vocabulary:
  domain: AWS Batch
  description: >-
    Taxonomy and vocabulary for AWS Batch covering batch computing concepts,
    resource types, job scheduling, and HPC workflows.

  concepts:
    - term: Compute Environment
      definition: >-
        A managed or unmanaged collection of compute resources (EC2, Fargate,
        or EKS) that AWS Batch uses to run containerized batch jobs.

    - term: Managed Compute Environment
      definition: >-
        A compute environment where AWS Batch automatically provisions, scales,
        and terminates EC2 or Fargate instances based on job demand.

    - term: Unmanaged Compute Environment
      definition: >-
        A compute environment where the customer manages their own EC2
        instances registered with an ECS cluster.

    - term: Job Queue
      definition: >-
        A queue that holds submitted jobs and routes them to compute environments
        based on priority ordering and scheduling policies.

    - term: Job Definition
      definition: >-
        A versioned template that specifies the container image, resource
        requirements (vCPUs, memory), command, retry strategy, and timeout
        for a batch job type.

    - term: Job
      definition: >-
        A unit of work submitted to AWS Batch. Can be a single-node job,
        an array job (multiple parallel instances), or a multi-node parallel
        job for MPI-style HPC workloads.

    - term: Array Job
      definition: >-
        A batch job that spawns N identical child jobs, each with a unique
        array index. Used for parameter sweeps, data parallelism, and
        large-scale simulations.

    - term: Multi-Node Parallel Job
      definition: >-
        A batch job that runs across multiple EC2 instances for tightly-coupled
        HPC workloads like MPI applications.

    - term: Scheduling Policy
      definition: >-
        A fair-share scheduling policy that distributes compute resources
        equitably across multiple users or job categories to prevent starvation.

    - term: Spot Instance
      definition: >-
        AWS EC2 Spot Instances used in AWS Batch compute environments to
        run batch jobs at a significantly reduced cost compared to On-Demand.

    - term: SPOT_CAPACITY_OPTIMIZED
      definition: >-
        An allocation strategy for Spot Instances that selects instance pools
        with the most available capacity to reduce interruption likelihood.

    - term: vCPU
      definition: >-
        Virtual CPU — the unit of compute capacity requested by batch job
        definitions and allocated in compute environments.

    - term: Retry Strategy
      definition: >-
        Configuration that specifies the number of times to retry a failed
        job attempt, with optional conditional exit code evaluation.

    - term: Job Status
      definition: >-
        The lifecycle state of a batch job. States: SUBMITTED, PENDING,
        RUNNABLE, STARTING, RUNNING, SUCCEEDED, FAILED.

    - term: Fair-Share Scheduling
      definition: >-
        A scheduling approach that ensures equitable distribution of compute
        resources based on share weights and historical usage.

  tags:
    - Batch Computing
    - HPC
    - High Performance Computing
    - Containers
    - EC2
    - Fargate
    - EKS
    - Job Scheduling
    - Data Processing
    - Scientific Computing
    - AWS
    - Amazon Web Services
    - Spot Instances
    - Serverless