cdk-library-cloudwatch-alarms

WIP - Library to provide constructs, aspects, and construct extensions to more easily set up alarms for AWS resources in CDK code based on AWS recommended alarms list. This project is still in early development so YMMV.

Usage

This library is flexible in its approach and there are multiple paths to configuring alarms depending on how you'd like to work with the recommended alarms.

Feature Availability

Intended feature list as of Aug 2024

• [x] Aspects to apply recommended alarms to a wide scope such as a whole CDK app
  • [x] Ability to exclude specific alarms
  • [x] Ability to define a default set of alarm actions
  • [x] Ability to modify the configuration of each alarm type
  • [ ] Ability to exclude specific resources
• [x] Constructs to ease alarm configuration for individual resources at a granular scope
  • [x] Constructs for each available alarm according to the coverage table
  • [x] Constructs for applying all recommended alarms to a specific resource
  • [x] Ability to exclude specific alarms from the all recommended alarms construct
• [x] Extended versions of resource constructs with alarm helper methods

Resource Coverage

If it's not shown it hasn't been worked on.

| Service | Status | Notes | | --- | --- | --- | | S3 | - [x] 4xxErrors- [x] 5xxErrors- [ ] OperationsFailedReplication | Replication errors are difficult to set up in CDK at the moment due to rule properties being IResolvables and replication rules not being available on the L2 Bucket construct | | SQS | - [x] ApproximateAgeOfOldestMessage- [x] ApproximateNumberOfMessagesNotVisible- [x] ApproximateNumberOfMessagesVisible- [x] NumberOfMessagesSent |- All alarms with the exception of number of messages sent require a user defined threshold because its very use-case specific.- The Aspect only assigns DLQs the ApproximateNumberOfMessagesVisible alarm with a default threshold of 0, unless dlqsGetFullRecommendedAlarms is true, in which case they get the same alarms as other queues. DLQs that belong to a main queue which isn't in the same scope as the Aspect is added to won't be detected as DLQs and they will be treated as normal queues. | | SNS | - [x] NumberOfMessagesPublished- [x] NumberOfNotificationsDelivered- [x] NumberOfNotificationsFailed- [x] NumberOfNotificationsFilteredOut-InvalidAttributes- [x] NumberOfNotificationsFilteredOut-InvalidMessageBody- [x] NumberOfNotificationsRedrivenToDlq- [x] NumberOfNotificationsFailedToRedriveToDlq- [ ] SMSMonthToDateSpentUSD- [ ] SMSSuccessRate | Some alarms require a threshold to be defined. SMS alarms are not implememented. | | Lambda | - [ ] ClaimedAccountConcurrency- [x] Errors- [x] Throttles- [x] Duration- [x] ConcurrentExecutions | ClaimedAccountConcurrency is account wide and one time so not covered by this library at this time | | RDS | For database & cluster instances- [x] CPUUtilization- [x] DatabaseConnections- [x] FreeableMemory- [x] FreeLocalStorage- [x] FreeStorageSpace- [x] ReadLatency- [x] WriteLatency- [x] DBLoadFor clusters- [x] AuroraVolumeBytesLeftTotal- [x] AuroraBinlogReplicaLag | Some alarms require a threshold to be defined. AuroraVolumeBytesLeftTotal and AuroraBinlogReplicaLag alarms are created only for Aurora MySQL clusters. | | ECS | - [x] CPUUtilization- [x] MemoryUtilization- [x] EphemeralStorageUtilized- [x] RunningTaskCount | The alarms are applied to FargateService constructs only. EphemeralStorageUtilized requires a threshold to be defined. | | EFS | - [x] PercentIOLimit- [x] BurstCreditBalance | The alarms are applied to FileSystem constructs. | | ApiGateway | - [x] 4XXError- [x] 5XXError- [x] Latency | The alarms are applied to RestApi constructs only. Alarms are automatically created using the ApiName and Stage dimensions. To create Latency alarms using the Resource and Method dimensions, the corresponding properties must be explicitly specified. | | CloudFront | - [x] 5xxErrorRate- [x] OriginLatency- [x] FunctionValidationErrors- [x] FunctionExecutionErrors- [x] FunctionThrottles | The alarms are applied to Distribution constructs only. Both 5xxErrorRate and OriginLatency require a threshold to be defined. To create Function level alarms using the FunctionName dimension, the corresponding properties must be explicitly specified. | | DynamoDB | Mandatory alarms- [x] ReadThrottleEvents- [x] SystemErrors- [x] WriteThrottleEventsReplication alarms (optional)- [x] AgeOfOldestUnreplicatedRecord- [x] FailedToReplicateRecordCount- [x] ThrottledPutRecordCount | The alarms are applied to Table constructs only. All the mandatory alarms require a threshold to be defined.Replication alarms are created only if the corresponding configuration is specified. Each replication alarm has a default DelegatedOperation dimension value:- AgeOfOldestUnreplicatedRecord: StreamRecords- FailedToReplicateRecordCount: StreamRecords- ThrottledPutRecordCount: PutItem | | EC2 | - [x] CPUUtilization- [x] StatusCheckFailed | The alarms are applied to Instance constructs. | | AutoScaling | - [x] GroupInServiceCapacity | The alarms are applied to AutoScalingGroup constructs. The alarm requires a threshold to be defined and the AutoScalingGroup should have this metric explicitly enabled. | | ElastiCache | - [x] DatabaseMemoryUsagePercentage- [x] EngineCPUUtilization- [x] ReplicationLag | The alarms are applied to CfnCacheCluster and CfnReplicationGroup constructs. DatabaseMemoryUsagePercentage and ReplicationLag require a threshold to be defined.| | PrivateLink | Endpoints- [x] PacketsDroppedEndpoint Services- [x] RstPacketsSent | The alarms are applied to InterfaceVpcEndpoint and VpcEndpointService constructs. Because these objects do not expose the attributes required by alarms, they cannot be implemented using the Aspect. In all cases, the threshold must be defined. | | VPN | - [x] TunnelState | The alarms are applied to CfnVPNConnection constructs. | | ELBv2 | For ApplicationLoadBalancer- [x] RejectedConnectionCount- [x] HTTPCode_ELB_4XX_Count- [x] HTTPCode_ELB_5XX_Count- [x] HTTPCode_Target_5XX_CountFor ApplicationTargetGroup- [x] HealthyHostCount- [x] UnHealthyHostCountFor NetworkLoadBalancer- [x] TCP_ELB_Reset_Count- [x] TCP_Target_Reset_CountFor NetworkTargetGroup- [x] HealthyHostCount- [x] UnHealthyHostCount |- For target groups, HealthyHostCount alarm triggers when count falls below threshold (default: 1) and UnHealthyHostCount alarm triggers when count exceeds threshold (default: 0). For load balancers, all alarms trigger when count exceeds threshold (default: 0).- The HTTPCode_ELB_4XX_Count and HTTPCode_ELB_5XX_Count alarms are defined as anomaly detection alarms instead of flat counts, because there is normally a constant background of such errors. | | DMS | For ReplicationInstances- [x] CPUUtilization- [x] FreeableMemory- [x] FreeStorageSpace- [x] WriteIOPS- [x] SwapUsageFor Replication Tasks- [x] CDCLatencySource- [x] CDCLatencyTarget | The alarms are applied to CfnReplicationInstance and CfnReplicationTask constructs. Extended versions ReplicationInstance and ReplicationTask are also available with alarm helper methods.Replication Instance Notes:- FreeableMemory and FreeStorageSpace alarms require a threshold to be defined.- SwapUsage alarm uses anomaly detection with a fixed 5-minute period (as required by AWS CloudWatch) to detect high swap usage that may indicate memory pressure or performance issues. By default, it triggers when swap usage exceeds the upper threshold of the anomaly detection band, but the comparisonOperator can be configured to detect other patterns.Replication Task Notes:- Alarms are created and validated based on the migrationType of the ReplicationTask. CDC-related alarms can only be created for cdc or full-load-and-cdc migration types.- CDC latency alarms (CDCLatencySource and CDCLatencyTarget) default to detecting high latency issues (threshold: 300 seconds, comparison: GREATER_THAN_THRESHOLD) which can indicate replication lag or database performance problems.- The CDCLatencySource alarm monitors the gap between the last event captured from the source endpoint and current system time, while CDCLatencyTarget monitors the gap between a change committed to the source and the same change committed to the target. |

Aspects

Below is an example of configuring the Lambda aspect. You must configure non-defaults for alarms which is most cases is only a threshold. Since the aspect is applied at the app level it applies to both the TestStack and TestStack2 lambda functions and will create all available recommended alarms for those functions. See references for additional details on Aspects which can be applied to the app, stack, or individual constructs depending on your use case.

import { App, Stack, Aspects, aws_lambda as lambda } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const stack2 = new Stack(app, 'TestStack2', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const appAspects = Aspects.of(app);

appAspects.add(
  new recommendedalarms.LambdaRecommendedAlarmsAspect({
    configDurationAlarm: {
      threshold: 15,
    },
    configErrorsAlarm: {
      threshold: 1,
    },
    configThrottlesAlarm: {
      threshold: 0,
    },
  }),
);

new lambda.Function(stack, 'Lambda', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

new lambda.Function(stack2, 'Lambda2', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

Recommended Alarm Constructs

You can also apply alarms to a specific resource using the recommended alarm construct for a given resource type. For example if you have an S3 Bucket you might do something like below. None of the S3 alarms require configuration so no config props are needed in this case:

import { App, Stack, Aspects, aws_s3 as s3 } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const bucket = new s3.Bucket(stack, 'Bucket', {});

new recommendedalarms.S3RecommendedAlarms(stack, 'RecommendedAlarms', {
  bucket,
});

Individual Constructs

You can also apply specific alarms from their individual constructs:

import { App, Stack, Aspects, aws_s3 as s3 } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const bucket = new s3.Bucket(stack, 'Bucket', {});

new recommendedalarms.S3Bucket5xxErrorsAlarm(stack, 'RecommendedAlarms', {
  bucket,
  threshold: 0.10,
});

Construct Extensions

You can use extended versions of the constructs you are familiar with to expose helper methods for alarms if you'd like to keep alarms more tightly coupled to specific resources.

import { App, Stack, Aspects, aws_s3 as s3 } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

  const bucket = new recommendedalarms.Bucket(stack, 'Bucket', {});

  bucket.applyRecommendedAlarms();

Alarm Actions

You can apply alarm actions using the default actions on an aspect or all recommended alarms construct or you can apply individual alarm actions for helper methods of individual constructs. See below where default actions are set but an override is set for a specific alarm for the alarm action to use a different SNS topic.

import { App, Stack, Aspects, aws_lambda as lambda } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const stack2 = new Stack(app, 'TestStack2', {
  env: {
    account: '123456789012',
    region: 'us-east-1',
  },
});

const alarmTopic = new sns.Topic(stack, 'Topic');
const topicAction =  new cloudwatch_actions.SnsAction(alarmTopic)

const alarmTopic2 = new sns.Topic(stack, 'Topic');
const topicAction2 =  new cloudwatch_actions.SnsAction(alarmTopic2)

const appAspects = Aspects.of(app);

appAspects.add(
  new recommendedalarms.LambdaRecommendedAlarmsAspect({
    defaultAlarmAction: topicAction,
    defaultOkAction: topicAction,
    defaultInsufficientDataAction: topicAction,
    configDurationAlarm: {
      threshold: 15,
      alarmAction: topicAction2,
    },
    configErrorsAlarm: {
      threshold: 1,
    },
    configThrottlesAlarm: {
      threshold: 0,
    },

  }),
);

new lambda.Function(stack, 'Lambda', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

new lambda.Function(stack2, 'Lambda2', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

Exclusions

You can exclude specific alarms or specific resources. Alarms use the available metrics enums and resources use the string used for a resources id. For example below Lambda1 will not have alarms created and there will be no alarm for the Duration metric for either lambda function.

import { App, Stack, Aspects, aws_lambda as lambda } from 'aws-cdk-lib';
import * as recommendedalarms from '@renovosolutions/cdk-library-cloudwatch-alarms';

const app = new App();
const stack = new Stack(app, 'TestStack', {
  env: {
    account: '123456789012', // not a real account
    region: 'us-east-1',
  },
});

const appAspects = Aspects.of(app);

appAspects.add(
  new recommendedalarms.LambdaRecommendedAlarmsAspect({
    excludeResources: ['Lambda1'],
    excludeAlarms: [recommendedalarms.LambdaRecommendedAlarmsMetrics.DURATION],
    configDurationAlarm: {
      threshold: 15,
    },
    configErrorsAlarm: {
      threshold: 1,
    },
    configThrottlesAlarm: {
      threshold: 0,
    },
  }),
);

new lambda.Function(stack, 'Lambda1', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

new lambda.Function(stack, 'Lambda2', {
  runtime: lambda.Runtime.NODEJS_20_X,
  handler: 'index.handler',
  code: lambda.Code.fromInline('exports.handler = async (event) => { console.log(event); }'),
});

References

• AWS Recommended Alarms
• Aspects and the AWS CDK