This is strange. I have been using time-slider and a synchronization script every 15 minutes since the OpenSolaris days. The recent update to 11.1 has caused the backup pool randomly to refuse to delete phantom snapshots during a weekly trim of the pool. Just now I saw several errors during a snapshot delete:
cannot destroy 'backup/export/site@zfs-auto-snap_daily-2012-11-05-14h48': snapshot already exists
The solution is to export then import the pool, but this happened earlier today and I could not export, even with the -f flag because the pool was busy (?). I suspended the synchronization and could find no processes with open files on mount points. I settled on a reboot.
Does this happen to anyone else? Is there a less disruptive way to recover?
Also, I have noticed that CPU utilization in pool and I/O operations is markedly higher on 11.1 vs. 11. Server is HP DL585 with 4x dual core Opteron 8216. Before the 11.1 upgrade, scrubbing three pools could sustain 400MB/s throughput with about 30% CPU utilization. After 11.1, the scrub will consume about 70% CPU with 320-ish MB/s throughput. Nothing else has changed. Similar effects happen during the regular snapshot synchronizations.
Any insight on either issue is appreciated.
I also get this - it's quite annoying.
This may help:
svcadm disable -t time-slider
for f in `zfs list -H -t snapshot -o name`; do zfs release -r org.opensolaris:time-slider-plugin:zfs-send $f; done
svcadm clear svc:/application/time-slider/plugin:zfs-send
svcadm enable time-slider