App-Bisect: Autonomous Healing for Microservice-Based Apps

The microservice and DevOps approach to software design has resulted in new software features being delivered immediately to users, instead of waiting for long refresh cycles. On the downside, software bugs and performance regressions have now become an important cause of downtime. We propose app-bisect, an autonomous tool to troubleshoot and repair such software issues in production environments. Our insight is that the evolution of microservices in an application can be captured as mutations to the graph of microservice dependencies, such that a particular version of the graph from the past can be deployed automatically, as an interim measure until the problem is permanently fixed. Using canary testing and version-aware routing techniques, we describe how the search process can be sped up to identify such a candidate version. We present the overall design and key challenges towards implementing such a system.