gistfile1.txt

apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: tpu-network-debug
  namespace: kube-system
  labels:
    k8s-app: tpu-network-debug
spec:
  selector:
    matchLabels:
      k8s-app: tpu-network-debug
  template:
    metadata:
      labels:
        k8s-app: tpu-network-debug
    spec:
      nodeSelector:
        # Replace with actual TPU VM name
        kubernetes.io/hostname: gke-tpu-fa1a3027-sf75
      priorityClassName: system-node-critical
      hostNetwork: true
      hostPID: true
      tolerations:
      - operator: "Exists"
        effect: "NoExecute"
      - operator: "Exists"
        effect: "NoSchedule"
      containers:
      - name: "tpu-network-debug"
        image: "ubuntu:latest"
        securityContext:
          privileged: true
        command:
        - bash
        - -c
        - |
          set -e
          INTERFACE=eth0
          iteration=0

          apt update -y
          apt install sysstat -y
          apt install ethtool -y
          apt install net-tools -y
          apt install iproute2 -y
          echo "Printing interface GRO settings"
          ethtool -k $INTERFACE
          while true
          do
            echo "Loop iteration timestamp: $(date -u)"

            echo "##### MPSTAT OUTPUTS #####"
            mpstat -P ALL

            echo "##### TRAFFIC STATS #####"
            # Reading from /proc/net/dev now relies on hostPID:true
            (for i in {1..10}; do sleep 0.1; echo $(date +%s.%N) $(grep $INTERFACE /proc/net/dev); done) | awk '{db=$3-b; if (db==0) next; b=$3; dt=$1-t; t=$1 ;printf "dt %8.6f  Mbps %12.3f\n", dt, 1e-6*db*8/dt}'

            iteration=$((iteration+1))
            if [ $((iteration % 10)) -eq 0 ]; then
              echo "##### OTHER NETWORK STATS #####"
              nstat
              ss -tmoi
              tc -s -d qdisc sh dev $INTERFACE
              
              echo "##### ETHTOOL -S OUTPUTS #####"
              ethtool -S $INTERFACE
            fi
            
            sleep 5
          done