This is a discussion on Live upgrade - experiance within the comp.unix.solaris forums, part of the Solaris Operating System category; --> After some advices on this USENet group couple months ago, I have decided to give live upgrade tehnology a ...
| |||||||
| Register | FAQ | Members List | Calendar | Search | Today's Posts | Mark Forums Read |
| ||||
| After some advices on this USENet group couple months ago, I have decided to give live upgrade tehnology a chance on one machine. Upgrade was from s10u4 to s10u5. Here are some obsevations of that process. There was one x86_64 machine with 2 SATA disks in SVM mirror 1 - 5 slices: /, /var, /opt, /export and swap. I have decided to temporary break mirror on /, /var and /opt and create 3 new mirrors with temporary attached 3 submirrors. Plan was to re-attach old sumbirrors from old mirrors after (sucessful) upgrade and metaclear old mirrors of /, /var and /opt. Everything was well and just great. Really nice user experiance, apart from some small annoyances: 1 - pkginfo files from SUNWlucfg and SUNWlur in /var/sadm/pkg/$PKG had been corrupted in the process of installing new live upgrade before actual upgrade - they were binary files full of null's. Breaks (terminates) pkginfo(1) output. Fixed with unpacking original packages and putting pkginfo files it in place. 2 - lucreate(1) command deleted with 'rm -rf' all direcrories which are being shared from global zone in read-write mode to non-global zones, and tried to delete those directories shared/mounted read-only - without sucess luckily. In my case, shared in read-write mode was /export/build, /export/shared, /export/software. Nothing terrible - just restored backup after upgrade horr... process. 3. Upgrade procedure installed unwanted packages which were not installed before - without obvious reason (there was no dependeny issues). Particulary, packages with drivers for non-existing hardware, Star Office, German and French localization. Nothing terrible. I just spended 2 hours diff-ing pkginfo output from 8/07 and 5/08, removing unwanted packages. 4. After I saw that upgrade was not too much broken, and system was manualy fixable, I have decided to remove old BE and reconstruct SVM RAID 1. This was not possible out of the terminal, because ludelete(1m) command tried to call lulib_relocate_grub_slice() from /usr/lib/lu/lulib which was absent from this shell library file. Added it manually witch my own idea of what it must do to the system. This worked. 5. After fixing /boot/grub/menu.lst, /etc/system and bootenv.rc, reinstalled grub just in case, and observing critical system files ad binaries just in case, I have sucessfuily rebooted system and reconstructed RAID. Solaris Live Upgrade - wonderful near-death experiance. -- Miroslav |
| |||
| In <fvvhmo$gvn$1@srv.home.local> Miroslav Zubcic <news@big-other.com> writes: >4. After I saw that upgrade was not too much broken, and system was >manualy fixable, I have decided to remove old BE and reconstruct SVM >RAID 1. This was not possible out of the terminal, because ludelete(1m) >command tried to call lulib_relocate_grub_slice() from /usr/lib/lu/lulib >which was absent from this shell library file. Added it manually witch >my own idea of what it must do to the system. This worked. I just opened a support case on this issue. It's supposed to finish with a bug report being filed. I'm sure it's been reported before, but I couldn't find the bug report. That shell function is in the Nevada version of lulib, so it shouldn't be a big problem to issue a patch for it in Solaris 10. -- -Gary Mills- -Unix Support- -U of M Academic Computing and Networking- |
| |||
| Gary Mills wrote: > I just opened a support case on this issue. It's supposed to finish > with a bug report being filed. I'm sure it's been reported before, > but I couldn't find the bug report. That shell function is in the > Nevada version of lulib, so it shouldn't be a big problem to issue a > patch for it in Solaris 10. In comparation with stange insane "rm -rf" prior to cpio-ing in non-global zones over directories re-mounted from global zone, this is really a tiny, very minor issue. Also, I had to remove newer Real Player and put old one from u4, because it has some "libeel" library dependency which is not installed on my system. Doing ldd * 2>/dev/null | grep "not found" in /usr/lib /usr/sfw/lib, /usr/openwin/lib and /usr/dt/lib is a good idea after this kamikaze inspired upgrade procedure. -- Miroslav |
| ||||
| Miroslav Zubcic wrote: > Also, I had to remove newer Real Player and put old one from u4, because > it has some "libeel" library dependency which is not installed on my system. Oh yes, one more thing: /tmp directory mount point in non-global zones was recreated in 0755 mode instead of 1777. -- Miroslav |