yahoo-eng-team team mailing list archive
-
yahoo-eng-team team
-
Mailing list archive
-
Message #95926
[Bug 2111894] [NEW] Evacuate fails to release PCI device mappings for GPU VMs
Public bug reported:
Description
===========
After evacuating a GPU-passthrough VM, the pci_devices table retains the PCI device mappings on the original host. This prevents the GPU from being reused, blocking new deployments and causing errors in subsequent operations.
Steps to reproduce
===================
- Deploy a GPU-passthrough VM.
- Evacuate the VM to another host.
- Check the pci_devices table ( SELECT * FROM pci_devices WHERE instance_uuid = '<VM_UUID>'; )
→ Entries remain on the source host with status=allocated.
- Try to launch a new VM using the same GPU → fails with "No valid PCI device available".
Environment
===========
- OpenStack Version: Caracal
- PCI Tracking: non-Placement (network required for SR-IOV).
References
===========
Related to Bug #2077009 but occurs in non-Placement PCI tracking scenarios.
** Affects: nova
Importance: Undecided
Status: New
--
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to OpenStack Compute (nova).
https://bugs.launchpad.net/bugs/2111894
Title:
Evacuate fails to release PCI device mappings for GPU VMs
Status in OpenStack Compute (nova):
New
Bug description:
Description
===========
After evacuating a GPU-passthrough VM, the pci_devices table retains the PCI device mappings on the original host. This prevents the GPU from being reused, blocking new deployments and causing errors in subsequent operations.
Steps to reproduce
===================
- Deploy a GPU-passthrough VM.
- Evacuate the VM to another host.
- Check the pci_devices table ( SELECT * FROM pci_devices WHERE instance_uuid = '<VM_UUID>'; )
→ Entries remain on the source host with status=allocated.
- Try to launch a new VM using the same GPU → fails with "No valid PCI device available".
Environment
===========
- OpenStack Version: Caracal
- PCI Tracking: non-Placement (network required for SR-IOV).
References
===========
Related to Bug #2077009 but occurs in non-Placement PCI tracking scenarios.
To manage notifications about this bug go to:
https://bugs.launchpad.net/nova/+bug/2111894/+subscriptions