vocabulary:
  name: Volcano Vocabulary
  description: >-
    Domain vocabulary for Volcano — the CNCF incubating batch scheduling and
    high-performance computing scheduler for Kubernetes. Covers scheduling
    concepts, CRD resources, and workload management terminology.
  version: '2026-05-03'
  terms:
    - term: Gang Scheduling
      definition: >-
        An all-or-nothing scheduling strategy where all pods in a group must
        be schedulable simultaneously before any are placed on nodes. Prevents
        partial deployments that would deadlock distributed jobs waiting for
        resources that will never arrive.
      tags:
        - Scheduling
        - Core Concept

    - term: Volcano Job (vcjob)
      definition: >-
        The primary Volcano CRD (Custom Resource Definition) representing a
        batch workload. A Job contains one or more task groups, lifecycle policies,
        queue assignment, and framework plugins. Extends Kubernetes Jobs with
        multi-task and gang scheduling support.
      tags:
        - CRD
        - Core Concept

    - term: Queue
      definition: >-
        A cluster-scoped Volcano CRD representing a scheduling partition.
        Queues have a weight for fair-share scheduling, optional resource
        capacity limits, and can be opened, closed, or placed in preemptable
        state. Jobs are submitted to a named queue.
      tags:
        - CRD
        - Resource Management

    - term: PodGroup
      definition: >-
        A namespaced Volcano CRD defining a set of pods that must be scheduled
        as a group. PodGroups are the atomic unit of gang scheduling — if
        minMember pods cannot be placed, none are placed.
      tags:
        - CRD
        - Gang Scheduling

    - term: Fair Share Scheduling
      definition: >-
        A scheduling algorithm where available cluster resources are distributed
        proportionally among queues based on their weight, preventing any single
        queue from monopolizing the cluster.
      tags:
        - Scheduling

    - term: MinAvailable
      definition: >-
        A Job or PodGroup field specifying the minimum number of pods that must
        be simultaneously schedulable. This is the threshold for gang scheduling.
      tags:
        - Configuration

    - term: Lifecycle Policy
      definition: >-
        A rule mapping job or pod events (PodFailed, PodEvicted, TaskCompleted)
        to actions (AbortJob, RestartJob, CompleteJob). Policies control how
        Volcano responds to runtime events in distributed workloads.
      tags:
        - Configuration

    - term: Plugin
      definition: >-
        A Volcano mechanism for framework-specific workload initialization.
        Plugins like pytorch, tensorflow, spark, and mpi inject environment
        variables (ranks, master addresses) and init containers into task pods.
      tags:
        - Framework Integration

    - term: CNCF Incubating
      definition: >-
        Volcano's graduation status in the Cloud Native Computing Foundation
        (CNCF) project lifecycle, indicating active development and production
        use but not yet graduated status.
      tags:
        - Governance

    - term: HPC
      definition: >-
        High-Performance Computing — computationally intensive workloads
        such as scientific simulations, molecular dynamics, and numerical analysis.
        One of Volcano's primary target use cases.
      tags:
        - Use Case

    - term: Task
      definition: >-
        A named group of pod replicas within a Volcano Job, each with an
        identical pod template. Different tasks in a job can represent different
        roles (master vs. worker in distributed training).
      tags:
        - Job Structure

    - term: Reclaimable Queue
      definition: >-
        A Queue configured to allow its allocated resources to be reclaimed by
        other queues when idle, enabling efficient cluster utilization.
      tags:
        - Resource Management

    - term: NUMA Topology Policy
      definition: >-
        A task-level configuration controlling how CPU and memory are allocated
        relative to NUMA (Non-Uniform Memory Access) nodes, important for
        latency-sensitive HPC workloads.
      tags:
        - HPC
        - Configuration