Re: yum and crash recovery

2007/3/2, seth vidal <[email protected]>:
> On Fri, 2007-03-02 at 12:35 -0500, Michel Salim wrote:
> > 2007/3/2, Fajar A. Nugraha <[email protected]>:
> > > Michel Salim wrote:
> > > > I have an unreliable system that is prone to overheating (at which
> > > > point the kill switch would mercifully save the CPU from a flaming
> > > > death, but at the expense of whatever is going on at the moment) --
> > > > and doing a large yum transaction is quite reliable at causing a
> > > > lock-up (I'm making sure I set the CPU speed to minimum before
> > > > updating now).
> > > >
> > > I may be saying the obvious here, but you should find out how to cool
> > > down that server, or throw it away. Unless it's NOT a server :-P
> >
> > That's the scary thing. It's a laptop (Turion 64)! And it's a good
> > suggestion -- I'm waiting for the next iteration of the Macbook.
> > Hopefully Linux has full support for it by then (apart from EFI)
> >
> > > > Anyway, the question is, if you're partway into the transaction, and
> > > > some packages have been installed (but the old ones not removed yet),
> > > > is there a way to tell yum to clean up the mess? There must be some
> > > > transaction logs somewhere that can be used to automate the process.
> > > >
> > > IMHO the best way is to recover rpmdb from backup, and restart the yum
> > > process.
> > > SUSE has /etc/cron.daily/suse.de-backup-rpmdb. I don't think Redhat has
> > > it by default.
> > >
> > Hmm .. the thing is, some RPMs have already been installed (e.g. if
> > foo-x.y is updated to foo-y.z, the system has both installed). If I
> > roll back the rpmdb, it will actually be inconsistent with what's on
> > the disk, right?
> >
> > What I want is a way for yum to say "oh, I detect an incomplete
> > transaction -- let me just continue it and make sure the old RPMs get
> > properly removed"
> >
>
> so there's no way for yum to know based on what information is in the
> rpmdb what has happened on the system. The mechanism we've been using
> thus far is to:
>
> 1. look for dupes in the rpmdb
> 2. remove the older version
> you can try running:
> http://linux.duke.edu/~skvidal/useful-scripts/dupes-cli.py
> to accomplish those two.
>
> 3. run rpm -Va and see what's broken
> 4. re-run the update command to see if it can fix the broken.
>
>
>
> it's worth a shot. not ideal.
>
Thanks! The script works fine for removing older packages -- the
problem is with RPM being hard-wired to ignore /usr/share and so files
owned by >1 packages there get removed.

If yum has an option to reinstall packages that are broken (as if they
were older) then the script would be a one-stop solution.

Thanks,


-- 
Michel Salim
http://hircus.wordpress.com/

My theology, briefly, is that the universe was dictated but not signed.
                -- Christopher Morley