[Linux-cluster] clvmd startup timed out

[Patrick Caulfield]

Kadlecsik Jozsi wrote:
> On Thu, 30 Aug 2007, Patrick Caulfield wrote:
> 
>>> This is simply ridiculous: after rebooting one of the machines, the same 
>>> behaviour appeared again: clvmd seems to wait for something indefinitely, 
>>> thus vgchange cannot proceed. Nothing interesting when clvmd started with 
>>> the debugging enabled. 
>> What do you define as "nothing interesting". It might not win a pulitzer 
>> prize but it could be an important clue ...
> 
> root at web1:~# clvmd -d
> CLVMD[b7dec6c0]: Aug 30 13:39:15 CLVMD started
> CLVMD[b7dec6c0]: Aug 30 13:39:15 Connected to CMAN
> CLVMD[b7dec6c0]: Aug 30 13:39:19 CMAN initialisation complete
>  
>> It might also be worth enabling LVM debugging in /etc/lvm.conf. If you have
>> clvmd from 2.02.28 then those values will also propogate to clvmd.
> 
> Nothing more emitted by clvmd. But lvm generated some log, I attached the 
> file (level = 7).
> 


So it's waiting for a DLM lock for the VG "data".

At this point it could be

a) another node is holding a lock or
b) A DLM bug, probably similar or identical to
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=238490


for a)

# mount -t debugfs debugfs /debug
# cat /debug/dlm/clvmd

and look for that lock and see which node is holding it.

for b)

Check the kernel you have and perhaps apply the patch shown in that bugzilla
comments 18 or 19.


Patrick