Communications of the ACM

practice

I L L U S T R A T I O N B Y G R E G M A B LY

MODERN SERVER SOFTWARE is demanding to develop and operate: It must be available at all times and in all locations; it must reply within milliseconds to user requests; it must respond quickly to capacity demands; it must process a lot of data and even more traffic; it must adapt quickly to changing product needs; and, in many cases, it must accommodate a large engineering organization, its many engineers the proverbial cooks in a big, messy kitchen.

What’s more, the best computers for these applications— whether you call them clouds, datacenters, or warehouse computers—are really bad. They are complex and unreliable, and prone to partial failures. They have asynchronous interconnects and deep memory hierarchies, and leave a lot of room for operator error. 1

Cloud computing thus forces us to confront the full complexity of distributed computing, where seemingly simple problems require complicated solutions.

While much of this complexity is inherent—the very nature of the problem precludes simpler solutions— much of it is also incidental, a simple consequence of using tools unfit for the purpose.

At Twitter, we use ideas from functional programming to tackle many of the complexities of modern server software, primarily through the use of higher-order functions and effects. Higher-order functions, or those that return other functions, let us combine simpler functions to define more complex ones, building up application functionality in a piecemeal fashion. Effects are values that represent some side-effecting operation; they are used in conjunction with higher-order functions to build complex effects from simpler ones. 7

Higher-order functions and effects help build scalable software in two ways: First, building complex software from simple parts makes it easy to understand, test, reuse, and replace individual components; second, effects make side-effecting operations tractable, promoting modularity and good separation of concerns.

This article explores three specific abstractions that follow this style of functional programming: futures are effects that represent asynchronous operations; services are functions that represent service boundaries; and filters are functions that encapsulate application-independent behavior. In turn, higher-order functions provide the glue used to combine these to create complex systems from simple parts.

Futures, services, and filters are thus combined to build server software in a piecemeal fashion. They let programmers build up complex software while preserving their ability to reason about the correctness of its constituent parts.