NUMATopologyFilter will not choose numa node which has sibling but not request and other vcpus match request when only one node has sibling while host has two node when cpu_thread_policy is prefer.
Bug #1599417 reported by
liuxiuli
This bug report is a duplicate of:
Bug #1578155: 'hw:cpu_thread_policy=prefer' misbehaviour.
Edit
Remove
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
New
|
Undecided
|
liuxiuli |
Bug Description
version:master
problem:
NUMATopologyFilter will not choose numa node which has sibling but not satisfies request and other vcpus satisfy request when cpu_thread_policy is prefer.
for example:
host has cpu info:
NUMA node0 CPU(s): 0-7,16-23
NUMA node1 CPU(s): 8-15,24-31
and vcpu_pin_set is 0,16-23, then the node 0's sibling is [0,16], and nod 1's sibling is [].
When I boot an instance with "hw:cpu_policy": "dedicated", "hw:cpu_
Changed in nova: | |
assignee: | nobody → liuxiuli (liu-lixiu) |
tags: | added: numa |
tags: | added: scheduler |
To post a comment you must log in.
Not sure whether it is the designed behavior. The 'prefer' policy will try to place vCPU according to the sibling information because it is available. And that's why it is failed.