[Discuss] BTRFS

[hag at linnaean.org (Daniel Hagerty)]

Edward Ned Harvey <blu at nedharvey.com> writes:

> Anybody using btrfs in production?  I know it says all over it, "not ready
> for production" and so forth.  But it's like dangling a big piece of candy
> in front of a child with a sticker that says "Do not eat."   ;-)
>
>  
>
> I've had a somewhat bad experience, I'd like to share, and see if others
> reason, performance would grind to a halt, and some processes were
> unkillable, and stuff like that.  I suspected btrfs, but didn't have any

    Unkillable processes almost always involve a kernel bug.  The major
exception I'm aware of is nfs hard mounts, and even that problem is
sometimes viewed as a design error in the typical unix vfs layer.

    When you see a process that won't die, it's usually in ps's state
"D": an uninteruptible short term wait (think microseconds) that
obviously isn't short term.  Ask ps for the wchan of these processes.
You might need to ask for it wider, since the symbol is frequently long
and truncating.

    This symbol will point at the subsystem participating in your bug;
it may not be the responsible one, but involved.  Maybe the thing going
to sleep in the first place, maybe the thing that it was supposed to
wake it up, maybe the sleeper's caller.  Hopefully you get the idea; it
narrows down the field quite a bit.