feat(dynamo): add ability to alarm on min TTL deletions

cdklabs · May 30, 2024 · 2444cf6 · 2444cf6
1 parent d1fda6f
commit 2444cf6
Show file tree

Hide file tree

Showing 5 changed files with 213 additions and 24 deletions.
diff --git a/API.md b/API.md
diff --git a/lib/monitoring/aws-dynamo/DynamoTableMetricFactory.ts b/lib/monitoring/aws-dynamo/DynamoTableMetricFactory.ts
@@ -4,13 +4,10 @@ import { BillingMode, ITable, Operation } from "aws-cdk-lib/aws-dynamodb";
 import { MetricFactory, MetricStatistic } from "../../common";
 
 const DynamoDbNamespace = "AWS/DynamoDB";
-const ProvisionedReadCapacityUnitsMetric = "ProvisionedReadCapacityUnits";
-const ProvisionedWriteCapacityUnitsMetric = "ProvisionedWriteCapacityUnits";
 const ProvisionedLabel = "Provisioned";
 const ConsumedLabel = "Consumed";
 const ReadThrottleEventsLabel = "Read";
 const WriteThrottleEventsLabel = "Write";
-const SystemErrorsLabel = "System Errors";
 
 export interface DynamoTableMetricFactoryProps {
   /**
@@ -39,7 +36,7 @@ export class DynamoTableMetricFactory {
 
   metricProvisionedReadCapacityUnits() {
     return this.metricFactory.adaptMetric(
-      this.table.metric(ProvisionedReadCapacityUnitsMetric, {
+      this.table.metric("ProvisionedReadCapacityUnits", {
         label: ProvisionedLabel,
         statistic: MetricStatistic.AVERAGE,
       })
@@ -48,7 +45,7 @@ export class DynamoTableMetricFactory {
 
   metricProvisionedWriteCapacityUnits() {
     return this.metricFactory.adaptMetric(
-      this.table.metric(ProvisionedWriteCapacityUnitsMetric, {
+      this.table.metric("ProvisionedWriteCapacityUnits", {
         label: ProvisionedLabel,
         statistic: MetricStatistic.AVERAGE,
       })
@@ -195,7 +192,16 @@ export class DynamoTableMetricFactory {
       // the metric is not emitted until error happens
       Object.keys(usingMetrics).join("+"),
       usingMetrics,
-      SystemErrorsLabel
+      "System Errors"
+    );
+  }
+
+  metricTimeToLiveDeletedItemCount() {
+    return this.metricFactory.adaptMetric(
+      this.table.metric("TimeToLiveDeletedItemCount", {
+        label: "TTL Deleted Item Count",
+        statistic: MetricStatistic.MAX,
+      })
     );
   }
 }
diff --git a/lib/monitoring/aws-dynamo/DynamoTableMonitoring.ts b/lib/monitoring/aws-dynamo/DynamoTableMonitoring.ts
@@ -5,6 +5,7 @@ import {
   IMetric,
   IWidget,
   LegendPosition,
+  TreatMissingData,
 } from "aws-cdk-lib/aws-cloudwatch";
 import {
   BillingMode,
@@ -36,12 +37,14 @@ import {
   LatencyThreshold,
   LatencyType,
   MetricWithAlarmSupport,
+  MinUsageCountThreshold,
   Monitoring,
   MonitoringScope,
   PercentageAxisFromZeroToHundred,
   QuarterWidth,
   ThrottledEventsThreshold,
   TimeAxisMillisFromZero,
+  UsageAlarmFactory,
 } from "../../common";
 import {
   MonitoringHeaderWidget,
@@ -68,6 +71,10 @@ export interface DynamoTableMonitoringOptions extends BaseMonitoringProps {
   >;
 
   readonly addSystemErrorCountAlarm?: Record<string, ErrorCountThreshold>;
+  readonly addMinTimeToLiveDeletedItemCountAlarm?: Record<
+    string,
+    MinUsageCountThreshold
+  >;
 
   readonly addAverageSuccessfulGetRecordsLatencyAlarm?: Record<
     string,
@@ -120,6 +127,7 @@ export class DynamoTableMonitoring extends Monitoring {
   readonly errorAlarmFactory: ErrorAlarmFactory;
   readonly latencyAlarmFactory: LatencyAlarmFactory;
   readonly dynamoCapacityAlarmFactory: DynamoAlarmFactory;
+  readonly usageAlarmFactory: UsageAlarmFactory;
 
   readonly latencyAnnotations: HorizontalAnnotation[];
   readonly errorCountAnnotations: HorizontalAnnotation[];
@@ -134,6 +142,7 @@ export class DynamoTableMonitoring extends Monitoring {
   readonly readThrottleCountMetric: MetricWithAlarmSupport;
   readonly writeThrottleCountMetric: MetricWithAlarmSupport;
   readonly systemErrorMetric: MetricWithAlarmSupport;
+  readonly timeToLiveDeletedItemCountMetric: MetricWithAlarmSupport;
   readonly latencyAverageSearchMetrics: IMetric;
   // keys are Operation, but JSII doesn't like non-string types
   readonly averagePerOperationLatencyMetrics: Record<
@@ -166,6 +175,8 @@ export class DynamoTableMonitoring extends Monitoring {
     this.errorAlarmFactory = new ErrorAlarmFactory(this.alarmFactory);
     this.latencyAlarmFactory = new LatencyAlarmFactory(this.alarmFactory);
     this.dynamoCapacityAlarmFactory = new DynamoAlarmFactory(this.alarmFactory);
+    this.usageAlarmFactory = new UsageAlarmFactory(this.alarmFactory);
+
     this.errorCountAnnotations = [];
     this.latencyAnnotations = [];
     this.dynamoReadCapacityAnnotations = [];
@@ -189,6 +200,8 @@ export class DynamoTableMonitoring extends Monitoring {
     this.writeThrottleCountMetric =
       metricFactory.metricThrottledWriteRequestCount();
     this.systemErrorMetric = metricFactory.metricSystemErrorsCount();
+    this.timeToLiveDeletedItemCountMetric =
+      metricFactory.metricTimeToLiveDeletedItemCount();
     this.latencyAverageSearchMetrics =
       metricFactory.metricSearchAverageSuccessfulRequestLatencyInMillis();
     this.averagePerOperationLatencyMetrics = {
@@ -293,6 +306,19 @@ export class DynamoTableMonitoring extends Monitoring {
       this.errorCountAnnotations.push(createdAlarm.annotation);
       this.addAlarm(createdAlarm);
     }
+    for (const disambiguator in props.addMinTimeToLiveDeletedItemCountAlarm) {
+      const alarmProps = {
+        // Missing data is bad if we expect TTL to be enabled
+        treatMissingDataOverride: TreatMissingData.BREACHING,
+        ...props.addMinTimeToLiveDeletedItemCountAlarm[disambiguator],
+      };
+      const createdAlarm = this.usageAlarmFactory.addMinCountAlarm(
+        this.timeToLiveDeletedItemCountMetric,
+        alarmProps,
+        disambiguator
+      );
+      this.addAlarm(createdAlarm);
+    }
     this.forEachOperationLatencyAlarmDefinition(
       Operation.GET_RECORDS,
       props.addAverageSuccessfulGetRecordsLatencyAlarm

diff --git a/test/monitoring/aws-dynamo/DynamoTableMonitoring.test.ts b/test/monitoring/aws-dynamo/DynamoTableMonitoring.test.ts
@@ -1,5 +1,6 @@
 import { Duration, Stack } from "aws-cdk-lib";
 import { Template } from "aws-cdk-lib/assertions";
+import { TreatMissingData } from "aws-cdk-lib/aws-cloudwatch";
 import { AttributeType, BillingMode, Table } from "aws-cdk-lib/aws-dynamodb";
 
 import { AlarmWithAnnotation, DynamoTableMonitoring } from "../../../lib";
@@ -62,6 +63,11 @@ test("snapshot test: all alarms", () => {
         evaluationPeriods: 8,
       },
     },
+    addMinTimeToLiveDeletedItemCountAlarm: {
+      Warning: {
+        minCount: 5,
+      },
+    },
     addReadThrottledEventsCountAlarm: {
       Warning: {
         maxThrottledEventsThreshold: 5,
@@ -125,7 +131,7 @@ test("snapshot test: all alarms", () => {
   });
 
   addMonitoringDashboardsToStack(stack, monitoring);
-  expect(numAlarmsCreated).toStrictEqual(14);
+  expect(numAlarmsCreated).toStrictEqual(15);
   expect(Template.fromStack(stack)).toMatchSnapshot();
 });
 
@@ -187,6 +193,12 @@ test("snapshot test: pay-per-request, all alarms", () => {
         evaluationPeriods: 8,
       },
     },
+    addMinTimeToLiveDeletedItemCountAlarm: {
+      Warning: {
+        minCount: 5,
+        treatMissingDataOverride: TreatMissingData.MISSING,
+      },
+    },
     addReadThrottledEventsCountAlarm: {
       Warning: {
         maxThrottledEventsThreshold: 5,
@@ -250,6 +262,6 @@ test("snapshot test: pay-per-request, all alarms", () => {
   });
 
   addMonitoringDashboardsToStack(stack, monitoring);
-  expect(numAlarmsCreated).toStrictEqual(14);
+  expect(numAlarmsCreated).toStrictEqual(15);
   expect(Template.fromStack(stack)).toMatchSnapshot();
 });