Kueue, placement on adjacent machines
author of the fork, 2026
This fork adds an ordered topology level to the Kubernetes job queue, assigning jobs to contiguous runs of neighboring machines to take advantage of faster cabling between adjacent hardware.
Kueue queues computing jobs for Kubernetes, and on hardware where cabling lets neighboring machines transfer data faster than distant ones, a job runs best across a continuous series of servers. This fork adds an ordered topology level that places workloads on adjacent machines and compacts the queue within a set budget, evicting and repacking tasks to defragment the chain while prioritizing machines that already hold warm caches.