Solved Sync-ing vnodes at ZFS shutdown failing

[semi-ambivalent]

Recently my localhost, a ZFS two disk mirror with geli, has been failing to sync vnodes before shutting down. Instead of getting the countdown to zero as you would with UFS there's just a long string of sixes or eights or threes, etc., then it times out, gives up and continues shutting down normally. I haven't noticed any indication of missing data but sometimes at boot the machine doesn't seem to see the keyboard or mouse. I just hit the power switch again to force reboot and that has always worked. I've run scrub three times and nothing is found. Once running all behavior and performance is nominal. Any guesses on what might be causing this?

Thanks,
s-a

 

[phoenix]

Which version of FreeBSD? There was some discussion about a similar issue on the mailing lists a month or three ago, with some fixes pushed into Subversion. Not sure if it made it into 11.1 or not, though. But, if you are using anything older than that, you may want to upgrade.

 

[semi-ambivalent]

Which version of FreeBSD? There was some discussion about a similar issue on the mailing lists a month or three ago, with some fixes pushed into Subversion. Not sure if it made it into 11.1 or not, though. But, if you are using anything older than that, you may want to upgrade.

Thanks Phoenix. Apologies for not mentioning that, I certainly should have. I'm running 11.1R (now at -P3).

I've played around a bit thinking that my nfs-mounted /home was not being cleanly unmounted so I tried doing

shutdown now

to drop to single user and then manually unmounting the nfs mount. Then doing

shutdown -h (-r) now

and there were no vnodes to sync. Just 000... and shutdown. This is what I was used to seeing after moving to ZFS from UFS. So it's good I guess. Further, I can drop to single user and shutdown fine without first unmounting the nfs mount. So it looks like localhost is running something that's not getting picked up by the shutdown scripts or is trying to die but can't fast enough before getting squashed, leaving dirty vnodes. I just do the 'shutdown now' first, it's takes no time at all.

s-a