dave at tamos
Jan 28, 1998, 9:46 AM
Post #1 of 1
High volume mailing list anyone? Hopefully, this will urge more people
to get involved.
I've started developing for a cluster networking protocol for
synchronizing nodes and for sending heartbeat packets. I don't have
much yet, but it's a start.
Anyway, in section 8 of the HOWTO, Harald talks about the behavior of
two nodes (as the easy case) and a chain of failures for network
interfaces. Well, this is easy for a two-node cluster. It's obvious
who gets to take over the new IP, but what happens when an n-node
cluster? What happens when at least two other nodes are configured to
provide that failed resource? Should there be some sort of voting going
on? Should there be a preferred chain of failure setup for a clustered
resource? (for speed maybe?) So what happens when the failed node comes
back up? =) I'm thinking maybe we ought to use IP aliasing for cluster
resources that are IP specific. Just a thought.