I wanted to reinstall the operating system on my server. After my totally
newbie install of Fedora Core 2 test 3, then using the development and
bleeding-edge repositories and adgering the Python system (including
up2date), and then using atrpms with innocent glee until I found out that
they had a totally different and even nastier version of 'bleeding edge',
the whole system felt as if it had been limping along being delicately
prodded to stay upright. The final straw was the
attack
on another machine in the lab which, while it didn't
seem to have actually broken into my machine, still didn't give
me the Ring Of Confidence.
But here's how not to do an install of Fedora Core 5:
- Start on the day you've just installed a Java system for another
person in the lab - the day before he goes overseas for a conference.
- Make sure the last disk image you have is slightly corrupted so that,
after getting through all the other packages, Fedora Core says it can't
install trivial-program-1.0.0.1-x86_64.rpm and has to reboot,
leaving the entire install adgered.
- Don't check your images against the SHA1SUM file but try for a while
to install using the net boot disk off your local hard drive, getting
the same install error as above in different places.
- Use the i386 net install disk when you want to install an x86_64
server. Waste some more time finding and burning the x86_64 net install
disk, as you can't quite get either the internal or the external burner
to erase and write the disk correctly. Waste some more time finding
this out the hard way by booting off a disk that's not correctly burnt.
- When you finally get the system upright, find out that the portable
USB disk that has its own power supply seems to not recognise any
system you plug it into, with the exception of the Windows XP system that
is, for some reason, adgered sufficiently that it cannot see anything on
the network anywhere.
- Once you've got the disk freed from its enclosure and plugged into
a separate machine (because finding an IDE cable and installing it in
the actual install machine is 'too hard'), remember to make the temporary
logical volume too small to take all the files. If you're lucky, you
can catch yourself doing this and re-make it before you actually copy
anything across the network.
- Remember to install specific rules in your backup scripts that don't
backup your Thunderbird email or your scratch directory full of music.
Especially, remember not to disable these rules when making that final
backup.
- Try restoring the differential backup first, without restoring the
base backup. Assume that doing the first automatically also does the
second. Install the files in the places they're supposed to go. Wonder
why half your configuration doesn't seem to be there, or working. Waste
some more time redoing it the right way. Here, again, you can be lucky
if you learn from previous mistakes and don't just restore over the
places things were backed up from but restore to a temporary directory
and be selective about what you copy back.
Why am I so stupid?
On the plus side, I learnt a few things:
- Booting from the net install disk and then installing off a local
hard drive's worth of ISOs is superior to burning a bunch of CDs in
every respect. The fact that you can do this from a removable USB disk
is all the sweeter.
- dar is the backup
program of the gods. It will back up and restore SELinux security
information in version 2.3.0. Put a copy of dar_static on your backup
media and away you go.
- Use LVM for flexibility. Use Software RAID to span your root
partition across two disks and get mirroring. Then use LVM to create
everything else. If it made any sense I'd make a LVM partition as a
hot spare for the software RAID, but they're all on the same disks
anyway. This way I can add another disk and span the LVM across it,
and then add the hot spare out of space on that Physical Volume.