Re: init scripts problems ?

"Also sprach denis bonnenfant:"
> I tried with +1, +2, and nothing, and I can't see any change !

OK - it needs some analysis then.

> But I'm in doubt :
> 
> when killing enbd-client, nda is marked faulty by enbd.

Well, it sends SET_FAULTY to the array, yes.

> So it remains in the array. But in this case, it must be totally removed from the array, as 

Why? We will do a HOT_ADD when contact is restored.

> everything related to enbd is going to be removed from the kernel.

Eh? No it isn't - nothing is removed or will be remved.

> This is a deadlock.

No it isn't! What is the deadlck supposed to be over?

> So i think that in this case, one have to make md to really unbind nda before suppressing it.

Are you saying that the md device wants to talk to nbd AFTER it has
been set faulty? That's clearly not part of any feasible MD design :-).
I don't think you are warm!

> setting SET_FAULTY is not sufficient.

It is sufficient. Thee is no mystery here- if you see clients that
should be dead not dead, then one has to work out why. So strace them as
they die (I presume you kill them or cause them to die by starting ne
daemons).  That will tell part of the story.  With that trace in hand we
can then add debugging to continue the trace in kernel.

> But how to do that 

> ? One have to do the same as mdadm --remove /dev/md0 /dev/nda before really destroy nda

No. No destruction is wanted.

Peter