[DRBD-user] heartbeat and drbd / Failover / Failback

[Florian Haas]

> I have 2 nodes with 2 nics and a serial.  Working on Primary/Standby model.
>
>     /----- eth0 ------\
> node1 --- Serial ---- node2
>     \----- eth1 ------/
>
> node1 eth0 192.168.151.91
> node1 eth1 192.168.1.91
>
> node2 eth0 192.168.151.91
> node2 eth1 192.168.1.91
>
> VirtualIP on eth0 192.168.151.90
>
>
> drbd uses eth1.
> heartbeat uses all three and has ping nodes to test on eth0 and eth1.
node1 is the primary to start.
>
> If, for any reason, node2 becomes primary, I want node1 to be outdated
so it can not become primary again without human intervention.
>
> Here's why:
> if the eth1 nic fails on node1, heartbeat will rollover to node2 as
primary and drbd will disconnect leaving
> node1 -> StandAlone st:Secondary/Unknown ds:UpToDate/DUnknown
> node2 -> WFConnection st:Primary/Unknown ds:UpToDate/DUnknown
>
> Both are showing UpToDate and so heartbeat can bring up either as primary.
>
> If node2 fails before human intervention can correct the problem on
node1, heartbeat rolls node1 back to the primary.  That could cause loss
of data that was written to node2 but not written to node1 because of the
disconnected state.  I want to fail over to node2 but I don't want it
coming back to node1 without human intervention.  (btw I'm not
talking about auto_failback.  I already have that off)
>
> Also, in the case where a secondary node2 has eth1 fail, I'd like to
outdate node2 so heartbeat can't rollover to it until human intervention
corrects the situation.
>
> I'm guessing there must be so way of doing this already before I start
hacking it up.
>
> Feel free to tell me I'm all wet and what I should by doing.  I'm pretty
new to drbd.  I looking for failover and NOT auto_failback.  Is this
making any sense.

Now I understand your comment in my blog better than I did earlier. :-)
And, as you correctly suspected, dopd is your friend here.

Cheers,
Florian