Re: debian in a lab

1997-04-16 Thread meierrj
Christoph,

 ..., I usually install [packages] ... in a directory
 /usr/local/packages/package_name ...
 I use a script to move files into the correct directories.

 The appeal of such an approach is [ability to] use any packages ... without
 [installing] anything locally except a set of well-defined links[,]
 and without using up local storage or worrying about updates. ...
 this is a way of providing software to people without having to be root
 on their machines and thus without being responsible for their mistakes...

 I found that this works extremely well with most packages. Is anyone thinking
 of adding a concept like this to standard Debian as an option?

Yes!  Ray Ingles and myself are currently defining this concept and
would appreciate your help.  Please see our prior mail (sent 4/15/97)
Re: DEITY TEAM -- REQUEST FOR FUNCTIONALITY and COMMENTS.

Robert Meier

FANUC Robotics North America, Inc.  Internet: [EMAIL PROTECTED]
Voice: 1-810-377-7469   Fax:  1-810-377-7363


dselect deity and debian in the lab

1997-04-14 Thread Pedro I. Sanchez
Hello,

I'd like to suggest that the deity team take into consideration the
open problem of installing and maintaining packages accross NFS-mounted
volumes (see the recent thread debian in the lab in this list).

I believe that making deity NFS-aware will take the debian packaging
system one further step ahead of competition!

Cheers,

-- 
Pedro I. Sanchez
Product Manager
CTI Datacom Inc.
514.683.6363 x31


Re: debian in a lab

1997-04-13 Thread Craig Sanders

On 9 Apr 1997, Graeme Stewart wrote:

 The ideal situation, I think, would be for dselect to have an option
 where by it can be told that certain directories are NFS mounted.
 It should then do the installation as normal, but not copy files to
 these directories (or attempt to delete them upon uninstalling).
 Would that be hard to implement? I don't think so, though I never
 got farther than thinking... If one wanted to be really clever,
 one could get dselect on a client to download the server's list of
 installed packages and issue a warning if the client tried to install
 a different version of the package or a package that the server didn't
 have. I think that with those sort of changes Debian could be made
 very user friendly to NFS networks like ours.

that's a really good idea.  

what is needed is a dpkg --option or wrapper script which communicates
with a specified NFS server (ssh $SERVER dpkg --get-selections?) and
extracts it's list of installed packages. 


For packages installed on the server but not on the client, it should
install ONLY the config files  package.{pre,post}{rm,inst} scripts,
plus any files/directories NOT in NFS mounted directories (e.g. /bin,
/sbin, /var, /etc)

There should be a command line option or config file for specifying which
directories are NFS, and an exclusion list for those which aren't (to
allow for situations like /usr is NFS, /usr/local is not).  Alternatively,
dpkg can simply be told to ignore errors caused by a file or directory
being read-only. 

The {pre,post}inst scripts should be run at install time on the client
machine.


For packages which are on the client but not on the server, it should
run the {pre,post}rm scripts and delete files which are not on NFS
mounted disks.

this should be as automated as possible, so that when the server is
upgraded or new packages are installed it should be possible to use a for
loop wrapper around rsh or ssh or something to automatically update the
clientsbut that depends a lot on the scripts in the individual
packages.


craig


Re: debian in a lab

1997-04-10 Thread Christoph Best
Hi,

[EMAIL PROTECTED] (Kay Nettle) wrote:
   We're planning on installing debian in a lab of about 50 machines 
and were wondering if anyone had come up with a good way to install it over
the net.  We want to spend as little time on each machine as possible.  

We have been running about a dozen Linux PC as desktop workstations for
several years in an IBM RS/6000 AIX environment.
Back in 1993, we simply installed Slackware on a master PC,
tar'red and zip'ed the disk, put this on one of our workstation, and then
NFS-mounted it from a bootdisk and untar'ed onto every new PC. The tar-files
are about 60 Megs for a basic system, and the installation takes around 20
minutes. A script fixes local things like hostnames, IP addresses, and
the XFree86 configuration. (I could add here that for desktop work a 32 MB
Pentium outperforms any AIX desktop system we have here).

Recently, with the arrival of some Pentiums, I started using Debian packages.
To avoid putting all packages on each disk, I usually install them (except
for basic system packages) in a directory 
/usr/local/packages/package_name which typically looks like this:

lx601:/usr/local/packages/pine# ls -l
total 5
drwxr-xr-x   2 root root 1024 Mar 10 07:10 DEBIAN-pine_3.96L-0/
drwxr-xr-x   2 root root 1024 Mar 10 07:10 etc/
drwxr-xr-x   2 root root 1024 Mar 10 07:10 usr.bin/
drwxr-xr-x   2 root root 1024 Mar 10 07:10 usr.doc.pine/
drwxr-xr-x   2 root root 1024 Mar 10 07:10 usr.man.man1/

I use a script to move files into the correct directories.
In theory, I could then mount /usr/local/packages on
each client and have a script that automatically checks which packages
are installed and update the links locally. (In praxi, this is still done
manually.) I usually look at the preinst and postinst scripts and run
them by hand on the master machine. Anything they modify or create should
also be in the package directory.

The appeal of such an approach is that any Debian user on our net could
mount this directory and use any packages he wants without ever having to
install anything locally except a set of well-defined links. 
And without using up local storage or worrying about updates.
(If he needs different configurations, he can just replace a link in /etc 
by a real file.) Some people here are using similar schemes to provide
software for other platforms centrally. From an institutional point of view,
this is a way of providing software to people without having to be root
on their machines and thus without being responsible for their mistakes...

I found that this works extremely well with most packages. Is anyone thinking
of adding a concept like this to standard Debian as an option?

-Christoph

Christoph Best[EMAIL PROTECTED] 
Institute for Theoretical Physics
Johann Wolfgang Goethe University 60054 Frankfurt am Main, Germany
  Using TSO is like kicking a dead whale down the beach.




debian in a lab

1997-04-09 Thread Kay Nettle
Hi,

We're planning on installing debian in a lab of about 50 machines 
and were wondering if anyone had come up with a good way to install it over
the net.  We want to spend as little time on each machine as possible.  
Each machine will be on the net.  We've thought of a couple of ways
to do this, like: creating a template machine and then stuffing the 
disk image on all the others, or just using the normal debian install (but
we're worried we'd have to spend to much time on each machine).

There are a couple of other questions we have as well.  Has anyone nfs 
mounted /usr and if so how do you handle updates to packages?  Also, what 
is the process for creating local packages that override standard 
packages?
 

Thanks,
Kay


Re: debian in a lab

1997-04-09 Thread Jason Killen
It seems to me the best way to do this would be to use the ftp install
but that may take to much time so making a template machine and then
copying everything from that machine to all the others (cp -Rap) would
work.   If you can open the machines and get the drives out this wouldn't
be to bad, if not you would have to at least get a kernel and filesystem and
nis working first, unless you feel like ftping a 300 meg file across the room,
that has been done, if you are going to use nfs instead of ftp then why not
just install everything.  I had /usr nfs mounted for about 10 minutes one
time and it was slow I could run ls and go get a coke.  The best way to
keep up with what is in /usr would probally be nis.

Sorry for the rambling.

In message [EMAIL PROTECTED] you wrote:
Hi,

   We're planning on installing debian in a lab of about 50 machines 
and were wondering if anyone had come up with a good way to install it over
the net.  We want to spend as little time on each machine as possible.  
Each machine will be on the net.  We've thought of a couple of ways
to do this, like: creating a template machine and then stuffing the 
disk image on all the others, or just using the normal debian install (but
we're worried we'd have to spend to much time on each machine).

There are a couple of other questions we have as well.  Has anyone nfs 
mounted /usr and if so how do you handle updates to packages?  Also, what 
is the process for creating local packages that override standard 
packages?
 

Thanks,
Kay
--
Jason Killen Question Stupidity
Ma ma's don't let your babies grow up to be Linux hackers 
Monolith : the new ANSI standard for humans 
PGP fingerprint = 64 71 48 14 31 AE C6 70  E4 4F 64 EB 3B AA 00 6B
[EMAIL PROTECTED] 

 


Re: debian in a lab

1997-04-09 Thread Graeme Stewart
Hi Kay,

We currently have 5 Debian clients running with /usr, /home and bits
of /var (shared tfm and pk font files). The system runs really well
over our not too busy 10Mbps ethernet. This may be because we have a
good server, PPro 200MHz with 64MB and a fast-wide SCSI disk, so your
milage might vary if your server's not so good (memory and disk are
probably more important than CPU here).

As to package control, I have to confess we did everything by hand,
first getting a workable small installation on each local machine
(about 100MB), then mounting /usr et al. over the top of this. At the
moment our clients then have a local /usr with the most basic packages
installed, but, in normal use, hidden under the NFS /usr. This is an
inefficient use of disk space, I know, but does mean that the machines
are usable in case of server failure, and it also simplifies upgrades
with dpkg when necessary. Most things work without any trouble then,
although on one very small disk machine I had to hand copy a few
config files across (e.g., /etc/lynx.cfg). It is a pain to upgrade: we
have to umount all network disks, run dselect, remount the network.

The ideal situation, I think, would be for dselect to have an option
where by it can be told that certain directories are NFS mounted. It
should then do the installation as normal, but not copy files to these
directories (or attempt to delete them upon uninstalling). Would that
be hard to implement? I don't think so, though I never got farther
than thinking... If one wanted to be really clever, one could get
dselect on a client to download the server's list of installed
packages and issue a warning if the client tried to install a
different version of the package or a package that the server didn't
have. I think that with those sort of changes Debian could be made
very user friendly to NFS networks like ours.

If I were in your shoes I'd test with one client, until everything
works well, then export the package list and run dselect on each
machine out of a script (mounting the packages NFS, of course!).

I'd be interested to hear how it goes, because I think that our setup
is a bit of a munge, even if it does work.

Cheers,
Graeme

-- 
| Graeme A Stewart, pgp public key  finger [EMAIL PROTECTED]|
|  Key fingerprint =  AF C7 BF A4 52 D5 3C 3B  17 A5 62 43 DA 15 E8 97  |
|   Keep a good head, and always carry a lightbulb.   Dylan   |