placement response timeout when create a vm with passthrough of 7GPUs

[zack chen]

When attempting to create a virtual machine with direct passthrough of 7 GPUs, nova-scheduler reports a placement response timeout. The placement logs do not indicate any errors but continuously print logs,
I enabled debug logging and executed a curl command that fetched over 200MB of log data.

The curl command is：

curl -g -i --insecure -X GET "https://192.168.1.100:8780//allocation_candidates?group_policy=none&limit=1000&member_of=in%3Aa58d4d6e-dd6f-44a0-9d84-c0da4edd28d9&resources=MEMORY_MB%3A32768%2CVCPU%3A16&resourcesdevice_profile_0=PGPU%3A1&resourcesdevice_profile_1=PGPU%3A1&resourcesdevice_profile_2=PGPU%3A1&resourcesdevice_profile_3=PGPU%3A1&resourcesdevice_profile_4=PGPU%3A1&resourcesdevice_profile_5=PGPU%3A1&resourcesdevice_profile_6=PGPU%3A1&root_required=COMPUTE_ACCELERATORS%2C%21COMPUTE_STATUS_DISABLED" -H "Accept: application/json" -H "OpenStack-API-Version: placement 1.36" -H "User-Agent: openstacksdk/0.101.0 keystoneauth1/5.0.0 python-requests/2.28.2 CPython/3.10.12" -H "X-Auth-Token: token"

The key log entry in placement-api.log is:

placement-api.log:498072:2024-06-22 20:26:19.825 738 DEBUG placement.objects.allocation_candidate [None req-08ad26e6-569d-4edd-bd70-56b579c17d86 1fc2e622be4d4d10ba4d9c9e1665abde 427edb8e80594a39ba95f18af64e6c42 - - default default] Merging candidates yields 20160 allocation requests and 10 provider summaries _merge_candidates /var/lib/kolla/venv/lib/python3.10/site-packages/placement/objects/allocation_candidate.py:809

Does this mean that looping 20160 times caused the timeout?