Performance testing of microservices in action

Internet changed our lives. We no longer go to book store since there is Amazon. We no longer watch TV since there are YouTube and Netflix. We tend to migrate into Internet and this is generates huge amount of traffic. This brings us new challenges but old solutions fail to keep pace with changes.

During past few years microservice oriented design became standard for distributed high load systems. It’s been proven by lots of companies including FANG (Facebook, Amazon, Netflix and Google) so at this stage efficiency of this approach is indisputable. Unfortunately there is no silver bullet and with all good parts we’ve received operation complexity, configuration management complexity, difficulties with debugging and tracing, instability of interfaces and performance drawback caused by expenses of IPC.

Goal of this talk is to get through performance analysis process of simple calculator application implemented as a set of microservices with following FOSS toolset: k6 and gatling for traffic generation, prometheus for monitoring, perf Linux profiler and FlameGraph project for flame graphs generation.