LQDeployConfig/kubernetes/mysql-master-slave/V1/troubleshoot.sh

#!/bin/bash

echo "=== MySQL主从故障排查 ==="

echo "1. 查看所有资源状态:"
kubectl get all -l app=mysql

echo -e "\n2. 查看Pod详细状态:"
kubectl describe pods -l app=mysql

echo -e "\n3. 查看Pod日志:"
for POD in $(kubectl get pods -l app=mysql -o name); do
  echo -e "\n=== $POD 日志 ==="
  kubectl logs $POD --tail=20
done

echo -e "\n4. 检查网络连通性:"
MASTER_POD=$(kubectl get pod -l role=master -o name 2>/dev/null | head -1)
if [ ! -z "$MASTER_POD" ]; then
  echo "从从节点测试连接主节点服务:"
  SLAVE_POD=$(kubectl get pod -l role=slave -o name | head -1)
  if [ ! -z "$SLAVE_POD" ]; then
    kubectl exec $SLAVE_POD -- sh -c "nc -zv mysql-master 3306 && echo '连接成功' || echo '连接失败'"
  fi
fi

echo -e "\n5. 检查MySQL进程:"
for POD in $(kubectl get pods -l app=mysql -o name); do
  echo -e "\n=== $POD MySQL进程 ==="
  kubectl exec $POD -- ps aux | grep mysql
done

echo -e "\n6. 手动检查复制状态:"
if [ ! -z "$SLAVE_POD" ]; then
  echo "从节点复制状态:"
  kubectl exec $SLAVE_POD -- mysql -uroot -proot123 -e "SHOW SLAVE STATUS\G" 2>/dev/null || echo "无法连接到MySQL"
fi