Once we have an initial node count, we should apply these guidelines to adjust the total count and node distribution:
- If the initial PostgreSQL node count is even, add one witness node.
- If the initial PostgreSQL node count is odd, convert one replica into a witness node.
- In the presence of two locations, the witness node should reside in the same data center as the primary node.
- If possible, allocate witness nodes in an independent tertiary location.