Zombie python processes accumulating

[david-becher]

I noticed that the aks-mcp accumulates python zombie processes when run, e.g., in ca Kubernetes cluster, see the screenshot below ("Z" indicates Zombie process).

This can lead to process exhaustion and file descriptor exhaustion (worst case scenario). It is also logged as an K8s event when to many Zombie processes accumulate.

ZombieProcessAccumulation: 20 zombie processes on node (threshold: 20)

One way to mitigate this, is either to handle SIGCHILD signals and effectively makeing a wait syscall on these processes, or just using something like tini or dumb-init in the dockerfile, which acts as a init process and reaps zombie processes.

To reproduce: run it in a container, exec into the running container and execute top.

[david-becher]

Another suggestion: in https://github.com/Azure/aks-mcp/blob/main/internal/command/command.go#L44, don't forget to call cmd.Wait after cmd.Run has been called, that would effectively clean up the resources and you would not need tini or dumb-init anymore. Assuming these python3 antries linger around because of the az command being called (which I think might be the case).