Thinking of using ZFS/FBSD for a backup system

[Kris Kennaway]

CZUCZY Gergely wrote:
> Yes Kris, but you've forgot something quite important.
> What you've just showed is -CURRENT, and how does that thumb-rule is
> about branches and (semi-)production systems? 
> My faint memories say something like "don't never ever even think of
> running -CURRENT on a production box", in a polite way.
> ZFS can be stable on -CURRENT but it's till -CURRENT, with its issues
> as a production system. So, the last we can go about a backup box is
> -STABLE, but i also wouldn't prefer that one, if I can. -RELEASE and
> patches for production, to be safe.
> 
> Give us a stable ZFS in -RELEASE and -STABLE and we will be statisfied
> and happy. -CURRENT is still not a way for production boxes, that's
> asking for trouble.

It's not relevant that I am running -CURRENT, there have been no changes 
in ZFS that are not also in -STABLE (and only one bug fix since 
7.0-RELEASE, I think -- that was important, but it fixes mmap 
corruption, not a panic).

I run -CURRENT to help debug it, but I am neither making use of ZFS 
fixes, nor encountering ZFS bugs.

> I've finetuned ZFS as much as I could, I've read every little tiny bit
> of hint/information/whatever that was available and I couldn't get rid
> of those kmem_size panics in -RELEASE and -STABLE.

Well, it's still almost certainly because you aren't setting kmem_size 
high enough.  As you saw, that is the only thing I tuned (disabling 
prefetch is just for performance in my environment).

If you can't set it high enough because you don't have enough RAM, that 
means your system does't have enough RAM to run ZFS, not that ZFS is 
unstable.

Kris