New: [AEA-4488] - Adds NPPTS Slack Alerts (#1339)

## Summary - ✨ New Feature ### Details Adds NPPTS Slack Alerts
NHSDigital · Oct 4, 2024 · 9bfb428 · 9bfb428
1 parent 08ab654
commit 9bfb428
Show file tree

Hide file tree

Showing 2 changed files with 85 additions and 7 deletions.
diff --git a/SAMtemplates/alarms/main.yaml b/SAMtemplates/alarms/main.yaml
@@ -6,11 +6,12 @@ Description: |
 Parameters:
   StackName:
     Type: String
-    Default: none
 
   GetMyPrescriptionsFunctionName:
     Type: String
-    Default: none
+
+  EnrichPrescriptionsFunctionName:
+    Type: String
 
   EnableAlerts:
     Type: String
@@ -23,14 +24,14 @@ Resources:
   ServiceSearchErrorsLogsMetricFilter:
     Type: AWS::Logs::MetricFilter
     Properties:
-      FilterName: GetMyPrescriptionsErrors
-      # Match logs with $.message starting with "call to service search unsuccessful"
-      FilterPattern: !Sub '{ ($.level = "WARN") && ($.function_name = "${GetMyPrescriptionsFunctionName}") && $.message = %call to service search unsuccessful% }' # function_name included to allow it to be set as a dimension on the metric
+      FilterName: ServiceSearchErrors
+      # Match logs with $.message starting with "error in request to serviceSearch"
+      FilterPattern: !Sub '{ ($.level = "ERROR") && ($.function_name = "${GetMyPrescriptionsFunctionName}") && $.message = %error in request to serviceSearch% }' # function_name included to allow it to be set as a dimension on the metric
       LogGroupName:
         Fn::ImportValue: !Sub ${StackName}:functions:${GetMyPrescriptionsFunctionName}:LambdaLogGroupName
       MetricTransformations:
         - MetricNamespace: LambdaLogFilterMetrics
-          MetricName: ErrorCount
+          MetricName: ServiceSearchErrorCount
           MetricValue: 1
           Unit: Count
           Dimensions: # dimensions for a logs filter metric can only be a field/value from the filter pattern
@@ -43,7 +44,7 @@ Resources:
       AlarmDescription: Count of Service Search errors
       AlarmName: !Sub ${StackName}_ServiceSearch_Errors
       Namespace: LambdaLogFilterMetrics
-      MetricName: ErrorCount
+      MetricName: ServiceSearchErrorCount
       Dimensions:
         - Name: FunctionName
           Value: !Ref GetMyPrescriptionsFunctionName
@@ -107,3 +108,79 @@ Resources:
         - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
       OKActions:
         - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+
+  GetMyPrescriptionsErrorsLogsMetricFilter:
+    Type: AWS::Logs::MetricFilter
+    Properties:
+      FilterName: GetMyPrescriptionsErrors
+      FilterPattern: !Sub '{ ($.level = "ERROR") && ($.function_name = "${GetMyPrescriptionsFunctionName}") && ($.message != %error in request to serviceSearch%) }'
+      LogGroupName:
+        Fn::ImportValue: !Sub ${StackName}:functions:${GetMyPrescriptionsFunctionName}:LambdaLogGroupName
+      MetricTransformations:
+        - MetricNamespace: LambdaLogFilterMetrics
+          MetricName: ErrorCount
+          MetricValue: 1
+          Unit: Count
+          Dimensions:
+            - Key: FunctionName
+              Value: $.function_name
+
+  GetMyPrescriptionsErrorsAlarm:
+    Type: AWS::CloudWatch::Alarm
+    Properties:
+      AlarmDescription: Count of GetMyPrescriptions errors
+      AlarmName: !Sub ${StackName}_GetMyPrescriptions_Errors
+      Namespace: LambdaLogFilterMetrics
+      MetricName: ErrorCount
+      Dimensions:
+        - Name: FunctionName
+          Value: !Ref GetMyPrescriptionsFunctionName
+      Period: 60 #seconds
+      EvaluationPeriods: 1
+      Statistic: Sum
+      ComparisonOperator: GreaterThanOrEqualToThreshold
+      Threshold: 1
+      Unit: Count
+      TreatMissingData: notBreaching
+      ActionsEnabled: !Ref EnableAlerts
+      AlarmActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+      InsufficientDataActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+      OKActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+
+  EnrichPrescriptionsErrorsLogsMetricFilter:
+    Type: AWS::Logs::MetricFilter
+    Properties:
+      FilterName: EnrichPrescriptionsErrors
+      FilterPattern: ERROR
+      LogGroupName:
+        Fn::ImportValue: !Sub ${StackName}:functions:${EnrichPrescriptionsFunctionName}:LambdaLogGroupName
+      MetricTransformations:
+        - MetricNamespace: LambdaLogFilterMetrics
+          MetricName: !Sub ${StackName}EnrichPrescriptionsErrorCount
+          MetricValue: 1
+          Unit: Count
+
+  EnrichPrescriptionsErrorsAlarm:
+    Type: AWS::CloudWatch::Alarm
+    Properties:
+      AlarmDescription: Count of EnrichPrescriptions errors
+      AlarmName: !Sub ${StackName}_EnrichPrescriptions_Errors
+      Namespace: LambdaLogFilterMetrics
+      MetricName: !Sub ${StackName}EnrichPrescriptionsErrorCount
+      Period: 60 #seconds
+      EvaluationPeriods: 1
+      Statistic: Sum
+      ComparisonOperator: GreaterThanOrEqualToThreshold
+      Threshold: 1
+      Unit: Count
+      TreatMissingData: notBreaching
+      ActionsEnabled: !Ref EnableAlerts
+      AlarmActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+      InsufficientDataActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
+      OKActions:
+        - !ImportValue lambda-resources:SlackAlertsSnsTopicArn
diff --git a/SAMtemplates/main_template.yaml b/SAMtemplates/main_template.yaml
@@ -136,5 +136,6 @@ Resources:
       Parameters:
         StackName: !Ref AWS::StackName
         GetMyPrescriptionsFunctionName: !GetAtt Functions.Outputs.GetMyPrescriptionsFunctionName
+        EnrichPrescriptionsFunctionName: !GetAtt Functions.Outputs.EnrichPrescriptionsFunctionName
         EnableAlerts: !Ref EnableAlerts
         ConcurrencyThreshold: 10000 #FIXME: Set appropriate threshold based on load tests.