Koordinator: QoS-Based Scheduling and Colocation for Kubernetes
Shriira Press
Run latency-sensitive services and batch jobs on the same nodes without one starving the other — QoS-based colocation, dynamic overcommitment, and smart scheduling for Kubernetes.
Welcome to Koordinator: QoS-Based Scheduling and Colocation for Kubernetes.
Koordinator is a CNCF sandbox scheduling system that lets microservices, web services, AI training, and big data jobs share the same Kubernetes nodes safely, raising cluster utilization without sacrificing the latency of critical workloads. It sits alongside vanilla Kubernetes — no forked kubelet, no patched API server — and adds the machinery that classic Kubernetes lacks: a two-dimensional priority-and-QoS model, dynamic resource overcommitment that reclaims allocated-but-idle capacity for batch pods, fine-grained CPU and NUMA orchestration, load-aware scheduling and descheduling, gang scheduling for all-or-nothing jobs, and multi-tenant elastic quotas. This book builds the picture from the ground up: the utilization problem and what colocation is, the priority and QoS classes that define how a pod is treated, the architecture of koord-manager, koord-scheduler, koord-descheduler, and the koordlet node agent, then dynamic overcommitment, batch colocation, CPU orchestration, load-aware placement, job scheduling with quotas, and finally putting it all together in practice. Nine focused chapters with clear diagrams that make a notoriously subtle topic concrete.
This title is part of the ShriIra library and is free to read in full, right here — our small contribution to making world-class knowledge easy to reach.
A note on reading it: open the Contents menu at the top of the reader to jump between chapters, use the Aa menu to set a comfortable text size, theme (light, sepia, or night), and single- or two-page layout. Your place is saved automatically, so you can always pick up where you left off.
We hope it serves you well.
— Shriira Press