managed-compute-environment.ts

import { Construct } from 'constructs';
import { CfnComputeEnvironment } from './batch.generated';
import { IComputeEnvironment, ComputeEnvironmentBase, ComputeEnvironmentProps } from './compute-environment-base';
import * as ec2 from '../../aws-ec2';
import * as eks from '../../aws-eks';
import * as iam from '../../aws-iam';
import { IRole } from '../../aws-iam';
import { ArnFormat, Duration, ITaggable, Lazy, Resource, Stack, TagManager, TagType } from '../../core';

/**
 * Represents a Managed ComputeEnvironment. Batch will provision EC2 Instances to
 * meet the requirements of the jobs executing in this ComputeEnvironment.
 */
export interface IManagedComputeEnvironment extends IComputeEnvironment, ec2.IConnectable, ITaggable {
  /**
   * The maximum vCpus this `ManagedComputeEnvironment` can scale up to.
   *
   * *Note*: if this Compute Environment uses EC2 resources (not Fargate) with either `AllocationStrategy.BEST_FIT_PROGRESSIVE` or
   * `AllocationStrategy.SPOT_CAPACITY_OPTIMIZED`, or `AllocationStrategy.BEST_FIT` with Spot instances,
   * The scheduler may exceed this number by at most one of the instances specified in `instanceTypes`
   * or `instanceClasses`.
   */
  readonly maxvCpus: number;

  /**
   * Specifies whether this Compute Environment is replaced if an update is made that requires
   * replacing its instances. To enable more properties to be updated,
   * set this property to `false`. When changing the value of this property to false,
   * do not change any other properties at the same time.
   * If other properties are changed at the same time,
   * and the change needs to be rolled back but it can't,
   * it's possible for the stack to go into the UPDATE_ROLLBACK_FAILED state.
   * You can't update a stack that is in the UPDATE_ROLLBACK_FAILED state.
   * However, if you can continue to roll it back,
   * you can return the stack to its original settings and then try to update it again.
   *
   * The properties which require a replacement of the Compute Environment are:
   *
   * @see https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-resource-batch-computeenvironment.html#cfn-batch-computeenvironment-replacecomputeenvironment
   * @see https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/using-cfn-updating-stacks-continueupdaterollback.html
   */
  readonly replaceComputeEnvironment?: boolean;

  /**
   * Whether or not to use spot instances.
   * Spot instances are less expensive EC2 instances that can be
   * reclaimed by EC2 at any time; your job will be given two minutes
   * of notice before reclamation.
   *
   * @default false
   */
  readonly spot?: boolean;

  /**
   * Only meaningful if `terminateOnUpdate` is `false`. If so,
   * when an infrastructure update is triggered, any running jobs
   * will be allowed to run until `updateTimeout` has expired.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/updating-compute-environments.html
   * @default 30 minutes
   */
  readonly updateTimeout?: Duration;

  /**
   * Whether or not any running jobs will be immediately terminated when an infrastructure update
   * occurs. If this is enabled, any terminated jobs may be retried, depending on the job's
   * retry policy.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/updating-compute-environments.html
   *
   * @default false
   */
  readonly terminateOnUpdate?: boolean;

  /**
   * The security groups this Compute Environment will launch instances in.
   */
  readonly securityGroups: ec2.ISecurityGroup[];

  /**
   * The VPC Subnets this Compute Environment will launch instances in.
   */
  readonly vpcSubnets?: ec2.SubnetSelection;

  /**
   * Whether or not the AMI is updated to the latest one supported by Batch
   * when an infrastructure update occurs.
   *
   * If you specify a specific AMI, this property will be ignored.
   *
   * Note: the CDK will never set this value by default, `false` will set by CFN.
   * This is to avoid a deployment failure that occurs when this value is set.
   *
   * @see https://github.com/aws/aws-cdk/issues/27054
   *
   * @default false
   */
  readonly updateToLatestImageVersion?: boolean;
}

/**
 * Props for a ManagedComputeEnvironment
 */
export interface ManagedComputeEnvironmentProps extends ComputeEnvironmentProps {
  /**
  * The maximum vCpus this `ManagedComputeEnvironment` can scale up to.
  * Each vCPU is equivalent to 1024 CPU shares.
  *
  * *Note*: if this Compute Environment uses EC2 resources (not Fargate) with either `AllocationStrategy.BEST_FIT_PROGRESSIVE` or
  * `AllocationStrategy.SPOT_CAPACITY_OPTIMIZED`, or `AllocationStrategy.BEST_FIT` with Spot instances,
  * The scheduler may exceed this number by at most one of the instances specified in `instanceTypes`
  * or `instanceClasses`.
  *
  * @default 256
  */
  readonly maxvCpus?: number;

  /**
  * Specifies whether this Compute Environment is replaced if an update is made that requires
  * replacing its instances. To enable more properties to be updated,
  * set this property to `false`. When changing the value of this property to false,
  * do not change any other properties at the same time.
  * If other properties are changed at the same time,
  * and the change needs to be rolled back but it can't,
  * it's possible for the stack to go into the UPDATE_ROLLBACK_FAILED state.
  * You can't update a stack that is in the UPDATE_ROLLBACK_FAILED state.
  * However, if you can continue to roll it back,
  * you can return the stack to its original settings and then try to update it again.
  *
  * The properties which require a replacement of the Compute Environment are:
  *
  * @see https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/aws-resource-batch-computeenvironment.html#cfn-batch-computeenvironment-replacecomputeenvironment
  * @see https://docs.aws.amazon.com/AWSCloudFormation/latest/UserGuide/using-cfn-updating-stacks-continueupdaterollback.html
  *
  * @default false
  */
  readonly replaceComputeEnvironment?: boolean;

  /**
   * Whether or not to use spot instances.
   * Spot instances are less expensive EC2 instances that can be
   * reclaimed by EC2 at any time; your job will be given two minutes
   * of notice before reclamation.
   *
   * @default false
   */
  readonly spot?: boolean;

  /**
   * Only meaningful if `terminateOnUpdate` is `false`. If so,
   * when an infrastructure update is triggered, any running jobs
   * will be allowed to run until `updateTimeout` has expired.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/updating-compute-environments.html
   *
   * @default 30 minutes
   */
  readonly updateTimeout?: Duration;

  /**
   * Whether or not any running jobs will be immediately terminated when an infrastructure update
   * occurs. If this is enabled, any terminated jobs may be retried, depending on the job's
   * retry policy.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/updating-compute-environments.html
   *
   * @default false
   */
  readonly terminateOnUpdate?: boolean;

  /**
   * VPC in which this Compute Environment will launch Instances
   */
  readonly vpc: ec2.IVpc;

  /**
  * The security groups this Compute Environment will launch instances in.
  *
  * @default new security groups will be created
  */
  readonly securityGroups?: ec2.ISecurityGroup[];

  /**
  * The VPC Subnets this Compute Environment will launch instances in.
  *
  * @default new subnets will be created
  */
  readonly vpcSubnets?: ec2.SubnetSelection;

  /**
  * Whether or not the AMI is updated to the latest one supported by Batch
  * when an infrastructure update occurs.
  *
  * If you specify a specific AMI, this property will be ignored.
  *
  * @default true
  */
  readonly updateToLatestImageVersion?: boolean;
}

/**
 * Abstract base class for ManagedComputeEnvironments
 * @internal
 */
export abstract class ManagedComputeEnvironmentBase extends ComputeEnvironmentBase implements IManagedComputeEnvironment {
  public readonly maxvCpus: number;
  public readonly replaceComputeEnvironment?: boolean;
  public readonly spot?: boolean;
  public readonly updateTimeout?: Duration;
  public readonly terminateOnUpdate?: boolean;
  public readonly securityGroups: ec2.ISecurityGroup[];
  public readonly updateToLatestImageVersion?: boolean;
  public readonly tags: TagManager = new TagManager(TagType.MAP, 'AWS::Batch::ComputeEnvironment');

  public readonly connections: ec2.Connections;

  constructor(scope: Construct, id: string, props: ManagedComputeEnvironmentProps) {
    super(scope, id, props);

    this.maxvCpus = props.maxvCpus ?? DEFAULT_MAX_VCPUS;
    this.replaceComputeEnvironment = props.replaceComputeEnvironment ?? false;
    this.spot = props.spot;
    this.updateTimeout = props.updateTimeout;
    this.terminateOnUpdate = props.terminateOnUpdate;
    this.updateToLatestImageVersion = props.updateToLatestImageVersion;
    this.securityGroups = props.securityGroups ?? [
      new ec2.SecurityGroup(this, 'SecurityGroup', {
        vpc: props.vpc,
      }),
    ];
    this.connections = new ec2.Connections({
      securityGroups: this.securityGroups,
    });
  }
}

/**
 * A ManagedComputeEnvironment that uses ECS orchestration on EC2 instances.
 */
export interface IManagedEc2EcsComputeEnvironment extends IManagedComputeEnvironment {
  /**
   * Configure which AMIs this Compute Environment can launch.
   *
   * Leave this `undefined` to allow Batch to choose the latest AMIs it supports for each instance that it launches.
   *
   * @default
   * - ECS_AL2 compatible AMI ids for non-GPU instances, ECS_AL2_NVIDIA compatible AMI ids for GPU instances
   */
  readonly images?: EcsMachineImage[];

  /**
   * The allocation strategy to use if not enough instances of
   * the best fitting instance type can be allocated.
   *
   * @default - `BEST_FIT_PROGRESSIVE` if not using Spot instances,
   * `SPOT_CAPACITY_OPTIMIZED` if using Spot instances.
   */
  readonly allocationStrategy?: AllocationStrategy;

  /**
   * The maximum percentage that a Spot Instance price can be when compared with the
   * On-Demand price for that instance type before instances are launched.
   * For example, if your maximum percentage is 20%, the Spot price must be
   * less than 20% of the current On-Demand price for that Instance.
   * You always pay the lowest market price and never more than your maximum percentage.
   * For most use cases, Batch recommends leaving this field empty.
   *
   * @default - 100%
   */
  readonly spotBidPercentage?: number;

  /**
   * The service-linked role that Spot Fleet needs to launch instances on your behalf.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/spot_fleet_IAM_role.html
   *
   * @default - a new Role will be created
   */
  readonly spotFleetRole?: iam.IRole;

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   */
  readonly instanceTypes: ec2.InstanceType[];

  /**
   * The instance classes that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   * Batch will automatically choose the size.
   */
  readonly instanceClasses: ec2.InstanceClass[];

  /**
   * Whether or not to use batch's optimal instance type.
   * The optimal instance type is equivalent to adding the
   * C4, M4, and R4 instance classes. You can specify other instance classes
   * (of the same architecture) in addition to the optimal instance classes.
   *
   * @default true
   */
  readonly useOptimalInstanceClasses?: boolean;

  /**
   * The execution Role that instances launched by this Compute Environment will use.
   *
   * @default - a role will be created
   */
  readonly instanceRole?: iam.IRole;

  /**
   * The Launch Template that this Compute Environment
   * will use to provision EC2 Instances.
   *
   * *Note*: if `securityGroups` is specified on both your
   * launch template and this Compute Environment, **the
   * `securityGroup`s on the Compute Environment override the
   * ones on the launch template.
   *
   * @default no launch template
   */
  readonly launchTemplate?: ec2.ILaunchTemplate;

  /**
   * The minimum vCPUs that an environment should maintain,
   * even if the compute environment is DISABLED.
   *
   * @default 0
   */
  readonly minvCpus?: number;

  /**
   * The EC2 placement group to associate with your compute resources.
   * If you intend to submit multi-node parallel jobs to this Compute Environment,
   * you should consider creating a cluster placement group and associate it with your compute resources.
   * This keeps your multi-node parallel job on a logical grouping of instances
   * within a single Availability Zone with high network flow potential.
   *
   * @see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html
   *
   * @default - no placement group
   */
  readonly placementGroup?: ec2.IPlacementGroup;

  /**
   * Add an instance type to this compute environment
   */
  addInstanceType(instanceType: ec2.InstanceType): void;

  /**
   * Add an instance class to this compute environment
   */
  addInstanceClass(instanceClass: ec2.InstanceClass): void;
}

/**
 * Base interface for containing all information needed to
 * configure a MachineImage in Batch
 */
interface MachineImage {
  /**
   * The machine image to use
   *
   * @default - chosen by batch
   */
  readonly image?: ec2.IMachineImage;
}

/**
 * A Batch MachineImage that is compatible with ECS
 */
export interface EcsMachineImage extends MachineImage {
  /**
   * Tells Batch which instance type to launch this image on
   *
   * @default - 'ECS_AL2' for non-gpu instances, 'ECS_AL2_NVIDIA' for gpu instances
   */
  readonly imageType?: EcsMachineImageType;
}

/**
 * A Batch MachineImage that is compatible with EKS
 */
export interface EksMachineImage extends MachineImage{
  /**
   * Tells Batch which instance type to launch this image on
   *
   * @default - 'EKS_AL2' for non-gpu instances, 'EKS_AL2_NVIDIA' for gpu instances
   */
  readonly imageType?: EksMachineImageType;
}

/**
 * Maps the image to instance types
 */
export enum EcsMachineImageType {
  /**
   * Tells Batch that this machine image runs on non-GPU instances
   */
  ECS_AL2 = 'ECS_AL2',

  /**
   * Tells Batch that this machine image runs on GPU instances
   */
  ECS_AL2_NVIDIA = 'ECS_AL2_NVIDIA',
}

/**
 * Maps the image to instance types
 */
export enum EksMachineImageType {
  /**
   * Tells Batch that this machine image runs on non-GPU instances
   */
  EKS_AL2 = 'EKS_AL2',

  /**
   * Tells Batch that this machine image runs on GPU instances
   */
  EKS_AL2_NVIDIA = 'EKS_AL2_NVIDIA',
}

/**
 * Determines how this compute environment chooses instances to spawn
 *
 * @see https://aws.amazon.com/blogs/compute/optimizing-for-cost-availability-and-throughput-by-selecting-your-aws-batch-allocation-strategy/
 */
export enum AllocationStrategy {
  /**
   * Batch chooses the lowest-cost instance type that fits all the jobs in the queue.
   * If instances of that type are not available, the queue will not choose a new type;
   * instead, it will wait for the instance to become available.
   * This can stall your `Queue`, with your compute environment only using part of its max capacity
   * (or none at all) until the `BEST_FIT` instance becomes available.
   * This allocation strategy keeps costs lower but can limit scaling.
   * `BEST_FIT` isn't supported when updating compute environments
   */
  BEST_FIT = 'BEST_FIT',

  /**
   * This is the default Allocation Strategy if `spot` is `false` or unspecified.
   * This strategy will examine the Jobs in the queue and choose whichever instance type meets the requirements
   * of the jobs in the queue and with the lowest cost per vCPU, just as `BEST_FIT`.
   * However, if not all of the capacity can be filled with this instance type,
   * it will choose a new next-best instance type to run any jobs that couldn’t fit into the `BEST_FIT` capacity.
   * To make the most use of this allocation strategy,
   * it is recommended to use as many instance classes as is feasible for your workload.
   */
  BEST_FIT_PROGRESSIVE = 'BEST_FIT_PROGRESSIVE',

  /**
   * If your workflow tolerates interruptions, you should enable `spot` on your `ComputeEnvironment`
   * and use `SPOT_CAPACITY_OPTIMIZED` (this is the default if `spot` is enabled).
   * This will tell Batch to choose the instance types from the ones you’ve specified that have
   * the most spot capacity available to minimize the chance of interruption.
   * To get the most benefit from your spot instances,
   * you should allow Batch to choose from as many different instance types as possible.
   */
  SPOT_CAPACITY_OPTIMIZED = 'SPOT_CAPACITY_OPTIMIZED',

  /**
   * The price and capacity optimized allocation strategy looks at both price and capacity
   * to select the Spot Instance pools that are the least likely to be interrupted
   * and have the lowest possible price.
   *
   * The Batch team recommends this over `SPOT_CAPACITY_OPTIMIZED` in most instances.
   */
  SPOT_PRICE_CAPACITY_OPTIMIZED = 'SPOT_PRICE_CAPACITY_OPTIMIZED',
}

/**
 * Props for a ManagedEc2EcsComputeEnvironment
 */
export interface ManagedEc2EcsComputeEnvironmentProps extends ManagedComputeEnvironmentProps {
  /**
   * Whether or not to use batch's optimal instance type.
   * The optimal instance type is equivalent to adding the
   * C4, M4, and R4 instance classes. You can specify other instance classes
   * (of the same architecture) in addition to the optimal instance classes.
   *
   * @default true
   */
  readonly useOptimalInstanceClasses?: boolean;

  /**
   * Configure which AMIs this Compute Environment can launch.
   * If you specify this property with only `image` specified, then the
   * `imageType` will default to `ECS_AL2`. *If your image needs GPU resources,
   * specify `ECS_AL2_NVIDIA`; otherwise, the instances will not be able to properly
   * join the ComputeEnvironment*.
   *
   * @default
   * - ECS_AL2 for non-GPU instances, ECS_AL2_NVIDIA for GPU instances
   */
  readonly images?: EcsMachineImage[];

  /**
   * The allocation strategy to use if not enough instances of
   * the best fitting instance type can be allocated.
   *
   * @default - `BEST_FIT_PROGRESSIVE` if not using Spot instances,
   * `SPOT_CAPACITY_OPTIMIZED` if using Spot instances.
   */
  readonly allocationStrategy?: AllocationStrategy;

  /**
   * The maximum percentage that a Spot Instance price can be when compared with the
   * On-Demand price for that instance type before instances are launched.
   * For example, if your maximum percentage is 20%, the Spot price must be
   * less than 20% of the current On-Demand price for that Instance.
   * You always pay the lowest market price and never more than your maximum percentage.
   * For most use cases, Batch recommends leaving this field empty.
   *
   * Implies `spot == true` if set
   *
   * @default 100%
   */
  readonly spotBidPercentage?: number;

  /**
   * The service-linked role that Spot Fleet needs to launch instances on your behalf.
   *
   * @see https://docs.aws.amazon.com/batch/latest/userguide/spot_fleet_IAM_role.html
   *
   * @default - a new role will be created
   */
  readonly spotFleetRole?: iam.IRole;

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   *
   * @default - the instances Batch considers will be used (currently C4, M4, and R4)
   */
  readonly instanceTypes?: ec2.InstanceType[];

  /**
   * The instance classes that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   * Batch will automatically choose the instance size.
   *
   * @default - the instances Batch considers will be used (currently C4, M4, and R4)
   */
  readonly instanceClasses?: ec2.InstanceClass[];

  /**
   * The execution Role that instances launched by this Compute Environment will use.
   *
   * @default - a role will be created
   */
  readonly instanceRole?: iam.IRole;

  /**
   * The Launch Template that this Compute Environment
   * will use to provision EC2 Instances.
   *
   * *Note*: if `securityGroups` is specified on both your
   * launch template and this Compute Environment, **the
   * `securityGroup`s on the Compute Environment override the
   * ones on the launch template.
   *
   * @default no launch template
   */
  readonly launchTemplate?: ec2.ILaunchTemplate;

  /**
   * The minimum vCPUs that an environment should maintain,
   * even if the compute environment is DISABLED.
   *
   * @default 0
   */
  readonly minvCpus?: number;

  /**
   * The EC2 placement group to associate with your compute resources.
   * If you intend to submit multi-node parallel jobs to this Compute Environment,
   * you should consider creating a cluster placement group and associate it with your compute resources.
   * This keeps your multi-node parallel job on a logical grouping of instances
   * within a single Availability Zone with high network flow potential.
   *
   * @see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html
   *
   * @default - no placement group
   */
  readonly placementGroup?: ec2.IPlacementGroup;
}

/**
 * A ManagedComputeEnvironment that uses ECS orchestration on EC2 instances.
 *
 * @resource AWS::Batch::ComputeEnvironment
 */
export class ManagedEc2EcsComputeEnvironment extends ManagedComputeEnvironmentBase implements IManagedEc2EcsComputeEnvironment {
  /**
   * refer to an existing ComputeEnvironment by its arn.
   */
  public static fromManagedEc2EcsComputeEnvironmentArn(
    scope: Construct, id: string, managedEc2EcsComputeEnvironmentArn: string,
  ): IManagedEc2EcsComputeEnvironment {
    const stack = Stack.of(scope);
    const computeEnvironmentName = stack.splitArn(managedEc2EcsComputeEnvironmentArn, ArnFormat.SLASH_RESOURCE_NAME).resourceName!;

    class Import extends Resource implements IManagedEc2EcsComputeEnvironment {
      public readonly computeEnvironmentArn = managedEc2EcsComputeEnvironmentArn;
      public readonly computeEnvironmentName = computeEnvironmentName;
      public readonly enabled = true;
      public readonly instanceClasses = [];
      public readonly instanceTypes = [];
      public readonly maxvCpus = 1;
      public readonly connections = { } as any;
      public readonly securityGroups = [];
      public readonly tags: TagManager = new TagManager(TagType.MAP, 'AWS::Batch::ComputeEnvironment');

      public addInstanceClass(_instanceClass: ec2.InstanceClass): void {
        throw new Error(`cannot add instance class to imported ComputeEnvironment '${id}'`);
      }
      public addInstanceType(_instanceType: ec2.InstanceType): void {
        throw new Error(`cannot add instance type to imported ComputeEnvironment '${id}'`);
      }
    }

    return new Import(scope, id);
  }
  public readonly computeEnvironmentArn: string;
  public readonly computeEnvironmentName: string;

  public readonly images?: EcsMachineImage[];
  public readonly allocationStrategy?: AllocationStrategy;
  public readonly spotBidPercentage?: number;
  public readonly spotFleetRole?: iam.IRole;
  public readonly instanceTypes: ec2.InstanceType[];
  public readonly instanceClasses: ec2.InstanceClass[];
  public readonly instanceRole?: iam.IRole;
  public readonly launchTemplate?: ec2.ILaunchTemplate;
  public readonly minvCpus?: number;
  public readonly placementGroup?: ec2.IPlacementGroup;

  private readonly instanceProfile: iam.CfnInstanceProfile;

  constructor(scope: Construct, id: string, props: ManagedEc2EcsComputeEnvironmentProps) {
    super(scope, id, props);

    this.images = props.images;
    this.allocationStrategy = determineAllocationStrategy(id, props.allocationStrategy, this.spot);
    this.spotBidPercentage = props.spotBidPercentage;

    this.spotFleetRole = props.spotFleetRole ?? (
      this.spot && this.allocationStrategy === AllocationStrategy.BEST_FIT
        ? createSpotFleetRole(this)
        : undefined
    );

    this.instanceTypes = props.instanceTypes ?? [];
    this.instanceClasses = props.instanceClasses ?? [];

    const { instanceRole, instanceProfile } = createInstanceRoleAndProfile(this, props.instanceRole);
    this.instanceRole = instanceRole;
    this.instanceProfile = instanceProfile;

    this.launchTemplate = props.launchTemplate;
    this.minvCpus = props.minvCpus ?? DEFAULT_MIN_VCPUS;
    this.placementGroup = props.placementGroup;

    validateVCpus(id, this.minvCpus, this.maxvCpus);
    validateSpotConfig(id, this.spot, this.spotBidPercentage, this.spotFleetRole);

    const { subnetIds } = props.vpc.selectSubnets(props.vpcSubnets);
    const resource = new CfnComputeEnvironment(this, 'Resource', {
      ...baseManagedResourceProperties(this, subnetIds),
      computeEnvironmentName: props.computeEnvironmentName,
      computeResources: {
        ...baseManagedResourceProperties(this, subnetIds).computeResources as CfnComputeEnvironment.ComputeResourcesProperty,
        minvCpus: this.minvCpus,
        instanceRole: this.instanceProfile.attrArn, // this is not a typo; this property actually takes a profile, not a standard role
        instanceTypes: Lazy.list({
          produce: () => renderInstances(this.instanceTypes, this.instanceClasses, props.useOptimalInstanceClasses),
        }),
        type: this.spot ? 'SPOT' : 'EC2',
        spotIamFleetRole: this.spotFleetRole?.roleArn,
        allocationStrategy: this.allocationStrategy,
        bidPercentage: this.spotBidPercentage,
        launchTemplate: this.launchTemplate ? {
          launchTemplateId: this.launchTemplate?.launchTemplateId,
        } : undefined,
        ec2Configuration: this.images?.map((image) => {
          return {
            imageIdOverride: image.image?.getImage(this).imageId,
            imageType: image.imageType ?? EcsMachineImageType.ECS_AL2,
          };
        }),
        placementGroup: this.placementGroup?.placementGroupName,
        tags: this.tags.renderedTags as any,
      },
    });

    this.computeEnvironmentName = this.getResourceNameAttribute(resource.ref);
    this.computeEnvironmentArn = this.getResourceArnAttribute(resource.attrComputeEnvironmentArn, {
      service: 'batch',
      resource: 'compute-environment',
      resourceName: this.physicalName,
    });

    this.node.addValidation({ validate: () => validateInstances(this.instanceTypes, this.instanceClasses, props.useOptimalInstanceClasses) });
  }

  public addInstanceType(instanceType: ec2.InstanceType): void {
    this.instanceTypes.push(instanceType);
  }

  public addInstanceClass(instanceClass: ec2.InstanceClass): void {
    this.instanceClasses.push(instanceClass);
  }
}

/**
 * A ManagedComputeEnvironment that uses EKS orchestration on EC2 instances.
 */
interface IManagedEc2EksComputeEnvironment extends IManagedComputeEnvironment {
  /**
   * The namespace of the Cluster
   *
   * Cannot be 'default', start with 'kube-', or be longer than 64 characters.
   *
   * @see https://kubernetes.io/docs/concepts/overview/working-with-objects/namespaces/
   */
  readonly kubernetesNamespace?: string;

  /**
   * The cluster that backs this Compute Environment. Required
   * for Compute Environments running Kubernetes jobs.
   *
   * Please ensure that you have followed the steps at
   *
   * https://docs.aws.amazon.com/batch/latest/userguide/getting-started-eks.html
   *
   * before attempting to deploy a `ManagedEc2EksComputeEnvironment` that uses this cluster.
   * If you do not follow the steps in the link, the deployment fail with a message that the
   * compute environment did not stabilize.
   */
  readonly eksCluster: eks.ICluster;

  /**
   * Configure which AMIs this Compute Environment can launch.
   *
   * @default
   * EKS_AL2 for non-GPU instances, EKS_AL2_NVIDIA for GPU instances,
   */
  readonly images?: EksMachineImage[];

  /**
   * The allocation strategy to use if not enough instances of
   * the best fitting instance type can be allocated.
   *
   * @default - `BEST_FIT_PROGRESSIVE` if not using Spot instances,
   * `SPOT_CAPACITY_OPTIMIZED` if using Spot instances.
   */
  readonly allocationStrategy?: AllocationStrategy;

  /**
   * The maximum percentage that a Spot Instance price can be when compared with the
   * On-Demand price for that instance type before instances are launched.
   * For example, if your maximum percentage is 20%, the Spot price must be
   * less than 20% of the current On-Demand price for that Instance.
   * You always pay the lowest market price and never more than your maximum percentage.
   * For most use cases, Batch recommends leaving this field empty.
   *
   * Implies `spot == true` if set
   *
   * @default - 100%
   */
  readonly spotBidPercentage?: number;

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   */
  readonly instanceTypes: ec2.InstanceType[];

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   */
  readonly instanceClasses: ec2.InstanceClass[];

  /**
   * The execution Role that instances launched by this Compute Environment will use.
   *
   * @default - a role will be created
   */
  readonly instanceRole?: iam.IRole;

  /**
   * The Launch Template that this Compute Environment
   * will use to provision EC2 Instances.
   *
   * *Note*: if `securityGroups` is specified on both your
   * launch template and this Compute Environment, **the
   * `securityGroup`s on the Compute Environment override the
   * ones on the launch template.
   *
   * @default - no launch template
   */
  readonly launchTemplate?: ec2.ILaunchTemplate;

  /**
   * The minimum vCPUs that an environment should maintain,
   * even if the compute environment is DISABLED.
   *
   * @default 0
   */
  readonly minvCpus?: number;

  /**
   * The EC2 placement group to associate with your compute resources.
   * If you intend to submit multi-node parallel jobs to this Compute Environment,
   * you should consider creating a cluster placement group and associate it with your compute resources.
   * This keeps your multi-node parallel job on a logical grouping of instances
   * within a single Availability Zone with high network flow potential.
   *
   * @see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html
   *
   * @default - no placement group
   */
  readonly placementGroup?: ec2.IPlacementGroup;

  /**
   * Add an instance type to this compute environment
   */
  addInstanceType(instanceType: ec2.InstanceType): void;

  /**
   * Add an instance class to this compute environment
   */
  addInstanceClass(instanceClass: ec2.InstanceClass): void;
}

/**
 * Props for a ManagedEc2EksComputeEnvironment
 */
export interface ManagedEc2EksComputeEnvironmentProps extends ManagedComputeEnvironmentProps {
  /**
   * The namespace of the Cluster
   */
  readonly kubernetesNamespace: string;

  /**
   * The cluster that backs this Compute Environment. Required
   * for Compute Environments running Kubernetes jobs.
   *
   * Please ensure that you have followed the steps at
   *
   * https://docs.aws.amazon.com/batch/latest/userguide/getting-started-eks.html
   *
   * before attempting to deploy a `ManagedEc2EksComputeEnvironment` that uses this cluster.
   * If you do not follow the steps in the link, the deployment fail with a message that the
   * compute environment did not stabilize.
   */
  readonly eksCluster: eks.ICluster;

  /**
   * Whether or not to use batch's optimal instance type.
   * The optimal instance type is equivalent to adding the
   * C4, M4, and R4 instance classes. You can specify other instance classes
   * (of the same architecture) in addition to the optimal instance classes.
   *
   * @default true
   */
  readonly useOptimalInstanceClasses?: boolean;

  /**
   * Configure which AMIs this Compute Environment can launch.
   *
   * @default
   * If `imageKubernetesVersion` is specified,
   * - EKS_AL2 for non-GPU instances, EKS_AL2_NVIDIA for GPU instances,
   * Otherwise,
   * - ECS_AL2 for non-GPU instances, ECS_AL2_NVIDIA for GPU instances,
   */
  readonly images?: EksMachineImage[];

  /**
   * The allocation strategy to use if not enough instances of
   * the best fitting instance type can be allocated.
   *
   * @default - `BEST_FIT_PROGRESSIVE` if not using Spot instances,
   * `SPOT_CAPACITY_OPTIMIZED` if using Spot instances.
   */
  readonly allocationStrategy?: AllocationStrategy;

  /**
   * The maximum percentage that a Spot Instance price can be when compared with the
   * On-Demand price for that instance type before instances are launched.
   * For example, if your maximum percentage is 20%, the Spot price must be
   * less than 20% of the current On-Demand price for that Instance.
   * You always pay the lowest market price and never more than your maximum percentage.
   * For most use cases, Batch recommends leaving this field empty.
   *
   * Implies `spot == true` if set
   *
   * @default - 100%
   */
  readonly spotBidPercentage?: number;

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   *
   * @default - the instances Batch considers will be used (currently C4, M4, and R4)
   */
  readonly instanceTypes?: ec2.InstanceType[];

  /**
   * The instance types that this Compute Environment can launch.
   * Which one is chosen depends on the `AllocationStrategy` used.
   * Batch will automatically choose the instance size.
   *
   * @default - the instances Batch considers will be used (currently C4, M4, and R4)
   */
  readonly instanceClasses?: ec2.InstanceClass[];

  /**
   * The execution Role that instances launched by this Compute Environment will use.
   *
   * @default - a role will be created
   */
  readonly instanceRole?: iam.IRole;

  /**
   * The Launch Template that this Compute Environment
   * will use to provision EC2 Instances.
   *
   * *Note*: if `securityGroups` is specified on both your
   * launch template and this Compute Environment, **the
   * `securityGroup`s on the Compute Environment override the
   * ones on the launch template.**
   *
   * @default - no launch template
   */
  readonly launchTemplate?: ec2.ILaunchTemplate;

  /**
   * The minimum vCPUs that an environment should maintain,
   * even if the compute environment is DISABLED.
   *
   * @default 0
   */
  readonly minvCpus?: number;

  /**
   * The EC2 placement group to associate with your compute resources.
   * If you intend to submit multi-node parallel jobs to this Compute Environment,
   * you should consider creating a cluster placement group and associate it with your compute resources.
   * This keeps your multi-node parallel job on a logical grouping of instances
   * within a single Availability Zone with high network flow potential.
   *
   * @see https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/placement-groups.html
   *
   * @default - no placement group
   */
  readonly placementGroup?: ec2.IPlacementGroup;
}

/**
 * A ManagedComputeEnvironment that uses ECS orchestration on EC2 instances.
 *
 * @resource AWS::Batch::ComputeEnvironment
 */
export class ManagedEc2EksComputeEnvironment extends ManagedComputeEnvironmentBase implements IManagedEc2EksComputeEnvironment {
  public readonly kubernetesNamespace?: string;
  public readonly eksCluster: eks.ICluster;

  public readonly computeEnvironmentName: string;
  public readonly computeEnvironmentArn: string;

  public readonly images?: EksMachineImage[];
  public readonly allocationStrategy?: AllocationStrategy;
  public readonly spotBidPercentage?: number;
  public readonly instanceTypes: ec2.InstanceType[];
  public readonly instanceClasses: ec2.InstanceClass[];
  public readonly instanceRole?: iam.IRole;
  public readonly launchTemplate?: ec2.ILaunchTemplate;
  public readonly minvCpus?: number;
  public readonly placementGroup?: ec2.IPlacementGroup;

  private readonly instanceProfile: iam.CfnInstanceProfile;

  constructor(scope: Construct, id: string, props: ManagedEc2EksComputeEnvironmentProps) {
    super(scope, id, props);

    this.kubernetesNamespace = props.kubernetesNamespace;
    this.eksCluster = props.eksCluster;

    this.images = props.images;
    this.allocationStrategy = determineAllocationStrategy(id, props.allocationStrategy, this.spot);
    if (this.allocationStrategy === AllocationStrategy.BEST_FIT) {
      throw new Error(`ManagedEc2EksComputeEnvironment '${id}' uses invalid allocation strategy 'AllocationStrategy.BEST_FIT'`);
    }
    this.spotBidPercentage = props.spotBidPercentage;
    this.instanceTypes = props.instanceTypes ?? [];
    this.instanceClasses = props.instanceClasses ?? [];

    const { instanceRole, instanceProfile } = createInstanceRoleAndProfile(this, props.instanceRole);
    this.instanceRole = instanceRole;
    this.instanceProfile = instanceProfile;

    this.launchTemplate = props.launchTemplate;
    this.minvCpus = props.minvCpus ?? DEFAULT_MIN_VCPUS;
    this.placementGroup = props.placementGroup;

    validateVCpus(id, this.minvCpus, this.maxvCpus);
    validateSpotConfig(id, this.spot, this.spotBidPercentage);

    const { subnetIds } = props.vpc.selectSubnets(props.vpcSubnets);
    const resource = new CfnComputeEnvironment(this, 'Resource', {
      ...baseManagedResourceProperties(this, subnetIds),
      computeEnvironmentName: props.computeEnvironmentName,
      eksConfiguration: {
        eksClusterArn: this.eksCluster.clusterArn,
        kubernetesNamespace: this.kubernetesNamespace,
      },
      computeResources: {
        ...baseManagedResourceProperties(this, subnetIds).computeResources as CfnComputeEnvironment.ComputeResourcesProperty,
        minvCpus: this.minvCpus,
        instanceRole: this.instanceProfile.attrArn, // this is not a typo; this property actually takes a profile, not a standard role
        instanceTypes: Lazy.list({ produce: () => renderInstances(this.instanceTypes, this.instanceClasses, props.useOptimalInstanceClasses) }),
        type: this.spot ? 'SPOT' : 'EC2',
        allocationStrategy: this.allocationStrategy,
        bidPercentage: this.spotBidPercentage,
        launchTemplate: this.launchTemplate ? {
          launchTemplateId: this.launchTemplate?.launchTemplateId,
        } : undefined,
        ec2Configuration: this.images?.map((image) => {
          return {
            imageIdOverride: image.image?.getImage(this).imageId,
            imageType: image.imageType ?? EksMachineImageType.EKS_AL2,
          };
        }),
        placementGroup: this.placementGroup?.placementGroupName,
        tags: this.tags.renderedTags as any,
      },
    });

    this.computeEnvironmentName = this.getResourceNameAttribute(resource.ref);
    this.computeEnvironmentArn = this.getResourceArnAttribute(resource.attrComputeEnvironmentArn, {
      service: 'batch',
      resource: 'compute-environment',
      resourceName: this.physicalName,
    });

    this.node.addValidation({ validate: () => validateInstances(this.instanceTypes, this.instanceClasses, props.useOptimalInstanceClasses) });
  }

  public addInstanceType(instanceType: ec2.InstanceType): void {
    this.instanceTypes.push(instanceType);
  }

  public addInstanceClass(instanceClass: ec2.InstanceClass): void {
    this.instanceClasses.push(instanceClass);
  }
}

/**
 * A ManagedComputeEnvironment that uses ECS orchestration on Fargate instances.
 */
export interface IFargateComputeEnvironment extends IManagedComputeEnvironment { }

/**
 * Props for a FargateComputeEnvironment
 */
export interface FargateComputeEnvironmentProps extends ManagedComputeEnvironmentProps { }

/**
 * A ManagedComputeEnvironment that uses ECS orchestration on Fargate instances.
 *
 * @resource AWS::Batch::ComputeEnvironment
 */
export class FargateComputeEnvironment extends ManagedComputeEnvironmentBase implements IFargateComputeEnvironment {
  /**
   * Reference an existing FargateComputeEnvironment by its arn
   */
  public static fromFargateComputeEnvironmentArn(scope: Construct, id: string, fargateComputeEnvironmentArn: string): IFargateComputeEnvironment {
    const stack = Stack.of(scope);
    const computeEnvironmentName = stack.splitArn(fargateComputeEnvironmentArn, ArnFormat.SLASH_RESOURCE_NAME).resourceName!;

    class Import extends Resource implements IFargateComputeEnvironment {
      public readonly computeEnvironmentArn = fargateComputeEnvironmentArn;
      public readonly computeEnvironmentName = computeEnvironmentName;
      public readonly enabled = true;
      public readonly maxvCpus = 1;
      public readonly connections = { } as any;
      public readonly securityGroups = [];
      public readonly tags: TagManager = new TagManager(TagType.MAP, 'AWS::Batch::ComputeEnvironment');
    }

    return new Import(scope, id);
  }

  public readonly computeEnvironmentName: string;
  public readonly computeEnvironmentArn: string;

  constructor(scope: Construct, id: string, props: FargateComputeEnvironmentProps) {
    super(scope, id, props);

    const { subnetIds } = props.vpc.selectSubnets(props.vpcSubnets);
    const resource = new CfnComputeEnvironment(this, 'Resource', {
      ...baseManagedResourceProperties(this, subnetIds),
      computeEnvironmentName: props.computeEnvironmentName,
      computeResources: {
        ...baseManagedResourceProperties(this, subnetIds).computeResources as CfnComputeEnvironment.ComputeResourcesProperty,
        type: this.spot ? 'FARGATE_SPOT' : 'FARGATE',
      },
    });
    this.computeEnvironmentName = this.getResourceNameAttribute(resource.ref);
    this.computeEnvironmentArn = this.getResourceArnAttribute(resource.attrComputeEnvironmentArn, {
      service: 'batch',
      resource: 'compute-environment',
      resourceName: this.physicalName,
    });
  }
}

function renderInstances(types?: ec2.InstanceType[], classes?: ec2.InstanceClass[], useOptimalInstanceClasses?: boolean): string[] {
  const instances = [];

  for (const instanceType of types ?? []) {
    instances.push(instanceType.toString());
  }
  for (const instanceClass of classes ?? []) {
    instances.push(instanceClass);
  }
  if (useOptimalInstanceClasses || useOptimalInstanceClasses === undefined) {
    instances.push('optimal');
  }

  return instances;
}

function createInstanceRoleAndProfile(scope: Construct, instanceRole?: iam.IRole) {
  const result: any = {};

  result.instanceRole = instanceRole ?? new iam.Role(scope, 'InstanceProfileRole', {
    assumedBy: new iam.ServicePrincipal('ec2.amazonaws.com'),
    managedPolicies: [iam.ManagedPolicy.fromAwsManagedPolicyName('service-role/AmazonEC2ContainerServiceforEC2Role')],
  });

  result.instanceProfile = new iam.CfnInstanceProfile(scope, 'InstanceProfile', {
    roles: [result.instanceRole.roleName],
  });

  return result;
}

function createSpotFleetRole(scope: Construct): IRole {
  return new iam.Role(scope, 'SpotFleetRole', {
    assumedBy: new iam.ServicePrincipal('spotfleet.amazonaws.com'),
  });
}

function determineAllocationStrategy(id: string, allocationStrategy?: AllocationStrategy, spot?: boolean): AllocationStrategy | undefined {
  let result = allocationStrategy;
  if (!allocationStrategy) {
    result = spot ? AllocationStrategy.SPOT_PRICE_CAPACITY_OPTIMIZED : AllocationStrategy.BEST_FIT_PROGRESSIVE;
  } else if (allocationStrategy === AllocationStrategy.SPOT_PRICE_CAPACITY_OPTIMIZED && !spot) {
    throw new Error(`Managed ComputeEnvironment '${id}' specifies 'AllocationStrategy.SPOT_PRICE_CAPACITY_OPTIMIZED' without using spot instances`);
  } else if (allocationStrategy === AllocationStrategy.SPOT_CAPACITY_OPTIMIZED && !spot) {
    throw new Error(`Managed ComputeEnvironment '${id}' specifies 'AllocationStrategy.SPOT_CAPACITY_OPTIMIZED' without using spot instances`);
  }

  return result;
}

function validateInstances(types?: ec2.InstanceType[], classes?: ec2.InstanceClass[], useOptimalInstanceClasses?: boolean): string[] {
  if (renderInstances(types, classes, useOptimalInstanceClasses).length === 0) {
    return ["Specifies 'useOptimalInstanceClasses: false' without specifying any instance types or classes"];
  }

  return [];
}

function validateSpotConfig(id: string, spot?: boolean, spotBidPercentage?: number, spotFleetRole?: iam.IRole): void {
  if (spotBidPercentage) {
    if (!spot) {
      throw new Error(`Managed ComputeEnvironment '${id}' specifies 'spotBidPercentage' without specifying 'spot'`);
    } else if (spotBidPercentage > 100) {
      throw new Error(`Managed ComputeEnvironment '${id}' specifies 'spotBidPercentage' > 100`);
    } else if (spotBidPercentage < 0) {
      throw new Error(`Managed ComputeEnvironment '${id}' specifies 'spotBidPercentage' < 0`);
    }
  }

  if (spotFleetRole) {
    if (!spot) {
      throw new Error(`Managed ComputeEnvironment '${id}' specifies 'spotFleetRole' without specifying 'spot'`);
    }
  }
}

function validateVCpus(id: string, minvCpus: number, maxvCpus: number): void {
  if (minvCpus < 0) {
    throw new Error(`Managed ComputeEnvironment '${id}' has 'minvCpus' = ${minvCpus} < 0; 'minvCpus' cannot be less than zero`);
  }
  if (minvCpus > maxvCpus) {
    throw new Error(`Managed ComputeEnvironment '${id}' has 'minvCpus' = ${minvCpus} > 'maxvCpus' = ${maxvCpus}; 'minvCpus' cannot be greater than 'maxvCpus'`);
  }
}

function baseManagedResourceProperties(baseComputeEnvironment: ManagedComputeEnvironmentBase, subnetIds: string[]) {
  return {
    serviceRole: baseComputeEnvironment.serviceRole?.roleArn,
    state: baseComputeEnvironment.enabled ? 'ENABLED' : 'DISABLED',
    computeResources: {
      maxvCpus: baseComputeEnvironment.maxvCpus,
      type: 'managed',
      updateToLatestImageVersion: baseComputeEnvironment.updateToLatestImageVersion,
      securityGroupIds: baseComputeEnvironment.securityGroups.map((securityGroup) => securityGroup.securityGroupId),
      subnets: subnetIds,
    },
    updatePolicy: {
      terminateJobsOnUpdate: baseComputeEnvironment.terminateOnUpdate,
      jobExecutionTimeoutMinutes: baseComputeEnvironment.updateTimeout?.toMinutes(),
    },
    replaceComputeEnvironment: baseComputeEnvironment.replaceComputeEnvironment,
    type: 'managed',
  };
}

const DEFAULT_MIN_VCPUS = 0;
const DEFAULT_MAX_VCPUS = 256;