When reading books about high availability and fault tolerance, in general, I always get the impression that most systems work on the assumption that crashes are atomic. What do I mean by that? Let's imagine there are two servers; one server has crashed and the other takes over. Unfortunately, crashes are not like this in many cases. In the real world, crashes are often far from atomic. Trouble is likely to build up gradually until things go really wrong. Consider memory corruption; when the memory goes wrong, it might not lead to an instant crash, and even when it does, the system might restart again without problems before troubles return. In many cases, a problem that accumulates silently is way more dangerous than a simple crash. The main danger in the case of silent corruption is that problems could gradually spread in the system without anybody noticing.

To prevent problems related to silent corruption caused by memory consumption, PostgreSQL...