Regarding regular zfs

[Peter Maloney]

On 2013-04-05 13:07, Damien Fleuriot wrote:
> -I've implemented mbuffer for the zfs send / receive operations. With
> mbuffer the sync went a lot faster, but still got the same symptoms
> when the zfs receive is done, the hang / unresponsiveness returns for
> 5-20 seconds
> -I've upgraded to 8.3-RELEASE ( + zpool upgrade and zfs upgrade to
> V28), same symptoms
> -I've upgraded to 9.1-RELEASE, still same symptoms

> So my question(s) to the list would be:
> In my setup have I taken the use case for zfs send / receive too far
> (?) as in, it's not meant for this kind of syncing and this often, so
> there's actually nothing 'wrong'.

I do the same thing on an 8.3-STABLE system, with replication every 20
minutes (compared to your 15 minutes), and it has worked flawlessly for
over a year. Before that point, it was hanging often, until I realized
that all hangs were from when there was more than 1 writing "zfs"
command running at the same time (snapshot, send, destroy, rename, etc.
but not list, get, etc.). So now *all my scripts have a common lock
between them* (just a pid file like in /var/run; cured the hangs), and I
don't run manual zfs commands without stopping my cronjobs. If the hang
was caused by a destroy or smething during a send, I think it would
usually unhang when the send is done, do the destroy or whatever else
was blocking, then be unhung completely, smoothly working. In other
cases, I think it would be deadlocked.


NAME                        USED  REFER  USEDCHILD  USEDDS  USEDSNAP 
AVAIL  MOUNTPOINT
tank                       38.5T   487G      37.4T    487G      635G 
9.54T  /tank
tank/backup                7.55T  1.01T      5.08T   1.01T     1.46T 
9.54T  /tank/backup
...

Sends are still quick with 38 T to send. The last replication run
started 2013-04-05 13:20:00 +0200 and finished 2013-04-05 13:22:18
+0200. I have 234 snapshots at the moment (one per 20 min today + one
daily for a few months).