How is CP technically implemented in PostgreSQL? What happens when one of the nodes stops working?

In PostgreSQL, the CP (Consistency and Partition tolerance) concept is implemented through replication mechanisms and transaction consistency.

PostgreSQL uses WAL (Write-Ahead Logging) to ensure data consistency. In a cluster with replication (for example, using Patroni, Patroni + etcd, or other solutions), there is a master node (primary) and replicas (standby).

When one of the nodes stops working:

• If it is a replica, it simply stops accepting read requests, and the master continues to operate.
• If it is the master, a failover occurs: one of the replicas becomes the new master to maintain availability and consistency.

Transactions in PostgreSQL provide ACID, which guarantees data consistency. During network partitioning, the cluster can choose either availability or consistency, depending on the configuration and the solution used for cluster management.

Example: Patroni uses consensus via etcd or Consul to elect the master, ensuring a CP approach — consistency and resilience to network partitioning, but it may temporarily reduce availability during conflicts.