New ZFSv28 patchset for 8-STABLE

[Attila Nagy]

  On 01/02/2011 05:06 AM, J. Hellenthal wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 01/01/2011 13:18, Attila Nagy wrote:
>>   On 12/16/2010 01:44 PM, Martin Matuska wrote:
>>> Link to the patch:
>>>
>>> http://people.freebsd.org/~mm/patches/zfs/v28/stable-8-zfsv28-20101215.patch.xz
>>>
>>>
>>>
>> I've used this:
>> http://people.freebsd.org/~mm/patches/zfs/v28/stable-8-zfsv28-20101223-nopython.patch.xz
>>
>> on a server with amd64, 8 G RAM, acting as a file server on
>> ftp/http/rsync, the content being read only mounted with nullfs in
>> jails, and the daemons use sendfile (ftp and http).
>>
>> The effects can be seen here:
>> http://people.fsn.hu/~bra/freebsd/20110101-zfsv28-fbsd/
>> the exact moment of the switch can be seen on zfs_mem-week.png, where
>> the L2 ARC has been discarded.
>>
>> What I see:
>> - increased CPU load
>> - decreased L2 ARC hit rate, decreased SSD (ad[46]), therefore increased
>> hard disk load (IOPS graph)
>>
>> Maybe I could accept the higher system load as normal, because there
>> were a lot of things changed between v15 and v28 (but I was hoping if I
>> use the same feature set, it will require less CPU), but dropping the
>> L2ARC hit rate so radically seems to be a major issue somewhere.
>> As you can see from the memory stats, I have enough kernel memory to
>> hold the L2 headers, so the L2 devices got filled up to their maximum
>> capacity.
>>
>> Any ideas on what could cause these? I haven't upgraded the pool version
>> and nothing was changed in the pool or in the file system.
>>
> Running arc_summary.pl[1] -p4 should print a summary about your l2arc
> and you should also notice in that section that there is a high number
> of "SPA Mismatch" mine usually grew to around 172k before I would notice
> a crash and I could reliably trigger this while in scrub.
>
> What ever is causing this needs desperate attention!
>
> I emailed mm@ privately off-list when I noticed this going on but have
> not received any feedback as of yet.
It's at zero currently (2 days of uptime):
kstat.zfs.misc.arcstats.l2_write_spa_mismatch: 0