Monday, 2026-06-22

ssergeyHello! I have a question regarding PCI tracking and scheduling in Placement that was introduced in Antelope. I've enabled pci.report_in_placement for all compute nodes with attached GPUs and then filter_scheduler.pci_in_placement for both nova-scheduler, nova-conductor and nova-api.15:02
ssergeyNow my Placement shows a tree of compute node resource providers with several children resource providers that looks like computeX-0000:00:00.0, computeX-0000:01:00.0 and so on.15:04
ssergeyIs there any way to have a flat structure (like root resource providers nodeX with a single resource class let's say PGPU with multiple available devices) instead of multiple child resource providers that have a single device each? Also how can I exclude compute nodes with disabled nova-compute from Placement output?15:07
ssergeyNow if I use allocation candidate list command it still shows devices that belong to those temporary disabled compute nodes15:08

Generated by irclog2html.py 4.1.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!