Wisdom-Data/deploy/health-check.sh

#!/bin/bash
# IoT Platform 健康检查脚本
# 用法: ./health-check.sh [host] [port] [max_wait_seconds]
# 返回: 0=健康, 1=不健康
#
# 检查策略:
#   1. 优先检查 /actuator/health (Spring Boot Actuator)
#   2. 如果 Actuator 返回 404，降级为检查 TCP 端口是否监听
#   3. 如果连端口都没监听，则判定为不健康

HOST=${1:-localhost}
PORT=${2:-8887}
MAX_WAIT=${3:-30}
HEALTH_URL="http://${HOST}:${PORT}/actuator/health"

INTERVAL=2
ATTEMPTS=$((MAX_WAIT / INTERVAL))

echo "[health-check] 开始检查: ${HOST}:${PORT}, 最多等待 ${MAX_WAIT} 秒"

for i in $(seq 1 $ATTEMPTS); do
    # 策略1: Actuator 健康端点
    RESPONSE=$(curl -s -o /dev/null -w "%{http_code}" --connect-timeout 2 --max-time 3 "${HEALTH_URL}" 2>/dev/null)

    if [ "$RESPONSE" = "200" ]; then
        BODY=$(curl -s --connect-timeout 2 --max-time 3 "${HEALTH_URL}" 2>/dev/null)
        if echo "$BODY" | grep -q '"status":"UP"'; then
            echo "[health-check] Actuator 健康检查通过 (${i}/${ATTEMPTS})"
            exit 0
        fi
    fi

    # 策略2: 如果 Actuator 404，降级为端口监听检查
    if [ "$RESPONSE" = "404" ]; then
        if nc -z "${HOST}" "${PORT}" 2>/dev/null || \
           ss -tlnp 2>/dev/null | grep -q ":${PORT} " || \
           netstat -tlnp 2>/dev/null | grep -q ":${PORT} "; then
            echo "[health-check] 端口监听检查通过 (${i}/${ATTEMPTS}) [Actuator 未启用，使用端口降级检查]"
            exit 0
        fi
    fi

    echo "[health-check] 第 ${i}/${ATTEMPTS} 次检查未通过 (HTTP ${RESPONSE}), ${INTERVAL} 秒后重试..."
    sleep $INTERVAL
done

echo "[health-check] 健康检查失败: 服务未在 ${MAX_WAIT} 秒内就绪"
exit 1