Many thanks to Graeme Fowler, whose suggestions led to the solution.
On Wed, Jun 28, 2006 at 04:02:14PM +0100, Graeme Fowler wrote:
> Decomplicate things a bit:
> 1. Remove the VRRP authentication completely (if you feel you can).
I wasn't aware that this worked.
> 2. You only have a single VRRP instance; you don't need a
> vrrp_sync_group. Comment it out.
And I wasn't aware that this was possible either.
> Start the one assumed to be MASTER.
> On the server assumed to be BACKUP, use tcpdump to make sure you can see
> the announcements - "tcpdump -qn net 126.96.36.199/8" should see you on your
> way with this. If you can, make sure that you don't have iptables rules
> sitting there getting in the way of them.
I was unaware of this; it showed that there were two machines using
the same vrid, and the other was of course my real server -- the
servers I am experimenting with are test servers. Hence:
> Also, as you don't describe your environment, make sure nobody else on
> the same network is doing VRRP. If they are, think about your
> inter-server communication by considering putting the announcements on a
> private or protected network instead.
Perceptive. I changed the vrid to another number and it worked fine.
On Wed, Jun 28, 2006 at 12:43:48PM -0700, Brad Dameron wrote:
> Your virtual_router_id on the backup needs to be incremented.
When I follow this advice, it stops working. Apparently the
virtual_router_id needs to be the same on both machines. So, the
failover is now working, with about a one second interval.
I presume the advert_int could be used to adjust this -- 1 second
is ideal for me in any case.
Many thanks for your help in solving this problem.