Attempts to use same node multiple times when many nodes created simultaneously (race condition?)

# Problem description

When instantiating a modest number (ten or so) Ironic nodes simultaneously via Nova using Terraform, I find that sometimes a few instances will immediately enter the Error state.

The final exception error message is "Failure prepping block device." which I suspect is a red herring since further up the exception trace is the following:

    nova.exception.InstanceDeployFailure: Failed to reserve
    node 228437bc-b091-4a8a-8b3d-7134214892e6 when provisioning
    the instance 168c5e97-bff2-4408-b12a-8158414fc5de

Here node 228437bc[...] is a (previously available) Ironic node which, upon inspection, is now running one of the other instances created in the batch. Instance 168c5e97[...] is the instance reporting this error.

Full stack trace here: https://paste.openstack.org/show/bSxVzPIVahJKYA5zGumX/

Also found in a Nova log file is:

2023-06-22 08:46:39.796 217494 WARNING nova.scheduler.client.report [None req-8e6b5289-cd5c-4499-bad5-e6b576a70e88 ce20f804a02fc250ec44394149d69a682aed245fd50194cd917fd238c2dc52c1 ffaaa801d7ac4b7e94945aa8b4b9d245 - - default default] Failed to save allocation for d43c698f-5ee7-445f-9653-a8d3a24e71d1. Got HTTP 409: {"errors": [{"status": 409, "title": "Conflict", "detail": "There was a conflict when trying to complete your request.\n\n Unable to allocate inventory: Unable to create allocation for 'CUSTOM_R730' on resource provider '228437bc-b091-4a8a-8b3d-7134214892e6'. The requested amount would exceed the capacity. ", "code": "placement.undefined_code", "request_id": "req-6bd98596-0a8a-48af-b1af-6c884c81085a"}]}

My interpretation of this situation is that there might be some kind of race condition which is causing a single Ironic node to be offered to two instances created in quick succession and obviously only the first one wins.

# Reproduction steps

1. Setup an Ironic system with a modest number of nodes (15 in my case).
2. Simultaneously create several instances (again 15 in my case -- one
   per available Ironic node). I used Terraform which internally calls
   the Nova API. Unfortunately it would not be helpful to share my
   Terraform script since it uses various internal Terraform modules.
3. Something like 10% of instances will immediately enter the Error
   state with an exception as described above. This appears to be a
   matter of random chance.

# Software Versions

Ironic: 06895641fb8a44caf4574919bd518f0de76cba3d
        with the following (unrelated) patches applied:
        Ic9dea4f51d82866be8ac16242a79237c789b9745
        I2a6bca3550819b98adbaffe315f77427b8a43d62

Nova: c9de185ea1ac1e8d4435c5863b2ad7cefdb28c76
      with the following (unrelated) patches applied:
      Ibd68fb72957ca850f3be4e7b4ea68af038fad07c
      Ifed0fa16053228990a6a8df8d4c666521db7e329
      I5a399f1d3d702bfb76c067893e9c924904c8c360