[Linux-HA] stonith and failover problem

[Andrew Beekhof]

On 9/20/07, FG <fabrice.grelaud at u-bordeaux1.fr> wrote:
> Andrew Beekhof a écrit :
> > On 9/19/07, FG <fabrice.grelaud at u-bordeaux1.fr> wrote:
> >
> >> Hi,
> >>
> >> I use heartbeat 2.1.1 in an active/passive configuration.
> >>
> >> I'am testing differents failover and how stonith reacts ?
> >>
> >>
> >> When my active node is down (hardware failure or simply kill -9 PID(hb master)), Stonith with apcmastersnmp from my standby node shoot the active node and  resources are failover on the standby node. GOOD...
> >>
> >> Now my problem:
> >> 1- If I unplug the network card, pingd reacts and failover the resources on the other node BUT stonith doesn't shoot the active node before.
> >>
> >
> > how many communication paths do you have?
> > is heartbeat using (only?) the network you unplugged?
> >
> I've two communication paths. eth0 for network
> production and eth1 for
> heartbeat communication (+ serial line soon).
> When i unplug eth0, i got "ping node dead" and so
> the resources are
> failover to the standby node but without shooting
> the active node.

why would you want to?
heartbeat knows the other node is there and that its not running the
resource, stonith achieves nothing in this case

[snip - please dont include logs in-line]

>
> Do i need a particular constraint between pingd
> and stonith ?

as above - there is no point