10.1 + ZFS snapshot eating diskspace

[Rick Romero]

  Ronald was on the right track. 
https://lists.freebsd.org/pipermail/freebsd-fs/2015-April/021144.html
The drives on the new system, while identical to the old, use native 4k
sectors. In FreeBSD 10.1, ZFS automatically used ashift=12 - the old 9.2
system used an ashift of 9.
Since my data is small files (mostly < 16k), the space is used VERY
inefficiently with an ashift of 12.

Setting vfs.zfs.max_auto_ashift: 9  prior to creating the new pool
resulted in a properly sized volume after the zfs receive.
zpool status complains about performace degradation, of which I'm
definitely choosing space+cost over performance, but at least zpool status
-x shows all good.

<now gluing hair back>

Quoting Rick Romero <rick at havokmon.com>:

> Try number two.   I built another new system, no encryption this time.
>
> I replacated ONE snapshot that is about 562GB of data.
> (I just found Ronalds reply in my Spam folder, sorry!)
> This new 10.1 system has the exact same 3 drives in RAIDZ1 as the
original
> source (9.2).  What's confusing is the original RAIDZ1 is replicated
> correctly to a 10 drive RAIDZ2 (10.1), but the RAIDZ2 source cannot
> replicate data correctly to a new 3 drive RAIDZ1.
> So not only is this a problem with the new system, but it concerns me
that
> if there were a problem with the old system that a full restore from
> backup
> would eat all the disk space.
>
> Source:
> # zfs get all sysvol/primessd_home |grep -i used
> sysvol/primessd_home  used                 
> 822G                   -
> sysvol/primessd_home  usedbysnapshots      
> 260G                   -
> sysvol/primessd_home  usedbydataset        
> 562G                   -
> sysvol/primessd_home  usedbychildren       
> 0                      -
> sysvol/primessd_home  usedbyrefreservation 
> 0                      -
> sysvol/primessd_home  logicalused          
> 811G                   -
>
> Right? 562 is the 'current' amount of space used? 
>
> So I send it to a new box, and this is the result
>
> # zfs list -t all
> NAME                        USED  AVAIL  REFER 
> MOUNTPOINT
> sysvol                      919G      0 
12.5G 
> /sysvol
> sysvol/home                 906G      0   898G 
> /sysvol/home
> sysvol/home at remrep-Week16  8.53G      -   898G  -
>
> I can see a possible sector size diff or recordsize affecting a few
bytes,
> but 400G is a bit excessive. The fact that it more closely matches the
> full
> dataset+snapshots, IMHO, is much more telling.
>
> # zfs get all sysvol/home | grep used
> sysvol/home  used                 
> 906G                   -
> sysvol/home  usedbysnapshots      
> 8.53G                  -
> sysvol/home  usedbydataset        
> 898G                   -
> sysvol/home  usedbychildren       
> 0                      -
> sysvol/home  usedbyrefreservation 
> 0                      -
> sysvol/home  logicalused          
> 574G                   -
>
> logical used is actual used, correct?   Why is it the 'full' amount,
when
> only one snapshot was replicated?
>
> So I thought maybe it's not reporting correctly
>
> # zfs list
> NAME          USED  AVAIL  REFER  MOUNTPOINT
> sysvol        907G  12.3G   256M  /sysvol
> sysvol/home   906G  12.3G   898G  /sysvol/home
>
> # dd bs=1M count=12560 if=/dev/zero of=test2
> dd: test2: No space left on device
> 12558+0 records in
> 12557+1 records out
> 13167886336 bytes transferred in 33.499157 secs (393081126 bytes/sec)
> # zfs list
> NAME          USED  AVAIL  REFER  MOUNTPOINT
> sysvol        919G      0  12.5G  /sysvol
> sysvol/home   906G      0   898G  /sysvol/home
> # dd bs=1M count=12560 if=/dev/zero of=test3
> dd: test3: No space left on device
>
> So what's going on?  Is this a known issue?
> I suppose I can take the new server down to the colo and replicate from
> the
> original, but that doesn't resolve the 'restore from backup' issue that I
> see happening...
> _______________________________________________
> freebsd-fs at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-fsTo unsubscribe, send
> any mail to "freebsd-fs-unsubscribe at freebsd.org"