[Pacemaker] Corosync over DHCP IP

Dennis Jacobfeuerborn dennisml at conversis.de
Tue Feb 12 16:28:44 EST 2013


On 02/12/2013 03:57 PM, Dejan Muhamedagic wrote:
> Hi,
> 
> On Tue, Feb 12, 2013 at 02:52:58PM +0100, Dennis Jacobfeuerborn wrote:
>> On 02/12/2013 03:15 AM, Andrew Beekhof wrote:
>>> On Tue, Feb 12, 2013 at 1:01 PM, Dennis Jacobfeuerborn
>>> <dennisml at conversis.de> wrote:
>>>> On 02/12/2013 02:38 AM, Andrew Beekhof wrote:
>>>>> On Tue, Feb 12, 2013 at 3:09 AM, Dennis Jacobfeuerborn
>>>>> <dennisml at conversis.de> wrote:
>>>>>> On 02/11/2013 11:30 AM, Andrew Beekhof wrote:
>>>>>>> On Mon, Feb 11, 2013 at 9:24 PM, Viacheslav Biriukov
>>>>>>> <v.v.biriukov at gmail.com> wrote:
>>>>>>>> It is VM in the OpenStack. So we can't use static IP.
>>>>>>>> Right now investigating why interface become down.
>>>>>>>
>>>>>>> Even if you solve that, dynamic IP addresses are fundamentally
>>>>>>> incompatible with cluster software.
>>>>>>> You're effectively trying to create a cluster out of nodes which
>>>>>>> change their name every time they boot.
>>>>>>
>>>>>> DHCP doesn't necessarily mean a dynamic IP.
>>>>>
>>>>> In most (if not all) openstack deployments, it does.
>>>>> Even better, the static IPs you can assign belong to the physical
>>>>> hosts and don't show up inside the guests - so corosync can't bind to
>>>>> them.
>>>>
>>>> You are probably talking about the floating IPs. The primary IPs (usually
>>>> in the 10.0.0.0/8 range) of the interfaces however should work fine for
>>>> this. There's no magic involved.
>>>
>>> If they don't show up in 'ip addr', which they don't in openstack
>>> guests, then corosync can't use them.
>>> There is no way to say "Your address is w.x.y.z but you'll get
>>> messages on a.b.c.d".
>>
>> They show up like like any other interface in 'ip addr'. There is no magic
>> involved in any of this.
>> The machine gets assigned a primary IP from the assigned network which is
>> stored in the dhcp leases database and usually (but not necessarily) it
>> gets assigned a floating ip that is reachable from the internet which is
>> NATed to the primary IP.
>> When the guest boots it discovers its primary IP via dhcp and assignes that
>> IP to eth0.
>> There is no reason why corosync shouldn't be able to use that IP to address
>> the cluster nodes.
> 
> As long as the lease time exceeds the node's uptime :)

The lease time is only relevant for "ad-hoc" ip assignments and not for
reserved/bootp dynamic lease entries. That's what I meant when I said dhcp
doesn't necessarily mean a dynamic ip. You can create a lease that is tied
to the MAC address of a system and then that ip will never be assigned to
another system.

Regards,
  Dennis






More information about the Pacemaker mailing list