Snapshot usage guidelines (to avoid stability issues)

[Cristian KLEIN]

Kostik Belousov wrote:
> On Thu, Jun 21, 2007 at 01:45:32AM +0300, Cristian KLEIN wrote:
>> Kostik Belousov wrote:
>>> On Wed, Jun 20, 2007 at 06:26:26PM +0200, Espen Skoglund wrote:
>>>> Hi,
>>>>
>>>> A couple of days ago I decided to start using snapshots on my system.
>>>> Knowing that the functionality has been available for quite some time
>>>> now I (apparently wrongly) assumed that it was ready for production
>>>> use.  My bad.
>>>>
>>>> My system, a 6-STABLE from week and a half ago, uses two 250GB
>>>> gstriped disks and has about 4-5 snapshots on one 215GB UFS partiton.
>>>> This morning I learned that things had gone terribly wrong during some
>>>> nightly cronjobs, hanging the whole system.  Suspecting that snapshots
>>>> were the culprit I soon learned after some investigation that the
>>>> snapshot functionality wasn't nearly as stable as I had hoped for.
>>>>
>>>> Looking at PRs and mailing lists there seems to be mainly two
>>>> outstanding stability issues with snapshots: a) snapshotted fs running
>>>> full, and b) deleting large amount of files on an fs with multiple
>>>> snapshots.  The former issue, while certainly annoying, one could be
>>>> able to work around.  The latter issue, on the other hand, seems more
>>>> like a definite show stopper.
>>>>
>>>> Am I right in inferring that the two above cases are main issues with
>>>> snaphots at this time, or are there other known gotchas that I have to
>>>> look out for.
>>> About the issue b). Are you system sources before or after 2007-06-11
>>> 10:53:48 UTC ? Or simply show me the version of sys/ufs/ffs/ffs_snapshot.c.
>>>
>>> If it is less then 1.103.2.24, see developers handbook for instruction on
>>> reporting deadlocks.
>>>
>> Do you think that 1.103.2.24 might solve this
>> http://lists.freebsd.org/pipermail/freebsd-fs/2007-May/003161.html ?
> I very much doubt it. AFAIR, I asked at least ddb backtrace for the crash,
> and you did not answered.

I'm really sorry, but I wasn't able to reproduce the problem on another
system. I tried taking / deleting many snapshots while running bonnie++,
but the panic would not occur.

I have 1,7mil inodes (4% utilization) on the server where the crash
occured. Might this be a prerequisite of the panic?