klm@gozer.UUCP (Kevin [My Amiga has e-mail] McBride) (12/07/89)
Has anybody ever seen anything like this before?
A client of mine is using Interactive 386/ix and has been having some
very werid problems. I've been using 386/ix regularly for almost a
year now and have never seen anything quite like it.
Early weirdness in my client's system(s) I believe to be caused by
yellow pages (which they just received a couple of weeks ago) The yp
installation was done correctly, but it is just really flaky. Machines
would stop recognizing each other's existence even though NFS continued
to function normally. Anyway, I ripped out yp and some of the problems
went away.
The really weird thing is the c-shell's behavior on NFS client machines.
For example:
On the machine I am doing development on, my home directory lives
on a remotely mounted NFS volume. i.e., my home directory is
something:/usr2/klm. If I cd down into one of my subdirectories:
cd ~/src/proto
and run, say an emacs or a vi, then do a shell escape from that
program, the new child shell is not in the directory I started
from. It is one directory higher. If I do this from /usr2/klm,
then the child shell's current directory will be /usr2.
Needless to say, this confuses the hell out of emacs. When I do a M-x
compile, the compilation process runs make in the wrong directory. I
know this is not an emacs bug because I was running the same copy of
the binary on a different machine at a different site last week and
everything worked just fine.
After further investigation I discovered that typing "exec csh" to the
c-shell would also cause the newly execed shell to be in the wrong
directory. Also, the cwd shell variable would be empty in the new
shell. (It exists but contains a null string.)
I have tried identical scenarios on both the NFS client and on the
server. The system behaves as expected on the server (not using NFS to
access the file system)
So, finally, my question: Does anybody know what's going on here?
Is NFS screwing us up?
How does the shell pass it's variables (not environment variables,
shell variables) from one process to another, or does it? If the
kernel ran out of some particular resource, would that cause the shell
to lose it's mind?
I tried upping a bunch of tunable parameters (NBUF, NBLKxxx,
semaphores, files, etc...) and got some improvement. My login shell
would create new shells that wouldn't get lost, but other shells
(running under xterms) would still be confused.
Both my client and I are in a real quandary here. I'll be calling
Interactive support tomorrow, but I don't expect much help given the
weirdness of this problem.
Anybody out there (Interactive developers?) care to take a stab at it?
AdThanksvance.
--
--
Kevin L. McBride, President // Amiga: | Brewmeister, VP of tasting,
McBride Software // The computer | and Bottle Washer,
Consulting Group, Inc. \\ // for the | McBeer Home Brewery
uunet!wang!gozer!klm \x/ creative mind | Nashua, NH