HEADS UP: New ZFS in the tree.

[Thomas Vogt]

Wes Morgan schrieb:
> On Thu, 4 Dec 2008, Thomas Vogt wrote:
> 
>> Hello Stefan
>>
>> Am 04.12.2008 um 00:33 schrieb Stefan Bethke:
>>> Am 03.12.2008 um 00:29 schrieb Peter Schuller:
>>>
>>>>> I've noticed the past couple of days, when using the server (not very
>>>>> often), every now and then, the GUI (KDE 4.1) will "hang" for up to 5
>>>>> minutes (no mouse movement, no keyboard events), while the drives 
>>>>> work like
>>>>> crazy.
>>>>
>>>> I was not explicit about it, but FWIW in my case the hang is not due
>>>> to drive saturation. The drives were mostly idle (except some stuff
>>>> triggered by a buildworld I had going) during the extended period of
>>>> ktorrent being unkillable. But again I never had this happen
>>>> pre-CURRENT.
>>>
>>> Just a very brief "me too" (but possibly different effect): I'm 
>>> stress testing two machines I put together over the weekend with an 
>>> endless loop of make -j4 universe, with /usr/obj on ZFS, with a 
>>> single disk. One of the two machines has now been stuck for a couple 
>>> of hours, and trying to access /tank results in a hung process, as 
>>> will zfs list.
>>>
>>> I'll reboot and see what happens, and if I can trigger it again, 
>>> willt try to produce more details.
>>>
>>> I have set
>>> vfs.zfs.arc_max="512M"
>>> vfs.zfs.prefetch_disable=1
>>> in loader.conf
>>>
>>> FreeBSD lokschuppen.lassitu.de 8.0-CURRENT FreeBSD 8.0-CURRENT #1: 
>>> Wed Dec 3 07:05:03 UTC 2008 
>>> root at lokschuppen.lassitu.de:/usr/obj/usr/src/sys/EISENBOOT  amd64
>>
>> Try to disable ZIL in loader.conf:
>> vfs.zfs.zil_disable="1"
>>
>> It helped me to stop deadlocks during rsync processes. As long as you 
>> don't run any databases or any fsync() intensiv applications,  i don't 
>> see any drawbacks in disabling zil. The drawbacks are for the 
>> applications itself not for ZFS. ZFS will be always consistent on disk 
>> due to its transaction model even without ZIL.
> 
> Are you sure about that? Without the ZIL, wouldn't you need to scrub 
> each pool after every crash, much as fsck with UFS?

AFAIK no. 
http://www.solarisinternals.com/wiki/index.php/ZFS_Evil_Tuning_Guide#Disabling_the_ZIL_.28Don.27t.29
The ZFS pool integrity itself is not compromised by this tuning.

There are many blog entries from ZFS developer with the same conclusion.

Regards,
Thomas