Running Multiple Schedulers in Kubernetes
summary
Huawei PaaS implements a multi-scheduler framework that adds an API-server conflict resolver and a scheduler-controller to pick instances by type and do instance-level load balancing. It replaces Kubernetes’ late 2–10s conflict detection with early conflict detection using binding requests and three criteria (strong/weak/boolean) plus priority and batch/group commit policies. The system supports per-pod scheduling policies, reduces reschedule latency for short jobs, integrates several schedulers (ported to Go), and is mostly implemented with some components still stabilizing for open source.
AI-written summary · about
chapters (11)
- 0:00Start
- 2:49Default Scheduling in Kubernetes
- 11:03Why Multiple Schedulers
- 24:52HOSS: Job Dispatch
- 26:52HOSS: Conflict Detection
- 28:24HOSS: Multi-level Conflict Criteria
- 31:44HOSS: Conflict Resolution Policies
- 34:30HOSS: All Schedulers
- 36:12Comparison: Multi-Schedulers in Mesos
- 39:59Summary
- 41:13References
related talks
34mKubernetes Scheduling Features or How Can I Make the System Do What I Want?
35mSIG Scheduling Deep Dive
42mDeep Dive Into the Latest Kubernetes Scheduler Features
10mFair Scheduling for Deep Learning Workloads in Kubernetes
25mKube-scheduler-evaluator: Evaluating Kubernetes Schedulers at Hyperscaler…
23mPractice of Fine-grained Cgroups Resources Scheduling in Kubernetes
30mSIG Scheduling Intro & Updates
24mKubernetes Batch Processing at Scale - A Scheduling Perspective