kubernetes: Dockerd process leaking handles in K8s Windows worker nodes

What happened:

The dockerd process is leaking handles (~2M) in Kubernetes Windows worker nodes, leading to huge (lot more than usual) memory and cpu consumption.

What you expected to happen:

The process is not expected to open so many handles as it is doing now. It’s usual inter-process communication is probably expected to open lot less than 2 million handles.

How to reproduce it (as minimally and precisely as possible):

Windows server minimal installation with Docker and Kubernetes should be able to reproduce this issue.

Anything else we need to know?:

The handle leak goes on increasing with time until and unless the system is rebooted for maintenance, which implies unless the dockerd process is restarted.

Environment:

• Kubernetes version (use kubectl version): 1.19.11
• Cloud provider or hardware configuration: Hyper-V Virtual Machine
• OS : Windows Server 20H2 SAC
• OS Version: 10.0.19042
• Install tools: Docker
• CNI : Calico (Vxlan)

####Other details:

`PS C:> get-process -Name dockerd

Handles NPM(K) PM(K) WS(K) CPU(s) Id SI ProcessName

---

1953515 27 145448 78128 250,658.44 2980 0 dockerd `