hy-app-monitor/localbox/check_local_services.sh

#!/usr/bin/env bash
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
# shellcheck disable=SC1091
source "$SCRIPT_DIR/common.sh"

load_local_env
wait_for_docker

APP_SERVICES=(
  auth
  gateway
  wallet
  other
  live
  order
  console
  scheduling
  external
  visual-monitor
  golang
)

INFRA_SERVICES=(
  mysql
  redis
  mongo
  nacos
  zookeeper
)

if ! local_rocketmq_disabled; then
  INFRA_SERVICES+=(
    rocketmq-namesrv
    rocketmq-broker
    rocketmq-proxy
  )
fi

FAILURES=0

container_name_for_service() {
  case "$1" in
    visual-monitor) printf '%s\n' "chatapp3-local-visual-monitor" ;;
    rocketmq-namesrv) printf '%s\n' "chatapp3-local-rocketmq-namesrv" ;;
    rocketmq-broker) printf '%s\n' "chatapp3-local-rocketmq-broker" ;;
    rocketmq-proxy) printf '%s\n' "chatapp3-local-rocketmq-proxy" ;;
    *) printf '%s\n' "chatapp3-local-$1" ;;
  esac
}

health_url_for_service() {
  local service="$1"
  case "$service" in
    gateway)
      printf 'http://127.0.0.1:%s/actuator/health\n' "$(service_port "$service")"
      ;;
    console)
      printf 'http://127.0.0.1:%s/console/actuator/health\n' "$(service_port "$service")"
      ;;
    golang)
      printf 'http://127.0.0.1:%s/health\n' "$(service_port "$service")"
      ;;
    *)
      printf 'http://127.0.0.1:%s/actuator/health\n' "$(service_port "$service")"
      ;;
  esac
}

http_probe() {
  local url="$1"
  local body_file http_code
  body_file="$(mktemp)"
  http_code="$(curl -sS --max-time 5 -o "$body_file" -w '%{http_code}' "$url" || true)"
  printf '%s\n' "$http_code"
  cat "$body_file"
  rm -f "$body_file"
}

print_row() {
  printf '%-16s %-10s %-12s %s\n' "$1" "$2" "$3" "$4"
}

published_host_port() {
  local container_name="$1"
  local target_port="$2"
  docker port "$container_name" "$target_port" 2>/dev/null | sed -n '1s/.*://p'
}

probe_app_service() {
  local service="$1"
  local container_name container_state health_url
  local probe_payload http_code body status detail

  container_name="$(container_name_for_service "$service")"
  container_state="$(docker inspect -f '{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' "$container_name" 2>/dev/null || printf 'missing')"
  health_url="$(health_url_for_service "$service")"

  probe_payload="$(http_probe "$health_url")"
  http_code="$(printf '%s\n' "$probe_payload" | sed -n '1p')"
  body="$(printf '%s\n' "$probe_payload" | sed '1d')"

  status="FAIL"
  detail="container=${container_state} http=${http_code}"

  if [ "$http_code" = "200" ]; then
    if [ "$service" = "golang" ]; then
      status="OK"
      detail="container=${container_state} http=200 path=/health"
    else
      local actuator_status reason
      actuator_status="$(printf '%s' "$body" | jq -r '.status // empty' 2>/dev/null || true)"
      if [ "$actuator_status" = "UP" ]; then
        status="OK"
        detail="container=${container_state} actuator=UP"
      else
        reason="$(printf '%s' "$body" | jq -r '.components.refreshScope.details.error // .components.db.details.error // .error // empty' 2>/dev/null || true)"
        detail="container=${container_state} actuator=${actuator_status:-unknown}"
        if [ -n "$reason" ]; then
          detail="$detail reason=$(printf '%s' "$reason" | tr '\n' ' ' | cut -c1-120)"
        fi
      fi
    fi
  else
    local reason
    reason="$(printf '%s' "$body" | jq -r '.errorMsg // .message // .status // empty' 2>/dev/null || true)"
    if [ -n "$reason" ]; then
      detail="$detail reason=$(printf '%s' "$reason" | tr '\n' ' ' | cut -c1-120)"
    fi
  fi

  if [ "$status" != "OK" ]; then
    FAILURES=$((FAILURES + 1))
  fi

  print_row "$service" "$status" "$container_state" "$detail"
}

probe_infra_service() {
  local service="$1"
  local container_name container_state status detail

  container_name="$(container_name_for_service "$service")"
  container_state="$(docker inspect -f '{{if .State.Health}}{{.State.Health.Status}}{{else}}{{.State.Status}}{{end}}' "$container_name" 2>/dev/null || printf 'missing')"
  status="FAIL"
  detail="container=${container_state}"

  case "$service" in
    nacos)
      local probe_payload http_code body nacos_status
      probe_payload="$(http_probe "http://127.0.0.1:${LIKEI_LOCAL_NACOS_PORT}/nacos/actuator/health")"
      http_code="$(printf '%s\n' "$probe_payload" | sed -n '1p')"
      body="$(printf '%s\n' "$probe_payload" | sed '1d')"
      nacos_status="$(printf '%s' "$body" | jq -r '.status // empty' 2>/dev/null || true)"
      if [ "$http_code" = "200" ] && [ "$nacos_status" = "UP" ]; then
        status="OK"
        detail="container=${container_state} actuator=UP"
      else
        detail="container=${container_state} http=${http_code} actuator=${nacos_status:-unknown}"
      fi
      ;;
    zookeeper)
      local zk_host_port
      zk_host_port="$(published_host_port "$container_name" "2181/tcp")"
      if [ -z "$zk_host_port" ]; then
        zk_host_port="${LIKEI_LOCAL_ZOOKEEPER_EXPOSE_PORT}"
      fi
      if nc -z 127.0.0.1 "$zk_host_port" >/dev/null 2>&1; then
        status="OK"
        detail="container=${container_state} tcp=open port=${zk_host_port}"
      else
        detail="container=${container_state} tcp=not_ready port=${zk_host_port}"
      fi
      ;;
    rocketmq-namesrv)
      if nc -z 127.0.0.1 "${LIKEI_LOCAL_ROCKETMQ_NAMESRV_PORT}" >/dev/null 2>&1; then
        status="OK"
        detail="container=${container_state} tcp=open port=${LIKEI_LOCAL_ROCKETMQ_NAMESRV_PORT}"
      else
        detail="container=${container_state} tcp=not_ready port=${LIKEI_LOCAL_ROCKETMQ_NAMESRV_PORT}"
      fi
      ;;
    rocketmq-broker)
      if docker exec chatapp3-local-rocketmq-broker sh mqadmin clusterList -n rocketmq-namesrv:9876 2>/dev/null | grep -q "DefaultCluster"; then
        status="OK"
        detail="container=${container_state} cluster=DefaultCluster"
      else
        detail="container=${container_state} cluster=not_ready"
      fi
      ;;
    rocketmq-proxy)
      if nc -z 127.0.0.1 "${LIKEI_LOCAL_ROCKETMQ_PROXY_GRPC_PORT}" >/dev/null 2>&1; then
        status="OK"
        detail="container=${container_state} tcp=open port=${LIKEI_LOCAL_ROCKETMQ_PROXY_GRPC_PORT}"
      else
        detail="container=${container_state} tcp=not_ready port=${LIKEI_LOCAL_ROCKETMQ_PROXY_GRPC_PORT}"
      fi
      ;;
    *)
      if [ "$container_state" = "healthy" ] || [ "$container_state" = "running" ]; then
        status="OK"
        detail="container=${container_state}"
      fi
      ;;
  esac

  if [ "$status" != "OK" ]; then
    FAILURES=$((FAILURES + 1))
  fi

  print_row "$service" "$status" "$container_state" "$detail"
}

echo "App Services"
print_row "SERVICE" "RESULT" "CONTAINER" "DETAIL"
for service in "${APP_SERVICES[@]}"; do
  probe_app_service "$service"
done

echo
echo "Infra Services"
print_row "SERVICE" "RESULT" "CONTAINER" "DETAIL"
for service in "${INFRA_SERVICES[@]}"; do
  probe_infra_service "$service"
done

echo
if [ "$FAILURES" -eq 0 ]; then
  echo "SUMMARY: all checked local backend services are healthy"
else
  echo "SUMMARY: $FAILURES service(s) need attention"
  exit 1
fi