hg repositories and Windows DFS Replication

[Matt Mackall]

[replying back to list]

On Tue, 2011-05-31 at 21:56 -0500, Robert Altman wrote:
> The only problems I have seen with DFS are when copying large files (>100's
> MB).  If you try moving or renaming a large files while it is being
> replicated, the operation may fail with file-in-use error.

Oh. That's quite unfortunate. I guess that answers the "is DFS in the
same class with GFS and Lustre" with a resounding no. 

If I had to guess, I would say that the above problem is encountered
with a probability proportional to size, and you will eventually
encounter it on a small file given enough time/use.

> Most version control files are small, so this seems unlikely to be an issue
> but I like to be certain when it comes to my own repositories.
> 
> How would hg handle this?

If Mercurial encounters an filesystem error, it aborts. Filesystem
operations are supposed to be maximally reliable, not a best-effort
try-again sort of affair; if the operating system gives up, Mercurial
assumes the situation is hopeless.

You may be in luck here though: the one exception Mercurial makes is for
deletion of files that are being renamed on Windows, because the
semantics here on Windows are a bit magical.

> 
> 
> On Tue, May 31, 2011 at 9:08 PM, Matt Mackall <mpm at selenic.com> wrote:
> 
> > On Tue, 2011-05-31 at 20:19 -0500, Robert Altman wrote:
> > > Has anyone ever used mercurial on Windows servers with DFS replicating
> > the
> > > repositories to another server.  (This would essentially server to
> > "mirror"
> > > the repositories to another computer.)
> > >
> > > As long as the changes are only made on one server (the web server), is
> > > there any risk to the data itself?  (I wouldn't think so, but I'm curious
> > if
> > > anyone has tried it.)
> >
> > First, Mercurial repos are just files; there's no intangible spirit that
> > needs to be copied. The only magic you need to be aware of is that
> > copies made during a write may not be completely valid. This generally
> > isn't an issue for backup scenarios.
> >
> > I'm not sure how DFS compares to industrial-strength distributed
> > filesystems like GFS and Lustre, but if it provides similar coherency
> > guarantees, it should be more or less invisible to Mercurial, even with
> > multiple writers. But given the reliability we've observed over CIFS,
> > I'd try to keep things simple.
> >
> > --
> > Mathematics is the supreme nostalgia of our time.
> >
> >
> >


-- 
Mathematics is the supreme nostalgia of our time.