I'm not a big fan of conflict resolution, and LWW is a pretty bad choice in most cases. Conflict resolution means data loss (it violates the D of ACID, as an user expects that a transaction that was committed would endure, but it is deleted or overwritten by another one), and most likely C.
The effects of this are hard to predict and can hit pretty hard your business logic. This is not the kind of expectation you get from Postgres. This would fail Jepsen at second 1.
So I'd advise not to use it. The only use case I see for bi-directional replication is where you ensure that all the master nodes have disjoint sets of data, and the logical replication just aggregates them, without a possibility for conflicts.
If you really want multi-master you need a system that works on top of a consensus protocol. And there are definitely tools out there for this. But not LWW.
The effects of this are hard to predict and can hit pretty hard your business logic. This is not the kind of expectation you get from Postgres. This would fail Jepsen at second 1.
So I'd advise not to use it. The only use case I see for bi-directional replication is where you ensure that all the master nodes have disjoint sets of data, and the logical replication just aggregates them, without a possibility for conflicts.
If you really want multi-master you need a system that works on top of a consensus protocol. And there are definitely tools out there for this. But not LWW.