infra: blue/green deployments

.github/workflows/ci.yml

@@ -24,7 +24,7 @@ jobs:
     needs: [lint, test]
     strategy:
       matrix:
-        app: [gateway, web, websockets, workers]
+        app: [gateway, websockets, workers]
     uses: ./.github/workflows/deploy-app.yml
     with:
       app-name: ${{ matrix.app }}

apps/infra/src/deployments/web.ts

@@ -140,7 +140,7 @@ const taskDefinition = pulumi
       }),
   )
 
-const targetGroup = new aws.lb.TargetGroup('LatitudeLLMAppTg', {
+const greenTargetGroup = new aws.lb.TargetGroup('LatitudeLLMAppTg', {
   port: 8080,
   vpcId,
   protocol: 'HTTP',
@@ -157,46 +157,94 @@ const targetGroup = new aws.lb.TargetGroup('LatitudeLLMAppTg', {
 
 const defaultListenerArn = coreStack.requireOutput('defaultListenerArn')
 
-new aws.lb.ListenerRule('LatitudeLLMAppListenerRule', {
-  listenerArn: defaultListenerArn,
-  actions: [
-    {
-      type: 'forward',
-      targetGroupArn: targetGroup.arn,
-    },
-  ],
-  conditions: [
-    {
-      hostHeader: {
-        values: [DNS_ADDRESS],
-      },
-    },
-  ],
-})
-
 const cluster = coreStack.requireOutput('cluster') as pulumi.Output<Cluster>
-new aws.ecs.Service('LatitudeLLMApp', {
+
+const ecsService = new aws.ecs.Service('LatitudeLLMApp', {
   cluster: cluster.arn,
   taskDefinition: taskDefinition.arn,
   desiredCount: 2,
   launchType: 'FARGATE',
   forceNewDeployment: true,
   enableExecuteCommand: true,
+  deploymentController: {
+    type: 'CODE_DEPLOY',
+  },
   networkConfiguration: {
     subnets: privateSubnets.ids,
     assignPublicIp: false,
     securityGroups: [ecsSecurityGroup],
   },
   loadBalancers: [
     {
-      targetGroupArn: targetGroup.arn,
+      targetGroupArn: greenTargetGroup.arn,
       containerName,
       containerPort: 8080,
     },
   ],
-  triggers: {
-    digest: image.repoDigest,
-    coreDigest: coreImage.repoDigest,
+})
+
+const blueTargetGroup = new aws.lb.TargetGroup('LatitudeLLMAppBlueTg', {
+  port: 8080,
+  vpcId,
+  protocol: 'HTTP',
+  targetType: 'ip',
+  healthCheck: {
+    path: '/api/health',
+    interval: 5,
+    timeout: 2,
+    healthyThreshold: 2,
+    unhealthyThreshold: 2,
+  },
+  deregistrationDelay: 5,
+})
+
+const codeDeployApp = new aws.codedeploy.Application(
+  'LatitudeLLMCodeDeployApp',
+  {
+    computePlatform: 'ECS',
+  },
+)
+
+new aws.codedeploy.DeploymentGroup('LatitudeLLMDeploymentGroup', {
+  appName: codeDeployApp.name,
+  serviceRoleArn: ecsTaskExecutionRole,
+  deploymentConfigName: 'CodeDeployDefault.ECSAllAtOnce',
+  deploymentGroupName: 'LatitudeLLMDeploymentGroup',
+  ecsService: {
+    clusterName: cluster.name,
+    serviceName: ecsService.name,
+  },
+  autoRollbackConfiguration: {
+    enabled: true,
+    events: ['DEPLOYMENT_FAILURE'],
+  },
+  blueGreenDeploymentConfig: {
+    deploymentReadyOption: {
+      actionOnTimeout: 'CONTINUE_DEPLOYMENT',
+      waitTimeInMinutes: 0,
+    },
+    terminateBlueInstancesOnDeploymentSuccess: {
+      action: 'TERMINATE',
+      terminationWaitTimeInMinutes: 5,
+    },
+  },
+  deploymentStyle: {
+    deploymentOption: 'WITH_TRAFFIC_CONTROL',
+    deploymentType: 'BLUE_GREEN',
+  },
+  loadBalancerInfo: {
+    targetGroupPairInfo: {
+      prodTrafficRoute: {
+        listenerArns: [defaultListenerArn],
+      },
+      testTrafficRoute: {
+        listenerArns: [defaultListenerArn],
+      },
+      targetGroups: [
+        { name: blueTargetGroup.name },
+        { name: greenTargetGroup.name },
+      ],
+    },
   },
 })