>Path: fnnews.fnal.gov!pogo!hanson
>From: [EMAIL PROTECTED] (Steve Hanson)
>Newsgroups: ssc.info.afs
>Subject: Batch processing under AFS
>Date: 29 Apr 93 16:46:50 GMT
>Organization: FERMILAB, Batavia, IL
>Lines: 49
>Message-ID: <hanson.736102010@pogo>
>Reply-To: [EMAIL PROTECTED]
>NNTP-Posting-Host: pogo.fnal.gov
>Keywords: BATCH AFS
Hi--
At FERMILAB we are currently being faced with the issue of getting
a number of batch processing systems running in an AFS environment.
This is proving not to be easy and I was wondering if anyone could
provide any hints, pointers, help, or aspirin, which I seem to be
using a lot of lately.
Anyway - the problem is that given a system in which there
is a master daemon that runs all the time, and which spawns off batch
processes according to its own scheduler, does anyone have a way to
get these child processes authenticated as the user who scheduled
the process in the first place? The system will run the process under
the UID of the person who scheduled it, but of course that's not
any help.
Several possible plans come to mind to get this working, none
of which are really satisfying:
1. Start up the daemon and keep it authenticated using reauth or
some such. Make it run as a master AFSbatch user. All output
needs to be open to this user. This stinks. Everyone could scribble
all over each other's files.
2. Run all the batch stuff outside of the AFS file system. Possible,
but hardly in the spirit of what we're trying to accomplish.
3. Try to figure out some way to authenticate the child processes.
Well, this is what we'd like but I don't see any way to do this
cleanly and in a secure manner, particularly since we don't have source
for the batch processor.
Off hand there's no elegant solution. Surely someone else is dealing
with this sort of thing.
We are also beginning to evaluate a batch system that IBM will be
marketing that is based on Condor. They tell us that the
release we have is "AFS aware" but so far no one has seemed to be
able to tell us what they mean by that. And we aren't far enough
into the process to have figured it out ourselves.
Anyway, although I believe this topic has been discussed in the
past, if anyone has any enlightenment to shed on me, please let me know.
--
Steve Hanson - FERMILAB, Batavia, Il.
[EMAIL PROTECTED] or [EMAIL PROTECTED]
