Considering the following CloudWatch alert on an API Gateway:

LatencyAlarmP90:
    Type: AWS::CloudWatch::Alarm
    Condition: Production
    Properties:
      AlarmDescription: latency P90 is lower then 1.5 sec
      AlarmName: FooApiP90LatencyAlarm
      ComparisonOperator: GreaterThanOrEqualToThreshold
      ExtendedStatistic: "p90"
      Period: 300
      MetricName: "Latency"
      DatapointsToAlarm: 5
      EvaluationPeriods: 5
      Threshold: 1500
      Namespace: AWS/ApiGateway
      Dimensions:
        - Name: ApiName
          Value: !Sub "bar"
      AlarmActions:
        - Fn::ImportValue: !Sub "foo"
      OKActions:
        - Fn::ImportValue: !Sub foo"

If this alert triggers, I have no way of knowing which specific route triggered it (multiple lambdas are on the same API Gateway).

I was thinking about monitoring each Lambda instead.

If I monitor a specific Lambda latency does it include API Gateway latency? (Cold starts) or only run time?

If not, is there a way to monitor a specific route/Lambda including API Gateway latency?

CodePudding user response：

If I monitor a specific Lambda latency, does it include API Gateway latency?

No.

The Duration performance metric does not include any API Gateway latency & is only Lambda function run time rounded up to the nearest millisecond as per docs.

If not, is there a way to monitor a specific route/Lambda including API Gateway latency?

Per route yes, per Lambda no.

For a metric that also includes API Gateway overhead, use the Latency metric for API GW.

Per Lambda function is not a supported dimension to assign to the metric and thus filter by but you can achieve the end result by using various other dimensions - including route - to filter this down.

Inevitably, you will need more alarms if the same Lambda function is responsible for multiple routes.

You will have to filter down your alarm by:

• Method
• ApiId & Stage
• Stage
• ApiId
• Api Name