I have PCI passthrough enabled for some of my hypervisior hosts. Initial
scheduling for flavors that have the passthrough assigned works properly, i.e.
instances are assigned to hypervisors with the resources and once the available
pool of resources is consumed creation of new instances fails. When instances
are shelved, I'm able to create new instances with the pci passthrough flavor
and all is good.
The problem that I'm running into is that when I try to unshelve an instance
that uses the pci passthrough flavor, the wrong hypervisor, i.e. one with no
available pci passthrough resources is selected.
For example, I have two hypervisors: one with no GPU cards and the other with 4
GeForce 1080 GPU cards. I have a flavor that allocates one GPU per instance
and when I try to unshelve an instance, it gets placed on the hypervisor
without any GPUs:
2017-12-01 13:55:52.843 2450 WARNING nova.scheduler.host_manager
[req-b7876409-1149-4f8a-837d-46f0023fed10 6136e80cc67e4759843a5f95f7a9a98b
7618d5a893f04349b9feb2ca541a6fce - default default] Selected host:
overcloud-novacompute-0.localdomain failed to consume from instance. Error: PCI
device request
[InstancePCIRequest(alias_name='GeForce1080',count=1,is_new=<?>,request_id=<?>,spec=[{dev_type='type-PCI',product_id='1b06',vendor_id='10de'}])]
failed: PciDeviceRequestFailed: PCI device request
[InstancePCIRequest(alias_name='GeForce1080',count=1,is_new=<?>,request_id=<?>,spec=[{dev_type='type-PCI',product_id='1b06',vendor_id='10de'}])]
failed
Is there a setting that I'm missing to use PciPassthroughFilter when unshelving
instances?
[root@overcloud-controller-0 nova]# nova
--version9.1.1[root@overcloud-controller-0 nova]#
Thanks in advance for any assistance.
Bryan
_______________________________________________
Mailing list: http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
Post to : openstack@lists.openstack.org
Unsubscribe : http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
