监控指标

{
  "version": "1",
  "statement": [
    {
      "effect": "Allow",
      "action": [
        "cms:DescribeCustomNamespace",
        "cms:PutHybridMonitorMetricData"
      ],
      "resource": "*"
    }
  ]
}

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "Statement1",
            "Effect": "Allow",
            "Action": [
                "cloudwatch:ListMetrics",
                "cloudwatch:GetMetricData",
                "cloudwatch:GetMetricStatistics",
                "cloudwatch:PutMetricData",
                "cloudwatch:PutMetricAlarm",
                "cloudwatch:TagResource",
                "cloudwatch:UntagResource"
            ],
            "Resource": [
                "*"
            ]
        }
    ]
}

# CPU利用率告警
- name: "Warebase CPU Usage High"
  metric: "warebase_resource_percent_normalized"
  threshold: 80
  comparison: ">"
  duration: "5m"
  severity: "warning"
  description: "Warebase CPU使用率超过80%，建议检查负载情况"
 
# 内存利用率告警
- name: "Warebase Memory Usage High"
  metric: "warebase_resource_percent_normalized"
  threshold: 85
  comparison: ">"
  duration: "5m"
  severity: "warning"
  description: "Warebase内存使用率超过85%，建议检查内存使用情况"
 
# QPS异常告警
- name: "Warebase QPS Spike"
  metric: "warebase_select_qps"
  threshold: 10000
  comparison: ">"
  duration: "1m"
  severity: "critical"
  description: "Warebase查询QPS突增，可能存在异常查询"
 
# 延迟异常告警
- name: "SQL Latency High"
  metric: "warebase_sql_service_p99_latency"
  threshold: 5000000000  # 5秒
  comparison: ">"
  duration: "5m"
  severity: "critical"
  description: "SQL P99延迟超过5秒，可能影响用户体验"
 
# 连接数异常告警
- name: "Active Connections High"
  metric: "warebase_active_sql_connections"
  threshold: 1000
  comparison: ">"
  duration: "5m"
  severity: "warning"
  description: "活跃连接数过多，可能影响系统性能"

# 存储空间使用率告警
- name: "Database Storage Usage High"
  metric: "database_size_bytes"
  threshold: 80  # 百分比
  comparison: ">"
  duration: "1h"
  severity: "warning"
  description: "数据库存储使用率超过80%，建议检查数据增长趋势"
 
# 存储空间增长异常告警
- name: "Database Storage Growth Anomaly"
  metric: "database_size_bytes"
  threshold: 50  # GB/天
  comparison: ">"
  duration: "1d"
  severity: "critical"
  description: "数据库存储日增长超过50GB，可能存在数据异常增长"

{
  "notification_channels": [
    {
      "type": "email",
      "recipients": [
        "admin@company.com",
        "dba@company.com"
      ],
      "subject_template": "[ProtonBase Alert] {{alert_name}} - {{severity}}",
      "body_template": "告警名称: {{alert_name}}\n告警级别: {{severity}}\n告警时间: {{timestamp}}\n告警详情: {{description}}\n当前值: {{current_value}}\n阈值: {{threshold}}"
    }
  ]
}

{
  "notification_channels": [
    {
      "type": "webhook",
      "url": "https://your-monitoring-system.com/webhook",
      "method": "POST",
      "headers": {
        "Content-Type": "application/json",
        "Authorization": "Bearer YOUR_WEBHOOK_TOKEN"
      },
      "payload_template": {
        "alert_name": "{{alert_name}}",
        "severity": "{{severity}}",
        "timestamp": "{{timestamp}}",
        "description": "{{description}}",
        "current_value": "{{current_value}}",
        "threshold": "{{threshold}}"
      }
    }
  ]
}