This is a slight detour, but I could not resist.
On Sunday, May 14, "Andy Glew" wrote:
>
> (1) I see that you are using CVS and AFS.
> Many readers of this list recommend using
> CVS server mode, rather than upon a network
> filesystem. I must admit that I think that is
> bogus - else what's a network filesystem for?
> - but there you have it.
IFF the network filesystem implemented the right filesystem
semantics correctly, I'd agree with you. However, for
efficiency reasons, I beg to differ. You meaning to tell me
that running an Oracle DB in this sort of distributed fashion
should work well? There are any number of reasons that CVS
system should not use network distributed filesystems. The
best I can think of, is to make the CVS "protocol" the defacto
"standard" way of interoperating with CVS, thereby decoupling
the repository from the client, enabling the possibilities of
actually doing some further development and implementation of
various features...
> (2) Are you, by any chance, using LINUX?
Ahh, the anti-unix. :-) <-- Notice the *smiley* people...
> I have been having problems left, right, and center
> with the combination of CVS, LINUX, and AFS.
> My sysadmins tell me that the LINUX AFS client
> has problems, doing things like returning incorrect
> status for directory queries.
The beauty of blaming the wrong piece of software is that it
is so convenient. I'm not 100% sure, but I thought that rcs/cvs
specified the type of filesystem semantics they assumed to be
working under/over/with. If these semantics are in error,
in whichever way that may be, then you can't really expect cvs
to its job correctly.
> I suspect that LINUX/AFS is also associated with
> broken timestamps, updates that don't really work, etc.
> "Suspect" because I haven't isolated the problem, but
> I have been having all sorts of issues since I started
> using CVS to do builds on LINUX.
On LINUX, or LINUX/AFS? I have my own pet-peeve with how linux
goes about ignoring various standards, and implements its own
semantics with various system calls. However, this is not the
forum to get into this. If it is LINUX/AFS you are having
trouble with, then I would suggest using client-server, and if
necessary, local filesystems on both ends (the server anyhow)
to do your work in. (local filesystems tend to be faster to
compile/work with anyhow).
> I am about to declare myself a personal rule:
> not to use CVS on AFS on LINUX, at least until a
> new rev of the AFS client for LINUX comes out.
> Only to use CVS on AFS on Solaris/x86.
One place I worked for had CVS performance trouble. They were
using it over NFS, client-server, and a number of other ways.
After cleaning things up, going completely client-server, and
using local disk for both the repo, and the client-side, they
saw at *least* a 10-fold improvement. (Ok, there was a server
upgrade in there as well...) My "rule" with CVS has always been:
Use client-server, use local disk.
--Toby.
