High CPU with Server 5.1

[MichaelNewbery]

Rather than reproduce everything here, may I direct you to the Apple Support Communities thread at https://discussions.apple.com/thread/7509348?

I wasn't sure it was a ZFS issue when I first posted the query, and I'm still not sure it is. However... If anyone here can offer guidance I'd appreciate it.

It used to work fine. It still works fine, except that since Server 5.1, servermgrd goes into a tizzy when it looks at the TM destinations and sits at 100% CPU forever (well, days).

Basically, I backup the family machines using Time Machine to a dedicated server. Each machine has its own TM Destination, which is a ZFS filesystem, which I snapshot daily.
The ZFS file systems are mounted under their own directory (/Wells), rather than than under /Volumes.
The whole server machine is then backed up offsite using Backblaze.

TM is a wonderfully convenient and painless way to back up the family machines. ZFS then keeps them safe against hardware failure (OWN Thunderbay 4, disks in ZFS mirror) and Backblaze maintains an offsite copy.

OS X version: Mac OS X 10.11.4 [15E65]
OS X Server version: Server 5.1
ZFS version: SPL: 1.4.5-1, ZFS:1.4.5-1

[Brendon]

For me - running OSX Server 5.1, it takes 30-60 seconds for the Time Machine tab to update itself, during which time there is some visible CPU load. Once displayed its fine though. There are only 3 backups hosted on this machine. The TM Backups are all on a ZFS dataset, its a thunderbolt 1/5 disk/raidz2 array on a mac pro, OSX 10.11.4, O3X close to master. Dunno if that helps or not.

Also after first update of that tab, it renders instantly thereafter.

Cheers
Brendon

[MichaelNewbery]

Cool! Thanks.

Are the TM backups in one dataset or multiple. It strongly appears as if it's having multiple datasets that makes the difference. What we don't know is if ZFS is also a necessary condition.

[Brendon]

one.

If you can prove that its ZFS, then feel free to raise an issue in the ZFS project. "Simple" test, reproduce symptoms (or not) on HFS.

Cheers
Brendon

[MichaelNewbery]

I upgraded to ZFS 1.5.2-1 (and thanks, as always!)

The high CPU seems to have stopped.

This may mean that the problem was with ZFS, but if so, it's fixed. However, since I was unable to reproduce the problem in a virtual machine, I'm not 100% convinced,

[MichaelNewbery]

Alas! I spoke too soon. servermgrd is back to its old tricks. So, either zfs is not the culprit, or whatever its interaction is with servermgrd, the latest upgrade did not fix the problem.

[lundman]

Does dtruss of servermgrd give any clues as to what its hung up on? Is it repeatedly querying ZFS due to some incompatibility?

[dabrahams]

ZFS then keeps them safe against hardware failure (OWN Thunderbay 4, disks in ZFS mirror) and Backblaze maintains an offsite copy.

OS X version: Mac OS X 10.11.4 [15E65]
OS X Server version: Server 5.1
ZFS version: SPL: 1.4.5-1, ZFS:1.4.5-1

I want to do this! But how did you get Backblaze to recognize your ZFS volumes? For me, ZFS volumes don’t appear in the listing in the BackBlaze GUI. I’m not running OSX Server at the moment; I used to, but it seemed worse for power consumption.

[MichaelNewbery]

I mount the ZFS filesystems other than under /Volumes. So, mount filesystems XXX/a and XXX/b as /XXX/a and /XXX/b rather than /Volumes/XXX/a, /Volumes/XXX/b