Volcano scheduler is the component responsible for Pod scheduling. It consists of a series of actions and plugins. Actions define what action should be executed in every step. Plugins provide the action algorithm detail in different scenes. Volcano scheduler is highly scalable. You can specify and implement actions and plugins according the needs.


Volcano scheduler workflow

Volcano scheduler works as follows:

  1. Jobs submitted by client are watched by scheduler and cached
  2. Open sessions periodically and a scheduling cycle begins
  3. Send jobs not scheduled in cache to to-be-scheduled-queue in session
  4. Traverse all jobs to be scheduled. Execute enqueue / allocate / preempt / reclaim / backfill actions in order they are defined and find the most suitable node for each job. Bind the job to the node. The specific algorithm logic executed in action depends on the implementation of each function in the registered plugins.
  5. Close this session



Enqueue action is responsible for filtering out the tasks to be scheduled that meet the requirements through a series of filtering algorithms and sending them to the queue. After the action, the status of the task changes from pending to inqueue.


Allocate action is responsible for selecting the most suitable node throughout a series of predication and optimization algorithm.


Preempt action is responsible for preemptive scheduling of high priority tasks in the same queue according to priority rules.


Reclaim action is responsible for reclaiming the resources due to the cluster based on the queue weight when a new task enters the queue and the cluster resources cannot meet the needs of the current queue.


Backfill action is responsible for backfilling the tasks in the pending state into the cluster node to maximize the resource utilization of the node.



The gang plugin considers tasks that are not in ready state(including Binding / Bound / Running / Allocated / Succeed / Pipelined) to have a higher priority. It will decide whether to schedule the task by checking if the resources due to the queue can meet the resources required by the task to run minavailable pods after trying to reclaim resources.


The conformance plugin considers the tasks in namespace kube-system have higher priority. These tasks will not be preempted.


The DRF plugin considers that tasks with fewer resources have higher priority. It tries to calculate the total amount of resources allocated by the preemptor and the preempted, and triggers the preemption when the preemptor has less resources.


The Nodeorder plugin returns the scores of all nodes for a task after passing a series of dimension scoring algorithms. The node with the highest score is considered to be the most suitable node for the task.


The predictions plugin determines whether a task is bound to a node through a series of dimensional evaluation algorithms.


The priority plugin is used to compare the priority of two jobs / tasks. For two jobs, it decides whose priority is higher by comparing job.spec.priorityClassName. For two tasks, it decides whose priority is higher by comparing task.priorityClassName / task.createTime / in order.


Volcano scheduler is highly scalable because of its composition pattern design. Users can decide which actions and plugins to use according to their personal needs, and they can also implement customization based on the interface Action or plugin. The scheduler configuration is located in the configmap named volcano-scheduler-configmap, which is mounted in the path /volcano.scheduler in the scheduler container as volume.

How to get configuration of Volcano scheduler

  • get the configmap named volcano-scheduler-configmap
# kubectl get configmap -nvolcano-system
NAME                          DATA   AGE
volcano-scheduler-configmap   1      6d2h
  • view the data part detail of configmap
# kubectl get configmap volcano-scheduler-configmap -nvolcano-system -oyaml
apiVersion: v1
  volcano-scheduler.conf: |
    actions: "enqueue, allocate, backfill"
    - plugins:
      - name: priority
      - name: gang
      - name: conformance
    - plugins:
      - name: drf
      - name: predicates
      - name: proportion
      - name: nodeorder
      - name: binpack
kind: ConfigMap
  annotations: |
      {"apiVersion":"v1","data":{"volcano-scheduler.conf":"actions: \"enqueue, allocate, backfill\"\ntiers:\n- plugins:\n  - name: priority\n  - name: gang\n  - name: conformance\n- plugins:\n  - name: drf\n  - name: predicates\n  - name: proportion\n  - name: nodeorder\n  - name: binpack\n"},"kind":"ConfigMap","metadata":{"annotations":{},"name":"volcano-scheduler-configmap","namespace":"volcano-system"}}
  creationTimestamp: "2020-08-15T04:01:02Z"
  name: volcano-scheduler-configmap
  namespace: volcano-system
  resourceVersion: "266"
  selfLink: /api/v1/namespaces/volcano-system/configmaps/volcano-scheduler-configmap
  uid: 1effe4d6-126c-42d6-a3a4-b811075c30f5

It includes actions and tiers in volcano-scheduler.conf. In actions, the comma is used as a separator to configure the actions to be executed by the scheduler. It should be noted that the order of configuration is the order of the scheduler’s execution, and Volcano itself will not check the rationality of the order. The list of plugins configured in tiers is the plugins registered with the scheduler. The specific algorithm implementation defined in plugins will be called in actions.