Re: init scripts problems ?

Peter T. Breuer wrote:
> "Also sprach denis bonnenfant:"
> 
>>>>>But I'm not even sure the code sticks in OPEN, rather than at the first
>>>>>read. The WCHAN value might indicate.
>>>
>>>
>>>OK, i can reproduce it, even with only one nbd device open.
>>>
>>>in this case, WCHAN is page_sync for one enbd-client, and select for the 
>>>others.
>>
>>oups, it's sync_page !
> 
> 
> Yes, well, there you are - the client daemons are in perfectly normal
> select-on-socket activity, except for one of them which is in
> "sync_page" somewhere way down in the kernel.
> 
> I don't call that function (at least directly) but it sounds like the
> result of a sync from either the kernel or from user space, and maybe
> the device isn't communicating with the server yet, so it can't flush
> ...
> 
> Really, one needs a kernel trace at this point when it's stuck. Anyone
> out there know how to induce one?
> 
> Maybe you can tell me how you get it wedged and I can try looking at it
> with one of my debug consle kernels (shudder - the kernel debugger is
> NOT a pretty thing).

rather simple :


on client
# modprobe enbd_ioctl


use these enbd.conf :
on server :

server A-1 12345 /dev/sda -b 512
server B-1 12350 /dev/sdb -b 512


on client :

client A-1 /dev/nda ldap-croise 12345 -n 2 -m  -b 512
client B-1 /dev/ndb ldap-croise 12350 -n 2 -m  -b 512

then on server :

# /etc/init.d/enbd start server

and on client :

# /etc/init.d/enbd start client

... normally you are stuck (but not everytime!)