Tracing processes in distributed container systems

Abstract

Techniques implemented by container services to provide users with utilization metrics indicating which processes running inside containers of distributed container systems are driving computing resource consumption. The container service may deploy agents in VMs alongside the containers that are supporting applications, and the agents may include profilers that inject eBPF programs into the kernels of each VM in which containers are running. The eBPF programs collect stack traces from the kernels that represent which processes were being executed when the stack traces were sampled. The profilers may use the stack traces to determine resource utilization for each process, and group the stack trace results based on the container in which the processes are executing. The utilization metrics may be converted into easily digestible visualizations and provided to a user to determine which processes are driving utilization in the containers, which in turn helps the users improve their application code.