Abstract:
Lazy-Koala is a low overhead Kubernetes framework to collect, store and process telemetry data using deep learning to help system operators detect anomalies earlier to reduce the MTTR when the system is experiencing an anomaly. To achieve that this project contains three components. Firstly, it comes with a telemetry data collector called gazer which relies on Linux eBPF to perform automatic service instrumentation. Secondly, there is a component that uses deep learning to analyze those collect telemetry data to detect anomalies. Last but not least the core of this project is a custom Kubernetes operator which manages all the resources and provides a simple UI for the user to use this system easily.
