Il giorno mer, 04/02/2009 alle 16.16 +0100, Michael Schwartzkopff ha
scritto:
> Am Mittwoch, 4. Februar 2009 12:27:44 schrieb Igor Neves:
> > Hi,
> >
> > I have done some work woth conntrackd and heartbeat a couple of time ago.
> >
> > Attached it's one conntrackd OCF script I made but when I finish I
> > realized that it was not working and would never work.
> > As you say in your HOWTO, conntrackd work with 2 caches.
> 
> I do start conntracd outside of heartbeat from init. So setup of sync is 
> already working before the cluster starts.
> Inside heartbeat I only dump the connection table from the cache into the 
> kernel (firewall starts) or clear the cache (firewall stops)
> 
I've also a 2-node active-standby firewall setup in production.
The problem with conntrackd is that it has only one sync connection with
the other node. To solve this SPOF I wrote two RA.
- the first one starts conntrackd and checks (in the monitor action) if
the other node is alive, otherwise, restarts conntrackd with another
configuration with another communication media.
- the second simply commits the conntrack tables from the other node
when it starts.
Obviously you must co-locate the second resource to an IP resource (or
in my case another custom RA that bridges some interfaces).
The two RA are still in a work-for-me status but they proved stable for
a while. Maybe in the next days I'll post them here to gather some
comments.

> If you want to write a OCF resource for that task to be done inside heartbeat 
> you need a stateful agent. You agent below is not stateful, i.e. it does not 
> unterstand promote and demote.
> 
> Re-thinking: Perhaps you also could state a conntrackd clone... 
In my implementation a clone (one for every node) of the table merging
RA is enough.

-- 
Michele Codutti
Centro Servizi Informatici e Telematici (CSIT)
Universita' degli Studi di Udine
via Delle Scienze, 208 - 33100 UDINE
tel +39 0432 558928
fax +39 0432 558911
e-mail: michele.codutti at uniud.it

_______________________________________________
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