differently, and done
> from clients, not from servers as it is now.
>
> Otherwise I feel that your expectation (a volume being visible
> as soon as it is created) is a very reasonable one - though incompatible
> with the existing implementation.
>
> Compared to AFS we have made a step in the right direction by the implicit
> volume naming in "cfs mkm", may be some day we can go further.
>
> Best regards,
> --
> Ivan
Received on 2005-09-12 17:28:32