- Never configure fewer than two communication paths. And I mean, never. Never. If you do, that’s an accident (read: split brain) waiting to happen.
- On the device that you run your DRBD replication over, if it uses a direct back-to-back connection to the peer node (like it should), configure a
bcastlink. Simple and easy to configure, and since you’re not sharing that link with anything other than DRBD replication, those broadcast packets won’t harm or affect anything else.
- For the device that’s connected to your LAN, use
mcast. If you are deploying multiple clusters, make sure you choose a unique multicast address for every cluster. If you don’t, you will get confusing log messages about the node receiving unknown heartbeat packets. Also, don’t accidentally disable multicast traffic on that interface (it’s normally enabled by default).
- Optionally, if you have a direct serial connection available between nodes, add a
serialpath. Important note: serial links tend to be very flaky if used with Heartbeat 2 CRM-enabled clusters. If for some reason you must use a serial link between two CRM-enabled cluster nodes, be sure to set your baud rate to 115200 and test thoroughly. See the
Here is a sample configuration (note that
eth2 is the DRBD interface, and
bond0 is the one connected to the LAN):
mcast bond0 18.104.22.168 694 1 0 bcast eth2 keepalive 2 warntime 15 deadtime 30 initdead 120 crm yes node joe node jane
By the way, you needn’t necessarily use a bonding device here. That
bond0 might just as well be
eth1, or a even a non-Ethernet device. As long as you can run IP over it, it’s good. This is true for
serial, well, that’ll work only over a serial link.
And, of course, you never, ever want to reuse signing keys (as specified in
authkeys) across cluster boundaries. Together with a re-used multicast address, that’s a sure-fire recipe for disaster, so stay away from that. If you’re getting tired of thinking up authentication keys, look here.