When creating a highly available PostgreSQL server, one important element to consider is server load. One database server, no matter how powerful its hardware may be, cannot scale infinitely. Regardless of any frontend application-side caching, the database should be able to weather cache failures or unexpected demand.
We can offset much of this risk by leveraging database replicas. Each replica is available for read-only use, and applications are welcome to use them instead of the primary server. Unfortunately, as the amount of replicas increase, the application must track the connection settings for each, and it may even need to know which is currently configured as the primary server.
Server additions, ...
