How to fix a partially successful upgrade?

JormaR · #1 19th January 2010, 06:45 PM

I hit some trouble upgrading from FC 10 to FC 12 using the netinst CD.

I had previously successfully upgraded the system from FC5 to FC10 with the same method - that time everything went well.

At first the upgrade went well - the installation program found my FC installation and detected the partitions right.

The problem appeared when about 90% of the rpm's were installed. Anaconda was complaining about a fatal error with a package and quit.

After rebooting and restarting the installation program it does not recognize my partitions anymore and suggests I'd format them for a fresh installation. As all my data is there I'd rather not do that.

I then rebooted again and chose "Repair" from the installation menu. I got to the console and noticed that all partitions mounted cleanly and all the data seems to be there.

Trying to reboot from the hard disk almost works. The system gets up, goes through grub, gets the kernel running and mounts the disks but fails to launch Xorg.

I get to a terminal login from here by pressing Ctrl+Alt+F2. When I type the user name and press enter the terminal seems to crash - does not ask for the password but simply returns to the user name query. An error message flashes just after pressing enter but too quickly for me to read.

When I boot from CD again and go to the terminal via the "Repair" option the previous events can be seen from /var/log/messages. Xorg tries to start and stops trying after a certain amount of failures.

After that it says this:

Code:

init: prefdm main process (2123) terminated with status 1
init: prefdm main process ended, respawning
init: prefdm respawning too fast,stopped
init: tty2 main process (1617) terminated with status 1
init: tty2 main process ended, respawning
init: tty2 main process (2184) terminated with status 1
init: tty2 main process ended, respawning
init: tty2 main process (2185) terminated with status 1
init: tty2 main process ended, respawning

My best guess is that this happens because the upgrade was aborted before it completed and hence some necessary rpm's were left out. I figured I could possibly fix this by updating the packages with yum from the terminal if I got to log in.

What should I do now? Could I for example try and manually copy some missing files to the disk as I can access it via the repair option?

And yes - I know upgrading is at one's own risk.

Nokia · #2 19th January 2010, 07:42 PM

Boot straight it runlevel 3, login as root and try

Code:

yum-complete-transaction
yum update --enablerepo=u*g

See if you get any errors and post it here, maybe we could salvage something...

JormaR · #3 19th January 2010, 09:48 PM

I did that - changed default runlevel to 3 and rebooted.

Unfortunately the problem remains - trying to log in from the log in prompt is impossible. After typing root <Enter> the prompt simply resets without asking for the password.

An error message flashes in the screen when the prompt resets but I can not make it out - it's so quick.

I thought of trying to log on from another computer via SSH as the network and services seem to get up and running normally. I'll try that next.

Nokia · #4 19th January 2010, 11:32 PM

Perhaps more suitable would be to boot a livecd, mount the installation in /mnt and chroot. I somehow doubt ssh would serve anything now.

JormaR · #5 20th January 2010, 10:15 AM

You are probably right. I had little luck connecting via ssh but that might be due to the firewall settings.

Anyhow - I can boot from the netinst CD and get to the file system. It mounts the disk correctly to to /mnt/sysimage. I said chroot /mnt/sysimage and many things worked correctly - chkconfig for example.

Trying to run yum did not work though - some Python libraries were missing. I'll burn a liveCD and try with that.

Thanks for the help so far!

Nokia · #6 20th January 2010, 01:02 PM

Quote:

Originally Posted by JormaR

You are probably right. I had little luck connecting via ssh but that might be due to the firewall settings.

Anyhow - I can boot from the netinst CD and get to the file system. It mounts the disk correctly to to /mnt/sysimage. I said chroot /mnt/sysimage and many things worked correctly - chkconfig for example.

Trying to run yum did not work though - some Python libraries were missing. I'll burn a liveCD and try with that.

Thanks for the help so far!

Using the LiveCD - or a USB stick

Code:

yum install yum-utils
yum deplist yum
yumdownloader pkg1 pkg2 ...   #### yum and the other dependencies
chroot /mnt/sysimage
rpm -Uvh --force <rpms>.rpm
yum-complete-transaction
yum update --enablerepo=u*g

If al l goes well, reboot.

Here's what deplist shoud look like

Code:

Finding dependencies: 
package: yum.noarch 3.2.25-1.fc12
  dependency: rpm >= 4.4.2
   provider: rpm.x86_64 4.7.1-6.fc12
   provider: rpm.x86_64 4.7.2-1.fc12
  dependency: pygpgme
   provider: pygpgme.x86_64 0.1-17.20090824bzr68.fc12
  dependency: python >= 2.4
   provider: python.x86_64 2.6.2-3.fc12
   provider: python.x86_64 2.6.2-2.fc12
  dependency: rpm-python
   provider: rpm-python.x86_64 4.7.1-6.fc12
   provider: rpm-python.x86_64 4.7.2-1.fc12
  dependency: python-urlgrabber >= 3.9.0-8
   provider: python-urlgrabber.noarch 3.9.1-2.fc12
   provider: python-urlgrabber.noarch 3.9.1-4.fc12
  dependency: /usr/bin/python
   provider: python.x86_64 2.6.2-3.fc12
   provider: python.x86_64 2.6.2-2.fc12
  dependency: python(abi) = 2.6
   provider: python.x86_64 2.6.2-3.fc12
   provider: python.x86_64 2.6.2-2.fc12
  dependency: yum-metadata-parser >= 1.1.0
   provider: yum-metadata-parser.x86_64 1.1.2-14.fc12
  dependency: config(yum) = 3.2.25-1.fc12
   provider: yum.noarch 3.2.25-1.fc12
  dependency: python-sqlite
   provider: python.x86_64 2.6.2-3.fc12
   provider: python.x86_64 2.6.2-2.fc12
  dependency: python-iniparse
   provider: python-iniparse.noarch 0.3.0-3.fc12
   provider: python-iniparse.noarch 0.3.1-2.fc12

Also, make sure you boot the livecd that matches your arch - either i686 or x86_64

JormaR · #7 20th January 2010, 07:54 PM

I am making progress - now writing this on the troubled machine running the live CD.

I did what you suggested above and manually installed the packages. I had to add the --nodeps option to make them go in but they eventually did.

yum still does not run though. It says -

Code:

There was a problem importing one of the Python modules
required to run yum. The error leading to this problem was:

   /lib/libfreebl3.so: version `NSSRAWHASH_3.12.3' not found (required by /lib/libcrypt.so.1)

ls on the other hand tells me that this very file is in place.

I am beginning to think that the problems may be caused by the failed installation having left FC 10 packages in the system.

Code:

rpm -qa |grep .fc10| wc -l

gives 1410
and

Code:

rpm -qa |grep .fc12 |wc -l

gives 1729

In other words - I have a huge amount of duplicate pacakges!

What if I just brutally remove all the fc 10 packages?

Or should I take a more surgical approach and remove them one by one when I find duplicates that sound relevant -starting with yum and python duplicates?

Nokia · #8 20th January 2010, 08:03 PM

At this point, backup data first.

When done, you could try tinkering with it a bit more, but in the end I suspect you'll need to do a fresh install. Either way, let me know what you're doing when the backup is secured.

JormaR · #9 20th January 2010, 09:05 PM

Words of wisdom. Will do.

After that I thought to check

Code:

yum deplist python

And make sure that I have all the packages python requires from FC 12 and possibly remove FC10 duplicates. Rpm is working correctly so I can upgrade and install packages manually.

I am running one konsole window with the chroot to my root partition done and one directly from the live CD where yum runs normally.

JormaR · #10 21st January 2010, 10:04 PM

I\ve now backed up the data and made some progress.

The yum error was because the mentioned library libfreebl3.so was provided by a different package in Fedora 10 and Fedora 12. Because the upgrade failed I had both packages installed.

Removing the Fedora 10 package with

Code:

rpm -e nss-devel-3.12.3.99.3-2.10.6.fc10.i386
rpm -e nss-3.12.3.99.3-2.10.6.fc10.i386

got me further. I also checked earlier that python by itself seems to work.

Then yum stopped when it tries to read /proc/cpuinfo

I had mounted my root partition from the disk to /mnt/rootpartition/. Then I had said

Code:

chroot /mnt/rootpartition/

. Due to this /proc/ is empty.

I copied /proc/cpuinfo to /mnt/rootpartition/proc and got further.

Now yum is complaining of missing /dev/urandom. I tried to be smart and created a symbolic link from /dev/urandom to /mnt/rootpartition/dev/urandom but had no that way. The error I got was -

Code:

 cannot access /dev/urandom: Too many levels of symbolic links

How to get further from here?

---------- Post added at 12:04 AM CST ---------- Previous post was at 12:01 AM CST ----------

Ah - how about using google?

I said

Code:

mknod -m 644 /dev/urandom c 1 9

And now yum runs!

JormaR · #11 23rd January 2010, 07:57 PM

The problem is now solved - writing this on the machine in question.

Most of the problem was caused by duplicate packages from FC10 and FC12. As the upgrade process was interrupted the installer apparently did not remove the FC10 packages - hence the trouble.

Luckily most of the FC12 packages were there - once I got yum running things got a lot easier. I had to manually remove all the FC10 packages not to mention packages from FC9 and even FC5 that were lurking in the depths of the system.

I still have some work to do - certain apps such as Konsole are missing - but now such things are easy to do.

Having struggled with this for several long nights I think I might go for the clean installation next time...

Thanks for the help - without the advice from Nokia I would have been lost.

Nokia · #12 23rd January 2010, 08:44 PM

Glad to hear you sorted it out.

Could you please explain the mknod statement you posted above ? I'm not sure what you need it for and what it did exactly.

JormaR · #13 27th January 2010, 12:51 PM

The mknod -command: there was no /dev/urandom as I was working in a directory tree created with chroot. As yum needed that to run I googled urandom and found this page:

http://linux.about.com/library/cmd/blcmdl4_urandom.htm

Then I simply applied that command which created the /dev/urandom under the directory I needed it ( in the full file system it was /mnt/rootpartition/dev/urandom.)