[svlug] Partitioning problem

[Rick Moen]

Quoting Bruce Coston (jane_ikari at yahoo.com):

> Assuming this is your server rebuild : Are those 18g drives likely to
> work out as a raid0 with most of your comonly used data on them backed
> to the 73g providing redundancy via something like dirvish under cron? 

I actually didn't intend to rely on intra-host data archiving:  Backup
will be to elsewhere, cutting out some common-mode failures.  (At some
point, the proverbial shoemaker's kids will cease going barefoot, and
there will be DLTtapeIV media & matching drive elsewhere in the house,
duplicated offsite, etc.)

> I'd also like serious internet caching from that 73g.

We'll see about Squid on that.  Alternatively, that function might be
served by the VA model 500 (the one of two that I won't give away) after
refit with a pair of 36GB drives and maybe an 18GB one.  That's next up,
after server cutover.

_Transparent_ proxy is probably not feasible on account of network
topology.  

(BTW, if anyone wants a few 1-2GB SCSI drives, drop by next Saturday's
CABAL meeting.  Who knows?  If I get busy, there might also be a free 2U
VA box.)

> I expect others to chime in about not using a single / partition to 
> do most of the work - which has benefits . 

Well, you might notice that my for-instance layout puts on /dev/sda only 
things that are either dispensible or easily rebuilt (well, except for
/root, which I don't rely on -- and I forgot to mention that /opt
always symlinks to /usr/local/opt ).

For the current server, I keep copies of these files off-host in
addition to my datafiles backup:

# fdisk -l /dev/sda > /root/partitions-sda-$(date -I)  
# fdisk -l /dev/sdb > /root/partitions-sdb-$(date -I)  
# dpkg --get-selections > /root/selections-$(date -I)  

Given that, I can lose / , /usr, and damned near everything else, and
recover pretty quickly like this:

1.  Do minimal base install.
2.  # dpkg --set-selections < selections-[date string]
    # apt-get -u dist-upgrade
3.  Rebuild /etc with reference to /etc backup.
4.  Copy other files back.
5.  Rebuild or reinstall local packages (just Leafnode 2.0 beta, 
    at the moment).
6.  Restart services. Check logfiles for errors.
7.  Profit.