/usr in NFS

cvelasquez · n00b Joined: 03 Feb 2007 Posts: 9

Hi!, I want to make an installation using only strictly needed disk space. Thus, I want to have one /usr directory in a server and then mount it in the clients (instead of only /usr/portage, for example). That'd really save me a lot of disk space, but it's not that easy, is it? If I do that then I'd have to mount almost everything in NFS (/etc, /opt, etc), and that'd be a diskless installation (a solution which doesn't fit my network installation). Am I only imagining that is more complicated that mounting /usr? If it isn't the case, can someone help me designing and ejecuting? Please tell me if it's possible without falling back to rootless installations (pls note that what I want to do is to share files, not processing power with a XDMCP server, for example).

no_hope · Guru Joined: 23 Jun 2003 Posts: 482

cvelasquez · n00b Joined: 03 Feb 2007 Posts: 9

no_hope · Guru Joined: 23 Jun 2003 Posts: 482

If you just want to save some space, I don't see why just adding an fstab entry is insufficient. Though you might want to be careful with AUTOCLEAN.

If you want all the involved machines to mirror each other configuration-wise (e.g. you want packages emerged on one machine to automatically become available on the rest), then yes, you have to mount pretty much everything or set up a bunch of scripts to sync files. The complexity of the script solution depends on details of what you want to do.

In any of these cases, diskless installation is not necessary.

I get a feeling I'm not helping. Do you have a specific concern you worry about? With an example preferab

cvelasquez · n00b Joined: 03 Feb 2007 Posts: 9

embobo · Guru Joined: 19 May 2003 Posts: 311

Mounting /usr won't work unless you guarantee all systems have the same packages and the same use flags. That would be hard to do because you couldn't emerge the packages on the clients.

Disk is cheap. Buy some more.

no_hope · Guru Joined: 23 Jun 2003 Posts: 482

let me try to give one more shot

I think what you really want to do is just image your nodes, as portage isn't meant for distributed package management. Taking and deploying an image can be done in under 20 minutes, but that means you would need to have a master node.

If you want all nodes to be equal and mirror each other (i.e. emerge on any machine effectively installs a package on all machines), I would recommend going diskless. Though you don't have to, I think that would make things easier, as I don't think you will gain much from nodes being able to boot on their own.

If you want to do the latter and mount as few directories as possible without going diskless.

bin -- nfs mount
boot -- local
dev -- local
etc -- nfs mount, note that by definition etc contains host-specific config files. you may have to play tricks if you want that dir shared. Whether you will run into difficulties depends on your setup. If all machines have identical hardware you should be fine.
home -- depends, probably nfs mount
lib -- local (*)
media -- local
mnt -- local
opt -- local
proc -- local
root -- local
sbin -- local (*)
sys -- local
tmp -- local
usr -- nfs mount
var -- local, assuming you only emerge packages on one master machine. Otherwise, nfs mount /var/db /var/lib/portage. Depending on your setup you might want to nfs mount more directories in /var/lib/. You could mount /var/lib but then you will run into problems because init scripts will see state from other machines.

* - depending on your setup you might have mount these or make links to files inside on the nfs server.

cvelasquez · n00b Joined: 03 Feb 2007 Posts: 9

thx for the answers

embobo, I also want to decrease mantaining time, not only disk space.

nohope, I want the latter, but I think it's too "dirty", I want it to be consistent.

Can't I instead mount the master / directory and then modify the PATH? Or It'd need a chroot?

no_hope · Guru Joined: 23 Jun 2003 Posts: 482

cvelasquez · n00b Joined: 03 Feb 2007 Posts: 9

thx nohop, but what is imaging, and how can I use it?

no_hope · Guru Joined: 23 Jun 2003 Posts: 482

augury · l33t Joined: 22 May 2004 Posts: 722 Location: philadelphia

rsync is also used to sync systems. Search rsync.

The unix/linux directories were created with the system's perspective.

System independent:

bin/
sbin/
lib/
usr/
opt/

System dependent:

root/
home
sys/
proc/
var/
tmp/
etc/

sbin and bin are required to init and sh the system to a console.
But portage only really installs into etc (protected) and var. AFAIK there are no ports that care if anything is in there or not, at least not etc. Parts of var could be copied into a ramfs /var at boot. I can't imagine it is anywhere as intensive or confusing as running portage all on a third party NFS. If you can't reduce the system down to eliminate the HDD it is really better to use the HDD. Not having the NFS available knocks down everything. Portage can grind on this type of system pretty hard.

But for thin-clients that run maybe a 1gig of flash and a hardened system on NFS that doesn't run portage except for managed installations -- I think this is very doable. Still there is the question "why bother?", especially considering the price of gigabit routers which are required to beat a $60 ide HDD. Many networks run select software from NFS. Usually because of licensing.