Patterns For Structuring Stacks

A key challenge with infrastructure design is deciding how to size and structure stacks. The infrastructure for a given system may be defined as a single stack, or spread across multiple stacks. Shared code may also be used across stacks.

It’s very common to start by defining infrastructure in a single stack. However, this tends to grow over time, and can become an unmanageable monolithic stack. The other end of the spectrum is the micro stack pattern, which divides the infrastructure into small, loosely coupled stacks that can be easily changed independently.

Deciding on appropriate sizing and boundaries for splitting stack is a particular challenge. One common strategy is the single service stack pattern, where each stack includes the infrastructure specific to a particular application or service. An alternative is cross-stack service, where the infrastructure to run a single application is broken out across multiple stacks, for example splitting web servers, application servers, database servers, etc. each into their own stack.

In the other direction, multiple services are run in a single multi-service stack. For example, a number of applications may combine to provide a single overall service, such as search, in which case it could make sense to combine them in a single infrastructure stack, although this risks becoming a monolith.

[Shared nothing stacks] take the single service pattern to the extreme of ensuring that stacks do not have dependencies on other stacks. More commonly, stacks do have dependencies, for example a service stack’s infrastructure may use networking structures managed by another stack. [Stack integration patterns] is an important topic in infrastructure design.

It is common to simplify infrastructure design by having a stack provide infrastructure used by other stacks. Another technique is to share common infrastructure code as a stack module, imported into multiple stack projects.