[Pacemaker] Pacemaker issues on Amazon EC2

[Andrew Beekhof]

On 18/06/2013, at 2:23 PM, Jon Eisenstein <jon at animoto.com> wrote:

> 
> On Jun 18, 2013, at 12:12 AM, Andrew Beekhof <andrew at beekhof.net> wrote:
> 
>> 
>> On 18/06/2013, at 1:46 PM, Jon Eisenstein <jon at animoto.com> wrote:
>> 
>>> 
>>> On Jun 17, 2013, at 11:31 PM, Andrew Beekhof <andrew at beekhof.net> wrote:
>>> 
>>>> 
>>>> On 18/06/2013, at 7:19 AM, Jon Eisenstein <jon at animoto.com> wrote:
>>>> 
>>>>> tl;dr summary: On EC2, we can't reuse IP addresses, and we need a reliable, scriptable procedure for replacing a dead (guaranteed no longer running) server with another one without needing to take the remaining cluster members down.
>>>> 
>>>> This is almost certainly the wrong approach.
>>>> Have you tried their Virtual Private Network feature?  This allows for the use of predictable IPs.
>>> 
>>> We've previously decided against it due to the performance impact of IPSec and the complications involved in managing the network routing against the rest of our stack. It's something that we're investigating more, but the timeline's a bit longer than our HA need here.
>> 
>> Ok.
>> 
>> First question - what version of corosync? 1 or 2?  Is cman in use?  Which distro?
>> I think 2 is the most likely to be made workable here.
> 
> The base distro is Ubuntu 12.04 (LTS), which appears to be Corosync 1.4.2, but we have a lot of leeway for building custom packages if needed. Same with cman -- it's not currently part of the stack, but if it's something that'll help, I'm all for adding it.

I would start with corosync 2.x and something really recent of pacemaker (the HEAD of my repo, that sort of recent).

The fun part will be that you probably can't use multicast and you don't want to use a node list.
But at least 2.x can allow the node list to be updated at runtime IIRC.