feat: Monitoring & Alarming

.env.example

@@ -19,4 +19,5 @@ VALHEIM_DOCKER_STEAMCMD_ARGS="validate"
 
 # Config values specific to this package
 APPGW_START_STOP_PASSWORD="changeme" # You should really change this if you want to use this feature
-BOOTSTRAP_WITH_WORLD_NAME=""
+BOOTSTRAP_WITH_WORLD_NAME=""
+DISCORD_WEBHOOK_URL_FOR_ALARMS=""

bin/valheim-server-aws-cdk.ts

@@ -3,11 +3,11 @@ import "source-map-support/register";
 import { ValheimServerAwsCdkStack } from "../lib/valheim-server-aws-cdk-stack";
 import { LambdaEcsFargateUpdownstatusStack } from '../lib/lambda-ecs-fargate-updownstatus-stack';
 import { Construct } from "constructs";
-import { Annotations, App, Fn } from "aws-cdk-lib";
+import { Annotations, App } from "aws-cdk-lib";
 import { config } from "dotenv";
-import { readdir } from "fs/promises";
 import WorldBootstrapResourcesStack from "../lib/world-bootstrap-resources-stack";
 import { readdirSync } from "fs";
+import { MonitoringStack } from "../lib/monitoring";
 config();
 
 const BOOTSTRAP_WITH_WORLD_NAME = process.env.BOOTSTRAP_WITH_WORLD_NAME;
@@ -48,6 +48,11 @@ class ValheimServer extends Construct {
             });
             lambdaStack.addDependency(ecsStack);
         }
+
+        new MonitoringStack(this, "ValheimServerMonitoringStack", {
+            fargateService: ecsStack.valheimService,
+            discordWebhookUrl: DISCORD_WEBHOOK_URL_FOR_ALARMS,
+        })
     }
 }
 
@@ -64,6 +69,7 @@ const getWorldAssetLocations = () => {
 const app = new App();
 const worldAssetLocations = getWorldAssetLocations();
 const APPGW_START_STOP_PASSWORD = process.env.APPGW_START_STOP_PASSWORD;
+const DISCORD_WEBHOOK_URL_FOR_ALARMS = process.env.DISCORD_WEBHOOK_URL_FOR_ALARMS;
 
 new ValheimServer(app, "ValheimServer", {
     worldAssetLocations,

lib/monitoring.ts

@@ -0,0 +1,90 @@
+import { Duration, Stack, StackProps } from "aws-cdk-lib";
+import { TreatMissingData } from "aws-cdk-lib/aws-cloudwatch";
+import { FargateService } from "aws-cdk-lib/aws-ecs";
+import { Runtime } from "aws-cdk-lib/aws-lambda";
+import { NodejsFunction } from "aws-cdk-lib/aws-lambda-nodejs";
+import { Topic } from "aws-cdk-lib/aws-sns";
+import { LambdaSubscription } from "aws-cdk-lib/aws-sns-subscriptions";
+import { AlarmFactoryDefaults, MonitoringFacade, SnsAlarmActionStrategy } from "cdk-monitoring-constructs";
+import { Construct } from "constructs";
+
+interface MonitoringStackProps extends StackProps {
+  fargateService: FargateService;
+  discordWebhookUrl?: string;
+}
+
+export class MonitoringStack extends Stack {
+  constructor(scope: Construct, id: string, props: MonitoringStackProps) {
+    super(scope, id, props);
+
+    const { fargateService } = props;
+
+    let alarmFactoryDefaults: AlarmFactoryDefaults = {
+      actionsEnabled: false,
+      alarmNamePrefix: "ValheimCDK",
+    };
+
+    if (props.discordWebhookUrl) {
+      const topicForDiscordNotif = new Topic(this, "AlarmTopic");
+      const embedAlarmInDiscordLambda = new NodejsFunction(this, "EmbedAlarmInDiscordLambda", {
+        runtime: Runtime.NODEJS_18_X,
+        entry: 'resources/embedAlarmInDiscordLambda.ts',
+        handler: "handler",
+        bundling: {
+          nodeModules: ['discord.js'],
+        },
+        environment: {
+          DISCORD_WEBHOOK_URL: props.discordWebhookUrl,
+        }
+      });
+      topicForDiscordNotif.addSubscription(new LambdaSubscription(embedAlarmInDiscordLambda));
+
+      alarmFactoryDefaults = {
+        ...alarmFactoryDefaults,
+        actionsEnabled: true,
+        action: new SnsAlarmActionStrategy({
+          onAlarmTopic: topicForDiscordNotif,
+          onOkTopic: topicForDiscordNotif,
+        }),
+      }
+    }
+
+    const monitoring = new MonitoringFacade(this, "Monitoring", {
+      alarmFactoryDefaults,
+    });
+
+    monitoring
+      .monitorSimpleFargateService({
+        fargateService,
+
+        addCpuUsageAlarm: {
+          "CRITICAL": {
+            maxUsagePercent: 95,
+            datapointsToAlarm: 3,
+            period: Duration.minutes(1),
+            treatMissingDataOverride: TreatMissingData.IGNORE,
+          },
+          "WARNING": {
+            maxUsagePercent: 90,
+            datapointsToAlarm: 10,
+            period: Duration.minutes(1),
+            treatMissingDataOverride: TreatMissingData.IGNORE,
+          }
+        },
+        addMemoryUsageAlarm: {
+          "CRITICAL": {
+            maxUsagePercent: 95,
+            datapointsToAlarm: 3,
+            period: Duration.minutes(1),
+            treatMissingDataOverride: TreatMissingData.IGNORE,
+          },
+          "WARNING": {
+            maxUsagePercent: 90,
+            datapointsToAlarm: 10,
+            period: Duration.minutes(1),
+            treatMissingDataOverride: TreatMissingData.IGNORE,
+          }
+        }
+      });
+  }
+}