Strange Loop

Resilient Microservices without the Chaos

Microservice applications introduce additional complexity: partial failure can render one or more required dependencies of a given service unavailable at any time. Therefore, we should ensure our applications are resilient to partial failure before we deploy them.

In this talk, we will discuss a new approach for resilience testing, Service-Level Fault Injection Testing, and a tool that implements it, called Filibuster. Filibuster verifies microservice application behavior under failure, starting from the existing functional tests that your organization is already writing. Filibuster combines static analysis, test synthesis, and principled fault injection to identify bugs before code ships to production. We will not only discuss the core algorithms behind Filibuster, but also discuss the challenges of technical transfer of academic code: from adapting algorithms to practical implementations of academic code based on a large microservice deployment that powers a popular app.

Christopher Meiklejohn

Ph.D. Student

Previously a practitioner, Christopher worked on large-scale distributed systems at Basho, Mesosphere, and Machine Zone. Before that, he managed software development for Berklee College of Music, and worked in a data center where he supported the Boston Marathon and the Kraft Group (i.e. New England Patriots and the New England Revolution.)