落地结构化日志与基础监控告警

scripts/ops/check_basic_metrics.sh

@@ -0,0 +1,67 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+# 用法：
+# METRICS_URL=http://127.0.0.1:8080/metrics/basic \
+# SERVER_ERROR_RATE_THRESHOLD=5 AVG_LATENCY_THRESHOLD_MS=800 \
+# OPS_ALERT_WEBHOOK=https://example.com/webhook ./scripts/ops/check_basic_metrics.sh
+
+METRICS_URL="${METRICS_URL:-http://127.0.0.1:8080/metrics/basic}"
+SERVER_ERROR_RATE_THRESHOLD="${SERVER_ERROR_RATE_THRESHOLD:-5}"
+AVG_LATENCY_THRESHOLD_MS="${AVG_LATENCY_THRESHOLD_MS:-800}"
+OPS_ALERT_WEBHOOK="${OPS_ALERT_WEBHOOK:-}"
+ALERT_TITLE="${ALERT_TITLE:-[wx_service] 基础监控告警}"
+
+send_alert() {
+  local message="$1"
+  if [[ -z "${OPS_ALERT_WEBHOOK}" ]]; then
+    echo "ALERT: ${message}" >&2
+    return
+  fi
+  curl -fsS -X POST "${OPS_ALERT_WEBHOOK}" \
+    -H "Content-Type: application/json" \
+    -d "{\"title\":\"${ALERT_TITLE}\",\"message\":\"${message}\"}" >/dev/null || true
+}
+
+response="$(curl -fsS "${METRICS_URL}" || true)"
+if [[ -z "${response}" ]]; then
+  send_alert "无法访问 metrics 接口：${METRICS_URL}"
+  exit 1
+fi
+
+set +e
+check_output="$(echo "${response}" | python3 - "${SERVER_ERROR_RATE_THRESHOLD}" "${AVG_LATENCY_THRESHOLD_MS}" <<'PY'
+import json
+import sys
+
+payload = json.load(sys.stdin)
+data = payload.get("data", {})
+server_error = float(data.get("server_error_rate_pct", 0.0))
+avg_latency = float(data.get("avg_latency_ms", 0.0))
+
+server_th = float(sys.argv[1])
+latency_th = float(sys.argv[2])
+
+problems = []
+if server_error > server_th:
+    problems.append(f"server_error_rate_pct={server_error:.2f}% > {server_th:.2f}%")
+if avg_latency > latency_th:
+    problems.append(f"avg_latency_ms={avg_latency:.2f} > {latency_th:.2f}")
+
+if problems:
+    print("; ".join(problems))
+    sys.exit(1)
+
+print(f"ok server_error_rate_pct={server_error:.2f}% avg_latency_ms={avg_latency:.2f}")
+PY
+)"
+check_exit=$?
+set -e
+
+if [[ ${check_exit} -ne 0 ]]; then
+  send_alert "基础监控阈值触发：${check_output}"
+  exit 1
+fi
+
+echo "${check_output}"
+