optaplanner solving of one dataset with multiple nodes

I'm interested in deploying an optaplanner solution on a Kubernetes cluster. The main interest would be to scale horizontaly the solving of a specific dataset.

I have seen an example of optaplanner with a cluster (implemented thanks to an Optaplanner operator) but in my understanding it scales when multiple problems are submitted (one pod per problem)

So this is not what I'm looking for.

I'm looking for a way to "multithread" the computation of the problem (like with the move-thread-count) but accross multiple pods.

Do you know or have heard of any way to do it? I'm not even sure that Optaplanner supports it...

this answer says it's not yet supported natively but speak about doing it manually.

Do you have any tips to give for a start?
I would rather find a way to not do partitionned solving since it can miss the optimal solution..

That depends.

Are we talking about multi-bet solving? Essentially running the solver in N clones simultaneously, and picking the best single result out of them? If so, that's where the Operator can help as well.

If we're talking about solving the same problem in a distributed manner, with different machines executing different moves and then incorporating them into one single solution, then no, there is no out-of-the-box solution for that, and I haven't seen any custom implementations of that either.

Honestly, I don't see the benefits of that - the async overhead (and network latency!) will kill performance. The multi-threaded solver already suffers from that, and that's running on a single machine.

My opinion: Having 1 pod run N threads makes just so much more sense to me than having N pods run 1 thread.