RE: CVS and AFS

[Ian Gilmour]

> -----Original Message-----
> From: Todd Denniston [mailto:address@hidden
> Sent: Friday, June 22, 2001 3:48 PM
> To: Michael Sperber [Mr. Preprocessor]
> Cc: address@hidden
> Subject: Re: CVS and AFS
> 
> 
> "Michael Sperber [Mr. Preprocessor]" wrote:
> > 
> > >>>>> "Gerhard" == Gerhard Sittig <address@hidden> writes:
> > 
> > Gerhard> On Thu, Jun 14, 2001 at 22:50 -0400, Charles Karney wrote:
> > >>
> > >> We use CVS in a mixed Windows and Linux environment.  Recently
> > >> we switched from accessing the CVS repository vis ssh to a
> > >> Linux machine on which CVSROOT was a local disk to having the
> 
> \/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
> > >> CVS repository in AFS and having all clients access this as a
> > >> local file system.
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> > 
> > Gerhard> You did read the few thousand horror stories of 
> what's going to
> > Gerhard> happen when accessing repos via networked 
> filesystems in the
> > Gerhard> list, didn't you?
> > 
> > ... because by far most of them don't apply to AFS.  If 
> you'd read the
> > report you'd seen that most of the "few thousand horror" 
> stories have
> > no relation to what Charles was asking about.
> > 
> although they do not sound like AFS because they usually say 
> NFS or SMB, they
> are exactly the case described by Charles!
> 
> Telling CVS on each machine that it accessing the repository 
> on a physical
> (local) drive so that all lock files are being cached in the 
> local machines
> File System cache.  However the case is that the lock files 
> are being cached
> locally, network latency added, and finally being put in the actually
> distributing computer's cache to be checked by other cvs 
> clients.  This allows 
> a race condition to exist where one client may not know 
> another has a lock and
> so it can (insert bad thing) which corrupts or causes other 
> problems with the
> repository.
> 
> Basic summary:
> If the repository is being accessed as a local File system, 
> then the repository
> should be on a physical disk of the computer, otherwise you 
> have different
> clients in a race condition.
> 
> Real use:
> If the repository is an network file system (NFS/SMB/AFS) 
> mount then it can be
> used as a local file system IF and ONLY IF you are using a 
> remote protocol
> (RSH/SSH/pserver) to connect to the single computer which 
> accesses the repo as
> a local file system, there by making all locks in only one 
> computer's cache.
> (Some even indicate this could be bad, but I have YET to 
> experience a problem
> with it. YMMV)
> 

Can someone explain why this "Real use:" is potentially bad. Because I was
thinking of doing it this way myself in the near future (single cvs pserver
m/c + NFS archive).

cheers,

Ian