Alan Hargreaves' Weblog

Live Upgrade and TimeSlider gotcha

Tried to upgrade my workstation over the weekend to snv_117. Apart from a little tridying up I had to do as a package didn't install correctly, all apeared to be going fine. I then went to unmount /.alt.snv_117, and it failed saying that the filesystem was busy.

fuser -c showed no processes using the mount point. What could it be?

A little bit of dtracing the umount2() system call was illuminating.

  1              <- zfsctl_umount_snapshots           0                0
  1            <- zfs_umount                          0               16

Hang on, snapshots? Although it returned 0, let's just check; as I do have timeslider enabled on this box.

rootksh@vesvi:~$ zfs list -t snapshot|grep 117                                                                     
pool/ROOT/snv_116@snv_117                                   4.03M      -  8.78G  -
pool/ROOT/snv_117@zfs-auto-snap:hourly-2009-06-19-09:00     43.8M      -  7.99G  -
pool/ROOT/snv_117@zfs-auto-snap:hourly-2009-06-19-10:00     48.9M      -  8.44G  -
pool/ROOT/snv_117@zfs-auto-snap:hourly-2009-06-19-11:00     43.7M      -  8.74G  -
pool/ROOT/snv_117@zfs-auto-snap:frequent-2009-06-19-11:15   42.6M      -  8.75G  -
pool/ROOT/snv_117@zfs-auto-snap:frequent-2009-06-19-11:30   45.8M      -  8.76G  -
pool/ROOT/snv_117@zfs-auto-snap:frequent-2009-06-19-11:45   38.1M      -  8.77G  -
pool/ROOT/snv_117@zfs-auto-snap:hourly-2009-06-19-12:00     38.5M      -  8.80G  -
pool/ROOT/snv_117@zfs-auto-snap:daily-2009-06-22-00:00          0      -  8.80G  -
pool/ROOT/snv_117@zfs-auto-snap:weekly-2009-06-22-00:00         0      -  8.80G  -
pool/ROOT/snv_117@zfs-auto-snap:hourly-2009-06-22-10:00         0      -  8.80G  -
pool/ROOT/snv_117@zfs-auto-snap:frequent-2009-06-22-10:30       0      -  8.80G  -
pool/ROOT/snv_117@zfs-auto-snap:frequent-2009-06-22-10:45       0      -  8.80G  -

Oh, timeslider was taking snapshots of the filesystem while it was upgrading. Hmm maybe we should be having that disabled on the target of a live upgrade (rfe coming, but I don't hold out a lot of hope).

Anyway, removing them was not difficult:

rootksh@vesvi:~$ zfs list -t snapshot|grep snv_117@zfs-auto|awk '{print $1}' | xargs -L 1 zfs destroy
rootksh@vesvi:~$ luumount snv_117                                                                                  
rootksh@vesvi:~$ 

Something to keep in mind if you are using timeslider, zfs root and live upgrade (I wonder if we would have the same issue with 'pkg image-update' in OpenSolaris).

Posted at 11:05AM Jun 22, 2009 by Alan Hargreaves in Solaris  |  Comments[1]