in-place clean installs with one / partition

[Andy Green]

Wolfgang S. Rupprecht wrote:

> A few months ago I did a "yum -y update ; shutdown -r now".  I don't
> know if the yum database needs a few 10's of seconds to get flushed
> cleanly, or what, but it got royally scrambled.  A "yum list" after
> the reboot showed that the set of *.db files was corrupted.  I could
> rebuild some of it from the yum cache directory, but roughly 1/3 of
> the rpm headers (~400) were no longer in the cache.  Nuking all of fc5
> would be a good way to make sure that everything in the filesystem is
> accounted for with no stragglers.

Yum does have some kind of database down /var/cache/yum/, but it is just 
holding what it last heard was available from the repos AIUI and can 
very safely be nuked with yum clean all and re-running yum.

RPM (yum uses the rpm libs to do the actual install/upgrade/erase) on 
the other hand holds a database of everything that is installed on the 
box, this one is down /var/lib/rpm.  RPM has a longstanding bug where 
some kind of internal db4 database cache files, /var/lib/rpm/__db*, can 
get left in an inconsistent state, until these are deleted rpm then 
hangs or chokes.  I don't know if that was the problem but with any 
funny behaviour from rpm that is the first port of call.

A nice shutdown like you did syncs stuff to disc and umounts your drives 
cleanly before restarting, being the textbook thing to do it should not 
have made any trouble.

I guess if it was my box I would muse about how important the data in 
the LVMs was to me, and take steps to back it up if it was that 
important.  Then I would use the rescue mode method to destroy /usr, 
/etc and /var at least and do an "install" of FC6 (without formatting, 
and taking care to list your LVMs in the fdisk part of anaconda).  I 
think that will work fine, but then my most deodorant-challenging 
adventures usually start that way.

> past I've always bought a newer bigger disk whenever I installed a new
> OS.  I also used to unplug the old disks just to prevent mishaps
> during the install.

Yes, it's a very cautious and satisfying way, especially if you find 
yourself installing Linux for an eager yet fretful Windows convert. 
Knowing they can go back makes it easier, nobody has gone back to their 
old drive yet though ;-)  But it allows for the fact that given a random 
machine from a random year with a random chipset and random video, one 
is never quite sure how much of it Linux will be able to bring up (I 
take a pile of outdated but supported video cards with me), given random 
BIOS versions, no Internet access until the box is working again, etc. 
Usually it is everything important, but if you are visiting and only 
have limited time at least you can know for certain you can back out 
100% to the old drive with 1 minute's notice.  If your life isn't 
exciting enough trying to grapple someone else's old machine into a good 
Linux shape with limited time will change that!

> I'm still waiting for price parity before I pick up one of the 750GB
> ones.  They still want $100 too much vs. buying 3 of the same series
> drives populated with 250GB worth of heads and platters.

ACK - 750GB are not at the $/MB sweetspot where one should buy.

-Andy