On Tue, Sep 28, 2010 at 11:13 AM, Dejan Muhamedagic <[email protected]> wrote:
> Hi,
>
> On Tue, Sep 28, 2010 at 06:27:17AM +0200, Michael Schhwartzkopff wrote:
>> On Thursday 23 September 2010 13:47:43 Dejan Muhamedagic wrote:
>> > Hi,
>> >
>> > On Wed, Sep 08, 2010 at 09:26:40PM +0200, Jonathan Petersson wrote:
>> > > Hi all,
>> > >
>> > > I haven't been active on this list for quite some time but I recall
>> > > conntrack-support for heartbeat/pacemaker has been on the wall a few
>> > > times. As I was in the process of installing a couple of new firewalls
>> > > I figured I would spend some time actually getting some support for it
>> > > now that the resource-based system has been put in place (great work
>> > > btw).
>> > >
>> > > Please notice that the code-set is still work in progress and I'll be
>> > > spending the next few days expanding it.
>> >
>> > Any new developments in the meantime?

There's been some modifications, I'll put it under git during the week.

>> >
>> > > The code is available at: http://pastebin.com/Bv060JvR
>> > >
>> > > Feel free to reply with comments and recommended changes.
>> >
>> > Isn't conntrack supposed to be a master-slave implementation,
>> > i.e. where one instance sends updates to other instances? I don't
>> > know if migrate can be used instead of demote/promote.
>>
>> Hi,
>>
>> A MS RA for conntrackd is not nescessary. conntrack publishes its state table
>> via multicast. You start it on all nodes of your firewall cluster as a clone
>> resource. The firewall that has the floating IP address sees new entries in 
>> the
>> state table and published it. All other nodes get the new entries.
>
> So, you could just as well let conntrack start by the boot
> process, right? I always wondered on the relative merit of
> cloning such resources or starting them via init.

I guess it makes sense to leave out starting the daemon using the OCF
resource since the resource doesn't really maintain service-state of
the daemon, just failover.

>
>> Passive nodes just do not get traffic and thus do not publish new entries.
>
> I wonder why then there is migrate_to/from in the RA.
>
>> Of course, you could write a MS RA. But that would be too much work.
>
> Well, that doesn't seem to be needed.
>
> Thanks,
>
> Dejan
>
>> Greetings,
>>
>> --
>> Dr. Michael Schwartzkopff
>> Guardinistr. 63
>> 81375 München
>>
>> Tel: (0163) 172 50 98
>> _______________________________________________
>> Linux-HA mailing list
>> [email protected]
>> http://lists.linux-ha.org/mailman/listinfo/linux-ha
>> See also: http://linux-ha.org/ReportingProblems
> _______________________________________________
> Linux-HA mailing list
> [email protected]
> http://lists.linux-ha.org/mailman/listinfo/linux-ha
> See also: http://linux-ha.org/ReportingProblems
>
_______________________________________________
Linux-HA mailing list
[email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha
See also: http://linux-ha.org/ReportingProblems

Reply via email to