What happened?

When running two pods that both use hostNetwork and have sidecar initContainers (restartPolicy: Always) using the same port, kube-scheduler allows the pods to be scheduled on the same node.

What did you expect to happen?

kube-scheduler does not schedule these two pods on the same node, following the behavior for pods using hostNetwork ports via a main container.

How can we reproduce it (as minimally and precisely as possible)?

1. Create a kind cluster: kind create cluster. The cluster only has one node.
2. Apply p1:

   apiVersion: v1
   kind: Pod
   metadata:
     name: p1
   spec:              
     hostNetwork: true
     dnsPolicy: ClusterFirstWithHostNet
     initContainers:
     - name: init
       restartPolicy: Always    
       image: python:3.12                                                                                                                                                                                                            
       command:
       - /bin/bash
       - -c
       - python -m http.server 8081
       ports:
       - containerPort: 8081
     containers:
     - name: main
       image: python:3.12
       command:
       - /bin/bash
       - -c
       - sleep 10000
   

3. Apply p2:

   apiVersion: v1
   kind: Pod
   metadata:
     name: p2
   spec:
     hostNetwork: true
     dnsPolicy: ClusterFirstWithHostNet
     initContainers:
     - name: init
       restartPolicy: Always    
       image: python:3.12                                                                                                                                                                                                            
       command:
       - /bin/bash
       - -c
       - python -m http.server 8081
       ports:
       - containerPort: 8081
     containers:
     - name: main
       image: python:3.12
       command:
       - /bin/bash
       - -c
       - sleep 10000
   

4. p2 is scheduled (and crashing because p1 is already using the port).

Anything else we need to know?

kube-scheduler already accounts for hostNetwork ports used by main containers. For example:

1. Create a kind cluster: kind create cluster. The cluster only has one node.
2. Apply p1:

   apiVersion: v1
   kind: Pod
   metadata:
     name: p1
   spec:              
     hostNetwork: true
     dnsPolicy: ClusterFirstWithHostNet
     containers:
     - name: main
       image: python:3.12
       command:
       - /bin/bash
       - -c
       - python -m http.server 8081
       ports:
       - containerPort: 8081
   

3. Apply p2:

   apiVersion: v1
   kind: Pod
   metadata:
     name: p2
   spec:
     hostNetwork: true
     dnsPolicy: ClusterFirstWithHostNet
     containers:
     - name: main
       image: python:3.12
       command:
       - /bin/bash
       - -c
       - python -m http.server 8081
       ports:
       - containerPort: 8081
   

4. p2 is not scheduled: Warning FailedScheduling 3m7s default-scheduler 0/1 nodes are available: 1 node(s) didn't have free ports for the requested pod ports. preemption: 0/1 nodes are available: 1 No preemption victims found for incoming pod.

Kubernetes version

$ kubectl version
Client Version: v1.31.9
Kustomize Version: v5.4.2
Server Version: v1.31.1

Cloud provider

OS version

# On Linux:
$ cat /etc/os-release
# paste output here
$ uname -a
# paste output here

# On Windows:
C:\> wmic os get Caption, Version, BuildNumber, OSArchitecture
# paste output here

Install tools

Container runtime (CRI) and version (if applicable)

Related plugins (CNI, CSI, ...) and versions (if applicable)