Microservices and Availability

I often hear the claim that microservices improve availability and reliability. It can be true when they are done well, but when these systems are built naively it actually reduces availability and reliability.

The networks that connect services and the services themselves are unreliable. If you require all your services to be available to function then adding additional services will reduce the availability of the system.

Take the example where service C depends on services A and B, either in breadth:

or in depth:

In terms of probabilities, if C requires both A and B to be available to function and, for example, we say at any point they each have a 0.9 independent probability of availability, C will effectively have a 0.9 x 0.9 x 0.9 = 0.73 probability of being available.

To put it another way, if your service relies on three services with 90% availability, you have a ceiling of 73% availability (on average - if you’re lucky and the failures overlap it’s still 90%, if you’re unlucky it’s 70%).

This gets worse quite quickly. If you rely on ten services with 90% availability, your service will never be available in the worst case!

In order to increase availability and improve reliability, C must be able to produce a useful result when A or B are not available.

I’m sure this will be obvious to many, but I have often encountered the opinion that microservices are magic and will somehow solve availability problems. In practice, I have seen them make availability worse more often than I’ve seen them make it better.

P.S. Microservices will not magically help you ‘scale’ either.

P.P.S. Please check my maths.