[erlang-questions] Forming a cluster

[Rick Pettit]

On Wed, Oct 24, 2007 at 09:48:09AM -0400, Jack Orenstein wrote:
> What is the recommended procedure for setting up a cluster of nodes,  
> each running an Erlang VM (all with the same .erlang.cookie)? The  
> nodes are all in the same subnet, behind a firewall. The node  
> population is reasonably static, but nodes there may be occasional  
> additions. When a node joins, the nodes already in the cluster need  
> to know about it. Each node has to know about every other node, (e.g.  
> by calling nodes()).
> 
> net_kernel:monitor_nodes will indicate when a node joins, but only if  
> the new node sends a message to some node in the cluster. How does  
> the new node know who is present? There seems to be a bootstrapping  
> problem. Also, requiring a new node to know about a cluster member is  
> problematic since the selected node might be down.
> 
> net_adm and the .erlang.hosts file is a possible solution, but that  
> seems static. It doesn't seem to accomodate nodes joining after the  
> initial cluster forms.
> 
> Is there a dynamic solution that doesn't require the new node to know  
> anything about the existing cluster members?

Most systems I've worked on involved a shared set of mnesia database nodes
containing configuration information for the entire cluster.

All other nodes in the cluster are started with -mnesia extra_db_nodes [...],
and thus no explicit net_adm:ping/1 is required (in my code).

Of course this is just another form of static configuration...

To do what you want you may need to take a close look at how epmd(1) works.

-Rick