Bulletproofing with synchronous replication

Sometimes, in order to provide acceptable data durability, a high availability configuration must utilize synchronous commits. Beginning with PostgreSQL 9.1, database servers can now refuse to commit a transaction until the data is located on at least one alternate server. Unlike asynchronous replication where this is optional, synchronous replicas enforce this requirement to a fault.

Discussions in the PostgreSQL mailing list suggest that there is a long-standing misconception that synchronous replication is similar to RAID-1 operation. In RAID-1, the same exact data exists on two disks (or two disk sets), and if one of the pair fails, it continues to operate in degraded mode until the problem is addressed. This is absolutely not the case with PostgreSQL synchronous replication.

Unlike a RAID-1, PostgreSQL replicas can exist on different servers, on different networks, or even in different countries. PostgreSQL synchronous replication is a guarantee...