Causal clustering of multiple graphs

[jim]

Suppose that I have multiple completely disconnected graphs and I'd like to cluster each of them. Could this be done on a single cluster?

If each graph is in its own DB (assuming Neo4j 4 here), then I read from the clustering documentation that one can cluster all of the DBs and each replica "will have access to all of the DBs"; so I could do what I want if I assign each of the replicas to just one of those DBs. But are there performance considerations? Would each replica need to have disk-space/memory for all of the DBs, even though it will only ever access one of them? I guess my real question is what does "has access to" entail for the replicas.

[david_allen · Accepted solution]

Typically read replicas have copies of all of the graphs in a given cluster. You can set each read replica to have a different default graph if you wish though. You do need to have the disk space for all of the graphs, but in terms of memory, if the read replica only ever touches one graph, you would only need to plan for heap/page cache for that one graph. If you're not hitting the other graphs on that replica, there's no point in planning extra memory for them.

[jim]

Excellent, exactly what I was hoping to hear -- many thanks