[Pacemaker] override node name when using cman

[Andrew Beekhof]

On Sun, Sep 2, 2012 at 12:08 AM, Patrick H. <pacemaker at feystorm.net> wrote:
>
>
> Sent: Thu Aug 30 2012 23:00:25 GMT-0400 (EDT)
> From: Andrew Beekhof <andrew at beekhof.net>
> To: Patrick H. <pacemaker at feystorm.net> The Pacemaker cluster resource
> manager <pacemaker at oss.clusterlabs.org>
>
> Subject: Re: [Pacemaker] override node name when using cman
>>
>> On Wed, Aug 29, 2012 at 11:59 PM, Patrick H.<pacemaker at feystorm.net>
>> wrote:
>>>
>>> Sent: Wed Aug 29 2012 08:00:53 GMT-0400 (EDT)
>>> From: Andrew Beekhof<andrew at beekhof.net>
>>> To: The Pacemaker cluster resource manager<pacemaker at oss.clusterlabs.org>
>>> Subject: Re: [Pacemaker] override node name when using cman
>>>
>>>> On Wed, Aug 29, 2012 at 4:56 AM, Patrick Hemmer<pacemaker at feystorm.net>
>>>> wrote:
>>>>>
>>>>> It looks like when using pacemaker with cman, pacemaker gets the name
>>>>> of
>>>>> the
>>>>> node from the<clusternode>   'name' attribute. Is there any way to
>>>>> override
>>>>> this?
>>>>
>>>> Basically no.  Nodes need to be addressable by whatever name is
>>>> supplied.
>>>
>>> Can this be put in as a feature request? It doesn't make sense to call a
>>> node by the name of it's interface. Also cman is more than happy to let
>>> you
>>> put IPs in the<clusternode>  and<altname>  'name' attribute (making
>>> things
>>> less ambiguous is always a good thing). But Having to identify the node
>>> (in
>>> crm shell) by the IP address of one of it's interfaces is just not right.
>>
>> Wait. What now?
>> We don't require (or want) people to use IP addresses as node names.
>>
>> The requirement is that the 'name' attribute needs to match the output
>> of 'uname -n'.
>> Although I just posted a lengthy email about a plan to remove the
>> 'uname -n' requirement in most cases.
>
> I completely agree, using an IP as a node name is not desirable. But as
> pacemaker pulls the node name from the <clusternode> 'node' parameter,
> that's what happens.
>
> Maybe if I provided an example you'd see why I'm considering this a
> significant issue. So lets assume I have the following config:
> <clusternode nodeid='1' name='192.168.0.1'>
> <altname name='hanode01' />

It turns out that 'altname' is how you enable redundant rings in cman.
It doesn't actually tell cman to call the node something else.

What I would recommend at this point is dropping the altname tag,
grabbing the patch/version below and using name="'hanode01-eth1".
With the patch, pacemaker should no longer care that it doesn't match
uname -n (be sure to change any location constraints and clean up the
nodes section though).

   https://github.com/beekhof/pacemaker/commit/82533dc

> </clusternode>
> <clusternode nodeid='2' name='192.168.0.2'>
> <altname name='hanode02' />
> </clusternode>
>
> Now in the above example, the 'hanode01' has 2 interfaces, the eth0 public
> interface, and an eth1 secondary interface. The eth1 interface is connected
> via crossover to 'hanode02', and I want to use this link as the primary link
> for communication between machines (maybe it's a faster connection, maybe
> it's more reliable, or something else). As such this interface has to be
> listed in the <clusternode> 'name' attribute (and not <altname>) for it to
> be used as the primary interface. In this example I put the IP in the config
> (but I could also just add an entry to /etc/hosts with '192.168.0.1
> hanode01-eth1' and use 'hanode01-eth1' as the name). So pacemaker ends up
> using '192.168.0.1' as the name of the node. If I were to put a
> 'hanode-eth1' entry in /etc/hosts, pacemaker would still end up using
> 'hanode-eth1' as the node's name. Neither one of these is the value of
> `uname -n` or `hostname`.
>
>
>>
>>>>> If the node has multiple interfaces, I may want corosync to use an
>>>>> interface
>>>>> other than the node's main interface as the main interface for
>>>>> communication
>>>>> (like if I had a crossover cable between nodes). If I use the IP of the
>>>>> second interface then pacemaker uses that IP as the node name. If I
>>>>> give
>>>>> the
>>>>> second interface a name, it still ends up using that name instead of
>>>>> the
>>>>> node's real name (`uname -n`).
>>>>>
>>>>> _______________________________________________
>>>>> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>>>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>>
>>>>> Project Home: http://www.clusterlabs.org
>>>>> Getting started:
>>>>> http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>>>> Bugs: http://bugs.clusterlabs.org
>>>>
>>>> _______________________________________________
>>>> Pacemaker mailing list: Pacemaker at oss.clusterlabs.org
>>>> http://oss.clusterlabs.org/mailman/listinfo/pacemaker
>>>>
>>>> Project Home: http://www.clusterlabs.org
>>>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf
>>>> Bugs: http://bugs.clusterlabs.org