Illegitimi non carborundum


Upgrading to Mandriva 2009.1

Well it's that time of the bi-year again. The twice annual Mandriva upgrade! Well in fairness, I actually skipped 2009.0 at work, so it's been a year since I last fiddled there. There was no particular reason for me skipping this release, I just think we were quite busy and a convenient time to do the upgrades didn't really pop up.

So, what happened I hear you cry? Well, surprisingly little drama thankfully! There were a few issues, but they were mostly minor so I only mention them here in case they trip up the less ninja skilled out there :p

LDAP: The upgrade of the LDAP server wasn't as smooth as it could be. When the database format format changes you need to run "service ldap recover" and after that it works fine. Half way through the upgrade the ldap server package is updated and the server restarted. When this happens, all my users go bye bye. Simply rinning the recover command is enough to allow you to restart the server. You can do this while the upgrade is running (provided you have a login window already opened or you can access they machine with a system user account.

MySQL: This one was a little tricky. If you've ever modified your /etc/sysconfig/mysqld file, the new one will be placed in a .rpmnew file rather than overwrite your changes. This is not generally a problem, but somewhere between 2008.1 and 2009.1 the temporary directory MySQL uses changed. I suspect this relates to the MySQL 5.1 upgrade. Sadly if you use the old directory (a .tmp folder inside the datadir), and MySQL cannot access and use this folder, InnoDB table engine does not initialise correctly leaving you without access to several tables. So make sure you get your sysconfig file right and things should be OK. You should run mysql_upgrade after installing 5.1 to make sure things are all in good shape for that.

AutoFS: There were a couple problems with NFS I encountered, the first being with AutoFS. I generally have an AutoFS setup in the office to pull the mount information from LDAP. This generally works fine. As it happens, my local RPM mirror is also accessed via NFS. Half way through the upgrade, autofs is upgraded and restarted. This time round this had the effect of unmounting the NFS shares. I've had this happen before, but this time it appears to be a Kernel related thing. After a bit of digging, I discovered that the new autofs was wanting to do something with UDP datagram connections that the currently running kernel didn't like. I'll be honest and say I didn't bother looking too hard to check that this is a 100% correct diagnosis, but it certainly looked like it. I just mounted the NFS share manually and carried on with the upgrade. After a reboot on the new kernel, autofs worked fine, so I'm not going to dwell on it 🙂

NFS locking: I had some issues with NFS Locking in that the file. I kept getting "lockd: cannot monitor nfs" in dmesg and looking further in the logs I saw: "rpc.statd[6138]: creat(/var/lib/nfs/sm/ failed: Permission denied". Looking at this file, it was owned by root. I removed it and it was recreated as the rpcuser and all seems well after that.

Ramdisk+LVM: There was a minor problem on one machine. As all IDE disks are now /dev/sd? rather /dev/hd? I had a very minor issue with an LVM setup. The kerne's ramdisk has a built in routine to wait for devices known to be part of the LVM to appear after the driver load before trying to activate them. Sadly, the ramdisk was created to wait for /dev/hd? rather than /dev/sd? as it had become. This basically added 1minute to the boot time, but it didn't cause any problems. Recreating the ramdisk after the first boot allowed things to work just fine.

That's about the extent of the problems I had here. All of them are fairly specialised problems resulting from a pretty non-standard setup, so I'm not really too worried about them. All in all, things went very smoothly!

Share and Enjoy:
  • Digg
  • StumbleUpon
  • Facebook
  • Yahoo! Buzz
  • Twitter
  • Google Bookmarks
  • Slashdot
  • Guy Bormann

    hi Colin, in my case the LVM issue made my laptop unbootable (after the wait I get a kernel panick). It looks like I have to split my /boot partition (1GB) in 2 (1 /boot, the other /) to make it bootable again. Let’s not mention the haldaemon service issues, the fundamental config error of the 64bit version of gdk-pixbuf (no loader entry for .png!), pulseaudio failure, … This is my worst install experience with Mandrakelinux/Mandriva (a user since 2000) ever.

    BTW, my machine is an HP Pavillion laptop (dv6158eu).


    • Colin

      Bummer you had such a bad experience 🙁 Has to be said, that other than the issues I encountered and documented, this upgrade was pretty smooth. Certainly not as bad as the last time round! Hopefully the 2009.1 -> 2010.0 will be smoother. It’s due out soon, so I’d recommend you jump straight to the RC2 released yesterday.

  • Guy Bormann

    Hi Colin,

    Spent the better part of the afternoon to get it up and running again. Doing a pure DVD-based __re__-install cured all problems : root on LVM, haldaemon service starts (I had a suspicion it was a permissions issue; it was and it was triggered by a lingering buggy update), pulseaudio starts, png loader sig in the gdk-pixbuf config, no more permission nags in GNOME, …
    After a regular boot, doing the updates is harmless, including the recommended kernel security update. So, apparently the fine-tuned install procedure doesn’t mix well with the intrusion of update packages.

    As an aside : I tasted some of the improved stability of the 2.6.29 kernel flavour after a first install. An upgrade was not save but the install tried to retain as much info of the 2008.0 version as possible. Due to a trigger-happy touchpad the updates got install while I didn’t get the chance to change anything in the final config summary phase (except timezone settings, network and graphics drv). Booting was still possible, probably because I didn’t touch the bootloader settings.
    Once I figured out something was really screwed with the permissions (especially the dbus ones; there are many related bug reports for various distributions), I tried to reinstall, first by doing an upgrade 2009.1->2009.1 (hey, I can’t help it was suggested by the installer) but it didn’t solve the non-fatal issues. Then I did complete reinstalls (keeping the partitionmount point mappings), but always with updates, with the aggravation of a booting problem after the first reboot.
    For some reason I didn’t trust the updates (I remember seeing a timestamped dbus rpm while the one on the DVD medium had no timestamp; there was a serious permission problem with a dbus launch helper), so I decided to do an install without defining any distro mirror and update source. And here we are again 🙂 : ALL ISSUES SOLVED, including the boot problem.

    Obviously, on my machine this is an Errata for 2009.1. The problem is : I didn’t keep record of all my trial and error work and I didn’t save all the intermediate install reports 🙁 So, I’m afraid I can’t write for the wiki. But I’ll get on the Forum. There are people with boot problems on RAID+LVM combo’s with similar symptoms. I will just wait for the release of 2010 although I don’t like .0 releases (I had issues with 2007.0 on my desktop and 2008.0 on this laptop) that got hammered out in the .1 releases, and later the .0 updates (took those instead 🙂

    Sorry for the long comment but this might be helpful to other and your blog shows up on Google 🙂



  • Guy Bormann

    Oh, oh, I really have to add a last remark.
    Don’t change the group of any users you add during install to a group id below 500. It might further complicate the permission problems for dbus/haldaemon and their clients’ access to them. If one likes to add users to a local sharing group, do it after the first regular boot into the new system.
    (I don’t understand the use of user-centric groups. Maybe it is meant to be an additional security layer for users unfamiliar with UNIX permission masks and management.)

    PS: I’ll stop hi-jacking your blog now 😛
    (And, most of the issues might have been ironed out anyway since the latest batch of updates!!)