Multi-master PostgreSQL promises many exciting benefits that are hard to ignore. Despite this, there are some inherent limitations that could potentially prevent safe adoption within the application stack. Failure to account for these before deploying multi-master PostgreSQL could result in data loss or disruption of the cluster itself.
Consider, for example, what happens if two nodes accept a write for the same tuple. How should the conflict be resolved? Are there any necessary changes to the schema, or the application itself, before the adoption of the multi-master approach is safe? What else must we plan for?
While not an exhaustive resource, this recipe will attempt to answer many pressing concerns related to multi-master PostgreSQL. Some of these are based on theoretical cluster concepts.