e1beceb0创建于 2023年3月10日历史提交

Volcano Roadmap

v1.0(Released on July 8, 2020)

The major target of this release to make Volcano more stable for product.

Stability and Resilience

Improve test coverage. In v1.0, more test cases will be added to improve Volcano stability for product.

Preemption/Reclaim Enhancement

Preemption and Reclaim are two import features for resource sharing; there're two actions for now, but unstable. In v1.0, those two features are going to be enhanced for elastic workload, e.g. stream job, bigdata batch job.

GPU Share (#624)

A better performance has its cost, including GPU; and there are several scenarios that a Pod can not consume one GPU, e.g. inference workload, dev environment. One of solutions is to support GPU share, including related enhancement to both scheduler and kubelet.

Flink is a widely used for Stateful Computations over Data Streams, but flink on kubernetes has some gaps now.

Integrate with argo to support job dependencies

Investigate to cooperate with argo to support job dependencies.

Support running MindSpore jobs

MindSpore is a deep learning training and inference framework, support running MindSpore training with volcano job.

v1.2(Released on Feb 27, 2021)

Queue Resource Reservation(Delay)

Fair Scheduling For Jobs Of Same Priority And Different Queue

Differentiated Scheduling Strategies For Different Queue

Support Hierarchy Queue(Delay)

Job PriorityClassName Update

Status Message Enhanced For CRD(Delay)

Support MinAvailable For Task

v1.3(Released on May 27, 2021)

Task-Topology

Support multiple scheduler

Stability and Resilience

Volcano Device Plugin enhancement

v1.4(Released on Sep 18, 2021)

Support NUMA-Awareness scheduling in Volcano

Support multi-scheduler by admission controller

Support scheduling with proportion of resources

Enhance the resource comparison functions for various of scenarios

System Stability Enhancement

v1.5(Released on Feb 20, 2022)

Support Hierarchy Queue(Delay)

Support Volcano scheduler in Spark community(Delay)

Monitoring: Cluster Resource(Delay)

Task Dag scheduling

Support configuration hot update(Delay)

v1.6(To Be Released around May 15, 2022)

Support Dynamic Scheduling Based on Realtime Load

Support Rescheduling Based on Realtime Load

Support Elastic Scheduling

  • Description: Support elastic scheduling for workloads.
  • Priority: High
  • Issue: TO BE ADDED
  • Owner: @qiankunli @Thor-wl

Later (To be updated)

Monitoring:

Improve resource calculation accuracy

Support job backfill

  • Description: Add backfill functionality to improve the resource utilization.

Improve the Autoscaling efficiency

  • Description: Combine the Autoscaler and scheduler to improve the scaling efficiency.