scheduling imp is just a combination of spark dirver's taskScheduler with FIFO and Fairness policy between pools and users and concurrent jobs, plus mesos DRF policy. And resource management is a slightly diff from YANR, but same at agent-executor style, providing more distributed system non-functionality like HA and reliability. use k8s declarive API for job interaction outside.
@DragonKlavier
4 жыл бұрын
Hello. Thank you very much for your contribution. It is very interesting. Is there a document where the Tangram scheduler is explained? I only found the slides.
Пікірлер: 2