[Linux-cluster] fence_tool -c -w join failing

[Gordan Bobic]

Just for reference, I've found out what was the problem that has been 
driving me nuts for a good few hours. It turns out I made a really lame 
schoolboy mistake.

Cluster components seem to rely heavily on the /etc/hosts file. If 
cluster.conf seems to be OK and all other settings check out, it could 
just be that there are incorrect references in the hosts file. I had 
cman_tool status reporting the node IP to be wrong, and it didn't tally 
up with any of the interfaces which were correctly set in cluster.conf 
(I changed the IP address since I first set up the cluster). It all 
seems to fail rather silently and without warnings. Correcting the 
/etc/hosts file makes the problem go away.

Anyway, I hope this comes in useful to somebody googling the archives in 
the future.

Gordan

Gordan Bobic wrote:
> Christine Caulfield wrote:
>> gordan at bobich.net wrote:
>>> Hi,
>>>
>>> I have a weird problem that I can't figure out. I have a two node
>>> cluster (<cman two_node="1" expected_votes="1">), which means that a
>>> node should be quorate on it's own, and there should be no problems for
>>> it to join it's own fence domain (presumably).
>>>
>>> But when I do:
>>> service cman start
>>>
>>> All the components up to fenced start, but:
>>> fence -c -w join
>>> waits until it times out and never starts.
>>>
>>> The logs don't seem to reveal anything other than that fenced failed to
>>> start. What are the likely causes of this kind of a problem? I don't
>>> have any real fencing devices defined (yet) in my cluster.conf, only the
>>> following stub entry:
>>>
>>> <fence>
>>>     <method name="1">
>>>         <device name = "foo"/>
>>>     </method>
>>> </fence>
>>
>> That's probably why. It's trying to fence the non-member nodes and its
>> failing because you don't have any valid fence devices.
> 
> I'm not sure that's right, because I've had dozens of clusters work 
> without fencing devices in various prototyping environments. So how does 
> one start up an incomplete cluster without any fence devices?
> 
>> Try starting fenced with '-D', it'll show more debugging information.
> 
> I am not starting fenced directly, I'm doing it via
> fence_tool -c -w join
> 
> That is supposed to start up fenced, according to documentation, but it 
> doesn't include the -D option.
> 
> Gordan
> 
> -- 
> Linux-cluster mailing list
> Linux-cluster at redhat.com
> https://www.redhat.com/mailman/listinfo/linux-cluster