Re: [Nbd] nbd-client: timeout option not working

[Paul Clements <paul.clements@...856...>]

On Wed, Jul 27, 2011 at 12:13 PM, Valerio Pachera <sirio81@...17...> wrote:
> 2011/7/21 Valerio Pachera <sirio81@...17...>:
>
>> If I unplug the ehternet cable from pc1, the guest is like frozen, the
>> kvm process goes 100% cpu, and the nbd device never fail.
>
> am I the only one with this problem?
>
> I also test raid1 and and directly on my laptop.
> Two computer with nbd-server, nbd-clinet on my laptop that run mdadm
> on nbd0 and nbd1.
> When I unplug one of the nbd-server, it's not possibile to write
> anything on the md0 device.
> It's not even possible to 'cat /proc/mdstat'.
>
> This time I used the right sintax for timeout option:
> nbd-client -timeout 3 nbd-server1 2000 /dev/nbd0
> nbd-client -timeout 3 nbd-server2 2000 /dev/nbd1

Which kernel? If you wait a while, does it eventually fail?

I have seen cases on recent kernels where it helps to remove:

-               printk(KERN_ERR "%s: Attempted send on closed socket\n",
-                      lo->disk->disk_name);

from nbd.c and rebuild nbd.ko. That makes nbd's time to fail much
shorter. Unfortunately, there appears to be some slow synchronous
behavior in the kernel/syslog path.

I need to commit that upstream actually...

--
Paul