init prometheus监控配置文件

2026-07-09 10:12:19 +08:00
commit 5e9088793c
55 changed files with 205440 additions and 0 deletions
@@ -0,0 +1,57 @@
+groups:
+- name: single_port_service_alarm
+  rules:
+  ########################### 严重告警 Critical ###########################
+  # 22 sshd 远程登录端口
+  - alert: Port_22_SSHD_Down
+    expr: probe_success{port="22"} == 0
+    for: 10s
+    labels:
+      severity: critical
+      service: sshd
+    annotations:
+      summary: "主机{{$labels.instance}} SSH登录端口{{$labels.port}}断开"
+      description: "端口{{$labels.port}} sshd服务异常，无法远程登录服务器，紧急处理！"
+
+  # 111 rpcbind RPC注册服务
+  - alert: Port_111_Rpcbind_Down
+    expr: probe_success{port="111"} == 0
+    for: 10s
+    labels:
+      severity: critical
+      service: rpcbind
+    annotations:
+      summary: "主机{{$labels.instance}} RPC注册端口{{$labels.port}}断开"
+      description: "端口{{$labels.port}} rpcbind异常，所有NFS/RPC相关服务全部失效，业务挂载会报错！"
+
+  # 2049 NFS存储端口
+  - alert: Port_2049_NFS_Down
+    expr: probe_success{port="2049"} == 0
+    for: 10s
+    labels:
+      severity: critical
+      service: nfs
+    annotations:
+      summary: "主机{{$labels.instance}} NFS存储端口{{$labels.port}}断开"
+      description: "端口{{$labels.port}} NFS内核端口无法连通，所有业务NFS读写中断！"
+
+  # 9100 node_exporter监控采集
+  - alert: Port_9100_NodeExporter_Down
+    expr: probe_success{port="9100"} == 0
+    for: 10s
+    labels:
+      severity: critical
+      service: node_exporter
+    annotations:
+      summary: "主机{{$labels.instance}} 监控采集端口{{$labels.port}}断开"
+      description: "端口{{$labels.port}} node_exporter离线，主机资源指标无法采集，无监控数据！"
+
+  ########################### 通用端口延迟告警 ###########################
+  - alert: Port_High_Latency
+    expr: probe_duration_seconds > 0.5
+    for: 1m
+    labels:
+      severity: warning
+    annotations:
+      summary: "主机{{$labels.instance}} {{$labels.service}}端口{{$labels.port}}连接延迟过高"
+      description: "端口{{$labels.port}}服务{{$labels.service}}网络耗时超过500ms，网络存在拥塞"