[lug] CVS temporary files

[Kevin Fenzi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

>>>>> "Jeff" == Jeff Schroeder <jeff at neobox.net> writes:

Jeff> Hey all-- As a longtime convert to CVS, I recommend it to many
Jeff> developers for software projects.  I helped a small company set
Jeff> it up on an in-house server, and they keep seeing problems when
Jeff> many people are making simultaneous updates.  Specifically, they
Jeff> get an error message like this:

Jeff> cannot create_adm_p /tmp/cvs-serv712/project No space left on
Jeff> device

Jeff> When I do a 'cvs checkout project' on my client machine, I see
Jeff> the CVS server creating a /tmp/cvs-serv[nnn[ directory and then
Jeff> filling it with a directory tree that mirrors the project.
Jeff> Inside each directory is a handful of text files (Entries,
Jeff> Entries.Log, Repository, and Root), none of which are more than
Jeff> a few kilobytes in size.  All told, a full checkout might
Jeff> presumably consume a megabyte of temporary space doing this.  As
Jeff> soon as the checkout is complete (or interrupted), the server
Jeff> wipes the tree.

Jeff> I have /tmp mounted as tmpfs with 128MB allocated to it-- in
Jeff> other words, the /tmp partition can't grow larger than 128MB.
Jeff> There are perhaps a dozen developers on the project, and it's
Jeff> highly doubtful that every one of them is doing a full checkout
Jeff> simultaneously.  Even if they were, I would expect no more than
Jeff> 15-20MB of space to be needed by the CVS server.

Jeff> I looked at the documentation and found a "LockDir" item, but
Jeff> that appears to control where CVS stores a record of what
Jeff> directory it's currently processing (so someone doesn't update a
Jeff> file in a directory that's being checked out).  And that takes
Jeff> only a handful of bytes, so I don't think it's the problem.

Jeff> Has anyone seen this kind of behavior, or have suggestions about
Jeff> how to avoid it?  I've never personally seen the error message
Jeff> above, but several fairly savvy developers have reported it.
Jeff> They're all using a Mac OS X client, whereas I'm on a Linux box.
Jeff> I can't imagine it makes that much difference, though.

How many inodes are available on your tmpfs /tmp?

df -i /tmp 

?
If it's running out of inodes for new files it will report that it's
out of disk space as well. (Or at least it used to). 

Jeff> Thanks, Jeff

kevin
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.6 (GNU/Linux)
Comment: Processed by Mailcrypt 3.5.8 <http://mailcrypt.sourceforge.net/>

iD8DBQFBooxy3imCezTjY0ERAjGZAJ9z2GErqBr9Tt6OO9XEoIkR7KLLcwCgjv2z
iK+wrKSj+rKh81PzRIxwgcU=
=hiro
-----END PGP SIGNATURE-----