zfs filesystem problem

[jhell]

On 06/07/2010 01:46, Sergiy Suprun wrote:
> On Mon, Jun 7, 2010 at 06:45, Jurgen Weber <jurgen at ish.com.au> wrote:
> 
>> Hello
>>
>> I have a FreeBSD 8.0-p2 system, which runs two pools. One with 6 disks all
>> mirrored for our data and another mirrored pool for the OS. The system has
>> 16GB of RAM.
>>
>> I have a nightly cron script running which takes a snapshot of a particular
>> file system within the storage pool. This has been running for just over a
>> month now without any issues until this weekend.
>>
>> Now we can not access the mentioned file system. If we try to `ls` to it or
>> `cd` into it the shell locks up (not even kill -9 can stop the `ls`
>> processes, etc) and top shows that the process state is `zfs`.

This is most likely caused by some bugs that were found and fixed in
stable/8. One of the commits that mm@ made has touched that zio->iowait
that you should see your processes are stuck in.

There still seems "at least in my case" some zio->iowait problems going
on but I have not pinned that down to the cause yet, but they have not
caused any of my system proccesses to freeze in that state.

Grab a kernel from one of the snapshots that were made sometime last
month to test this out just to be sure so your not upgrading for no
reason. When I say kernel I mean kernel & modules that go with it as ZFS
is a module and you will obviously need that.

Please report back on your findings if the kernel from stable fixed your
problem.

URL to retrieve snapshots: http://bit.ly/aLoXXV

Good Luck!,


<Irrelevant>
> Hello.
> How about scrub ?
> And which size of your pools and how many place used by data+snapshots?
</Irrelevant>.

-- 

 jhell