feat: introduce ComputeStrategy interface and extract AgentCoreComputeStrategy

cdk/package.json

@@ -18,6 +18,7 @@
     "@aws-cdk/mixins-preview": "2.238.0-alpha.0",
     "@aws-sdk/client-bedrock-agentcore": "^3.1021.0",
     "@aws-sdk/client-bedrock-runtime": "^3.1021.0",
+    "@aws-sdk/client-ecs": "^3.1021.0",
     "@aws-sdk/client-dynamodb": "^3.1021.0",
     "@aws-sdk/client-lambda": "^3.1021.0",
     "@aws-sdk/client-secrets-manager": "^3.1021.0",

cdk/src/constructs/ecs-agent-cluster.ts

@@ -0,0 +1,150 @@
+/**
+ *  MIT No Attribution
+ *
+ *  Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
+ *
+ *  Permission is hereby granted, free of charge, to any person obtaining a copy of
+ *  the Software without restriction, including without limitation the rights to
+ *  use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of
+ *  the Software, and to permit persons to whom the Software is furnished to do so.
+ *
+ *  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ *  IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ *  FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ *  AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ *  LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ *  OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ *  SOFTWARE.
+ */
+
+import { RemovalPolicy } from 'aws-cdk-lib';
+import * as dynamodb from 'aws-cdk-lib/aws-dynamodb';
+import * as ec2 from 'aws-cdk-lib/aws-ec2';
+import * as ecr_assets from 'aws-cdk-lib/aws-ecr-assets';
+import * as ecs from 'aws-cdk-lib/aws-ecs';
+import * as iam from 'aws-cdk-lib/aws-iam';
+import * as logs from 'aws-cdk-lib/aws-logs';
+import * as secretsmanager from 'aws-cdk-lib/aws-secretsmanager';
+import { NagSuppressions } from 'cdk-nag';
+import { Construct } from 'constructs';
+
+export interface EcsAgentClusterProps {
+  readonly vpc: ec2.IVpc;
+  readonly agentImageAsset: ecr_assets.DockerImageAsset;
+  readonly taskTable: dynamodb.ITable;
+  readonly taskEventsTable: dynamodb.ITable;
+  readonly userConcurrencyTable: dynamodb.ITable;
+  readonly githubTokenSecret: secretsmanager.ISecret;
+  readonly memoryId?: string;
+}
+
+export class EcsAgentCluster extends Construct {
+  public readonly cluster: ecs.Cluster;
+  public readonly taskDefinition: ecs.FargateTaskDefinition;
+  public readonly securityGroup: ec2.SecurityGroup;
+  public readonly containerName: string;
+
+  constructor(scope: Construct, id: string, props: EcsAgentClusterProps) {
+    super(scope, id);
+
+    this.containerName = 'AgentContainer';
+
+    // ECS Cluster with Fargate capacity provider and container insights
+    this.cluster = new ecs.Cluster(this, 'Cluster', {
+      vpc: props.vpc,
+      containerInsights: true,
+    });
+
+    // Security group — egress TCP 443 only
+    this.securityGroup = new ec2.SecurityGroup(this, 'TaskSG', {
+      vpc: props.vpc,
+      description: 'ECS Agent Tasks - egress TCP 443 only',
+      allowAllOutbound: false,
+    });
+
+    this.securityGroup.addEgressRule(
+      ec2.Peer.anyIpv4(),
+      ec2.Port.tcp(443),
+      'Allow HTTPS egress (GitHub API, AWS services)',
+    );
+
+    // CloudWatch log group for agent task output
+    const logGroup = new logs.LogGroup(this, 'TaskLogGroup', {
+      logGroupName: '/ecs/abca-agent-tasks',
+      retention: logs.RetentionDays.THREE_MONTHS,
+      removalPolicy: RemovalPolicy.DESTROY,
+    });
+
+    // Task execution role (used by ECS agent to pull images, write logs)
+    // CDK creates this automatically via taskDefinition, but we need to
+    // grant additional permissions to the task role.
+
+    // Fargate task definition
+    this.taskDefinition = new ecs.FargateTaskDefinition(this, 'TaskDef', {
+      cpu: 2048,
+      memoryLimitMiB: 4096,
+      runtimePlatform: {
+        cpuArchitecture: ecs.CpuArchitecture.ARM64,
+        operatingSystemFamily: ecs.OperatingSystemFamily.LINUX,
+      },
+    });
+
+    // Container
+    this.taskDefinition.addContainer(this.containerName, {
+      image: ecs.ContainerImage.fromDockerImageAsset(props.agentImageAsset),
+      logging: ecs.LogDrivers.awsLogs({
+        logGroup,
+        streamPrefix: 'agent',
+      }),
+      environment: {
+        CLAUDE_CODE_USE_BEDROCK: '1',
+        TASK_TABLE_NAME: props.taskTable.tableName,
+        TASK_EVENTS_TABLE_NAME: props.taskEventsTable.tableName,
+        USER_CONCURRENCY_TABLE_NAME: props.userConcurrencyTable.tableName,
+        LOG_GROUP_NAME: logGroup.logGroupName,
+        ...(props.memoryId && { MEMORY_ID: props.memoryId }),
+      },
+    });
+
+    // Task role permissions
+    const taskRole = this.taskDefinition.taskRole;
+
+    // DynamoDB read/write on task tables
+    props.taskTable.grantReadWriteData(taskRole);
+    props.taskEventsTable.grantReadWriteData(taskRole);
+    props.userConcurrencyTable.grantReadWriteData(taskRole);
+
+    // Secrets Manager read for GitHub token
+    props.githubTokenSecret.grantRead(taskRole);
+
+    // Bedrock model invocation
+    taskRole.addToPrincipalPolicy(new iam.PolicyStatement({
+      actions: [
+        'bedrock:InvokeModel',
+        'bedrock:InvokeModelWithResponseStream',
+      ],
+      resources: ['*'],
+    }));
+
+    // CloudWatch Logs write
+    logGroup.grantWrite(taskRole);
+
+    NagSuppressions.addResourceSuppressions(this.taskDefinition, [
+      {
+        id: 'AwsSolutions-IAM5',
+        reason: 'DynamoDB index/* wildcards generated by CDK grantReadWriteData; Bedrock InvokeModel requires * resource; Secrets Manager wildcards from CDK grantRead; CloudWatch Logs wildcards from CDK grantWrite',
+      },
+      {
+        id: 'AwsSolutions-ECS2',
+        reason: 'Environment variables contain table names and configuration, not secrets — GitHub token is fetched from Secrets Manager at runtime',
+      },
+    ], true);
+
+    NagSuppressions.addResourceSuppressions(this.cluster, [
+      {
+        id: 'AwsSolutions-ECS4',
+        reason: 'Container insights is enabled via the containerInsights prop',
+      },
+    ], true);
+  }
+}

cdk/src/constructs/task-orchestrator.ts

@@ -100,6 +100,32 @@ export interface TaskOrchestratorProps {
    * and writes fallback episodes during finalization.
    */
   readonly memoryId?: string;
+
+  /**
+   * ARN of the ECS cluster for ECS compute strategy.
+   * When provided, ECS-related env vars and IAM policies are added.
+   */
+  readonly ecsClusterArn?: string;
+
+  /**
+   * ARN of the ECS task definition for ECS compute strategy.
+   */
+  readonly ecsTaskDefinitionArn?: string;
+
+  /**
+   * Comma-separated subnet IDs for ECS tasks.
+   */
+  readonly ecsSubnets?: string;
+
+  /**
+   * Security group ID for ECS tasks.
+   */
+  readonly ecsSecurityGroup?: string;
+
+  /**
+   * Container name in the ECS task definition.
+   */
+  readonly ecsContainerName?: string;
 }
 
 /**
@@ -152,6 +178,11 @@ export class TaskOrchestrator extends Construct {
           USER_PROMPT_TOKEN_BUDGET: String(props.userPromptTokenBudget),
         }),
         ...(props.memoryId && { MEMORY_ID: props.memoryId }),
+        ...(props.ecsClusterArn && { ECS_CLUSTER_ARN: props.ecsClusterArn }),
+        ...(props.ecsTaskDefinitionArn && { ECS_TASK_DEFINITION_ARN: props.ecsTaskDefinitionArn }),
+        ...(props.ecsSubnets && { ECS_SUBNETS: props.ecsSubnets }),
+        ...(props.ecsSecurityGroup && { ECS_SECURITY_GROUP: props.ecsSecurityGroup }),
+        ...(props.ecsContainerName && { ECS_CONTAINER_NAME: props.ecsContainerName }),
       },
       bundling: {
         externalModules: ['@aws-sdk/*'],
@@ -192,6 +223,33 @@ export class TaskOrchestrator extends Construct {
       resources: runtimeResources,
     }));
 
+    // ECS compute strategy permissions (only when ECS is configured)
+    if (props.ecsClusterArn) {
+      this.fn.addToRolePolicy(new iam.PolicyStatement({
+        actions: [
+          'ecs:RunTask',
+          'ecs:DescribeTasks',
+          'ecs:StopTask',
+        ],
+        resources: ['*'],
+        conditions: {
+          ArnEquals: {
+            'ecs:cluster': props.ecsClusterArn,
+          },
+        },
+      }));
+
+      this.fn.addToRolePolicy(new iam.PolicyStatement({
+        actions: ['iam:PassRole'],
+        resources: ['*'],
+        conditions: {
+          StringEquals: {
+            'iam:PassedToService': 'ecs-tasks.amazonaws.com',
+          },
+        },
+      }));
+    }
+
     // Per-repo Secrets Manager grants (e.g. per-repo GitHub tokens from Blueprints)
     for (const [index, secretArn] of (props.additionalSecretArns ?? []).entries()) {
       const secret = secretsmanager.Secret.fromSecretCompleteArn(
@@ -229,7 +287,7 @@ export class TaskOrchestrator extends Construct {
       },
       {
         id: 'AwsSolutions-IAM5',
-        reason: 'DynamoDB index/* wildcards generated by CDK grantReadWriteData; AgentCore runtime/* required for sub-resource invocation; Secrets Manager wildcards generated by CDK grantRead; AgentCore Memory wildcards generated by CDK grantRead/grantWrite',
+        reason: 'DynamoDB index/* wildcards generated by CDK grantReadWriteData; AgentCore runtime/* required for sub-resource invocation; Secrets Manager wildcards generated by CDK grantRead; AgentCore Memory wildcards generated by CDK grantRead/grantWrite; ECS RunTask/DescribeTasks/StopTask conditioned on cluster ARN; iam:PassRole conditioned on ecs-tasks.amazonaws.com',
       },
     ], true);
   }

cdk/src/handlers/orchestrate-task.ts

@@ -19,6 +19,8 @@
 
 import { withDurableExecution, type DurableExecutionHandler } from '@aws/durable-execution-sdk-js';
 import { TaskStatus, TERMINAL_STATUSES } from '../constructs/task-status';
+import { resolveComputeStrategy } from './shared/compute-strategy';
+import { logger } from './shared/logger';
 import {
   admissionControl,
   emitTaskEvent,
@@ -28,7 +30,7 @@ import {
   loadBlueprintConfig,
   loadTask,
   pollTaskStatus,
-  startSession,
+  transitionTask,
   type PollState,
 } from './shared/orchestrator';
 import { runPreflightChecks } from './shared/preflight';
@@ -116,27 +118,71 @@ const durableHandler: DurableExecutionHandler<OrchestrateTaskEvent, void> = asyn
     }
   });
 
-  // Step 4: Start agent session — invoke runtime and transition to RUNNING
-  await context.step('start-session', async () => {
+  // Step 4: Start agent session — resolve compute strategy, invoke runtime, transition to RUNNING
+  // Returns the full SessionHandle (serializable) so ECS polling can use it in step 5.
+  const sessionHandle = await context.step('start-session', async () => {
     try {
-      return await startSession(task, payload, blueprintConfig);
+      const strategy = resolveComputeStrategy(blueprintConfig);
+      const handle = await strategy.startSession({ taskId, payload, blueprintConfig });
+
+      await transitionTask(taskId, TaskStatus.HYDRATING, TaskStatus.RUNNING, {
+        session_id: handle.sessionId,
+        started_at: new Date().toISOString(),
+      });
+      await emitTaskEvent(taskId, 'session_started', {
+        session_id: handle.sessionId,
+        strategy_type: handle.strategyType,
+      });
+
+      logger.info('Session started', {
+        task_id: taskId,
+        session_id: handle.sessionId,
+        strategy_type: handle.strategyType,
+      });
+
+      return handle;
     } catch (err) {
       await failTask(taskId, TaskStatus.HYDRATING, `Session start failed: ${String(err)}`, task.user_id, true);
       throw err;
     }
   });
 
   // Step 5: Wait for agent to finish
-  // NOTE: Polls DynamoDB every 30s rather than re-invoking the AgentCore session.
-  // The agent writes terminal status directly to DDB. If the agent crashes without
-  // writing a terminal status, we detect it via the HYDRATING early-exit check
-  // (MAX_NON_RUNNING_POLLS ~5min); otherwise the loop runs up to MAX_POLL_ATTEMPTS
-  // (~8.5h). A future improvement could add AgentCore session status checks for
-  // faster crash detection.
+  // Polls DynamoDB every 30s. For ECS compute, also polls the ECS task to detect
+  // container crashes that don't write terminal status to DDB. For AgentCore,
+  // crash detection relies on the HYDRATING early-exit (MAX_NON_RUNNING_POLLS ~5min).
   const finalPollState = await context.waitForCondition<PollState>(
     'await-agent-completion',
     async (state) => {
-      return pollTaskStatus(taskId, state);
+      const ddbState = await pollTaskStatus(taskId, state);
+
+      // ECS compute-level crash detection: if DDB is not terminal, check ECS task status
+      if (
+        ddbState.lastStatus &&
+        !TERMINAL_STATUSES.includes(ddbState.lastStatus) &&
+        blueprintConfig.compute_type === 'ecs'
+      ) {
+        try {
+          const strategy = resolveComputeStrategy(blueprintConfig);
+          const ecsStatus = await strategy.pollSession(sessionHandle);
+          if (ecsStatus.status === 'failed') {
+            const errorMsg = 'error' in ecsStatus ? ecsStatus.error : 'ECS task failed';
+            logger.warn('ECS task failed before DDB terminal write', {
+              task_id: taskId,
+              error: errorMsg,
+            });
+            await failTask(taskId, ddbState.lastStatus, `ECS container failed: ${errorMsg}`, task.user_id, true);
+            return { attempts: ddbState.attempts, lastStatus: TaskStatus.FAILED };
+          }
+        } catch (err) {
+          logger.warn('ECS pollSession check failed (non-fatal)', {
+            task_id: taskId,
+            error: err instanceof Error ? err.message : String(err),
+          });
+        }
+      }
+
+      return ddbState;
     },
     {
       initialState: { attempts: 0 },