Things always go wrong with individual components of the networks—whether it be power failures or cable cuts. Networks are typically architected at the hardware level using a set of redundant links and software is responsible for providing an ordered, reliable message pipe between two systems. This software is typically referred to as the networking layers (TCP/IP). Even with the layers trying to achieve sanity across unreliable links, often, applications get exposed to failures and need to handle things appropriately. 

This vulnerability is particularly acute for distributed data stores, and we will look at how to enable consistency in such scenarios in the Consistency section, later. For external calls, in Chapter 7, Building APIs, we will look at how to insulate application from network outages using a pattern called Hystrix.