如何为Lambda调用Dynamo场景在CloudFormation CloudWatch告警中新增latency监控

完全可以在你现有CloudFormation样例的基础上新增latency告警，根据你的监控精度需求，可以选择以下两种方案：

方案1：监控Lambda整体运行延迟（覆盖调用Dynamo的全链路耗时） #

直接复用Lambda自带的Duration指标即可，无需额外埋点，只需要在现有模板的Resources块下新增一个告警资源，原有错误告警保持不动即可，完整示例如下：

# 原有Lambda错误告警保留
MyAlarm:
  Type: AWS::CloudWatch::Alarm
  Properties:
    AlarmName: "superalarm"
    AlarmDescription: "Alarm if lambda errors out "
    Namespace: "AWS/Lambda"
    MetricName: "Errors"
    Dimensions:
    - Name: "FunctionName"
      Value: "mylambda"
    Statistic: "Sum"
    ComparisonOperator: "GreaterThanThreshold"
    Threshold: 0
    EvaluationPeriods: 5
    Period: 60
    TreatMissingData: "breaching"

# 新增的Lambda延迟告警
LambdaLatencyAlarm:
  Type: AWS::CloudWatch::Alarm
  Properties:
    AlarmName: "lambda_dynamo_latency_alarm"
    AlarmDescription: "Alarm when lambda execution latency exceeds threshold"
    Namespace: "AWS/Lambda"
    MetricName: "Duration"
    Dimensions:
    - Name: "FunctionName"
      Value: "mylambda"
    Statistic: "Average" # 可根据需求替换为P95、P99等分位统计值
    ComparisonOperator: "GreaterThanThreshold"
    Threshold: 1000 # 阈值单位为毫秒，此处为1秒，可根据业务容忍度调整
    EvaluationPeriods: 3
    Period: 60
    TreatMissingData: "notBreaching"

方案2：精准监控Lambda调用Dynamo的单独请求延迟 #

如果需要排除Lambda其他逻辑的耗时，只统计Dynamo调用的延迟，可以使用DynamoDB自带的SuccessfulRequestLatency指标配置告警，示例如下：

DynamoCallLatencyAlarm:
  Type: AWS::CloudWatch::Alarm
  Properties:
    AlarmName: "dynamo_call_latency_alarm"
    AlarmDescription: "Alarm when DynamoDB request latency exceeds threshold"
    Namespace: "AWS/DynamoDB"
    MetricName: "SuccessfulRequestLatency"
    Dimensions:
    - Name: "TableName"
      Value: "你的Dynamo表名" # 替换为实际调用的Dynamo表名
    - Name: "Operation"
      Value: "Query" # 替换为实际调用的操作类型，如PutItem、GetItem、Scan等
    Statistic: "Average"
    ComparisonOperator: "GreaterThanThreshold"
    Threshold: 200 # 阈值单位为毫秒，可自行调整
    EvaluationPeriods: 3
    Period: 60
    TreatMissingData: "notBreaching"

注意事项 #

• 阈值、统计周期、统计方式都可以根据业务实际需求调整，对延迟敏感的场景推荐使用P95/P99分位值作为统计指标
• 如果Lambda会调用多个Dynamo表或者多种操作类型，需要为每个表+操作的组合单独配置告警
• 如果需要更精准的只统计该特定Lambda发起的Dynamo调用延迟，可以在Lambda代码中埋点上报自定义监控指标，再针对自定义指标配置告警即可

内容的提问来源于stack exchange，提问作者ozil