)]}'
{"nova/compute/manager.py":[{"author":{"_account_id":9708,"name":"Balazs Gibizer","display_name":"gibi","email":"gibizer@gmail.com","username":"gibi"},"change_message_id":"a9273b305ebbc135e693c8e3aed93627d1acec40","unresolved":true,"context_lines":[{"line_number":3963,"context_line":"                self._notify_instance_rebuild_error(context, instance, e, bdms)"},{"line_number":3964,"context_line":"                raise"},{"line_number":3965,"context_line":"            else:"},{"line_number":3966,"context_line":"                # NOTE: During evacuate, explicitly clean up PCI devices on the"},{"line_number":3967,"context_line":"                # source host since they aren\u0027t automatically released. This is"},{"line_number":3968,"context_line":"                # critical in non-Placement environments where PCI tracking"},{"line_number":3969,"context_line":"                # relies solely on Nova\u0027s DB."}],"source_content_type":"text/x-python","patch_set":3,"id":"5d5df74d_46c90732","line":3966,"updated":"2025-05-28 12:27:39.000000000","message":"This should happen when the source compute is restarted after the outage triggered the evacuation. The dest computes should not interfere with the resource allocation of the source compute","commit_id":"381afadd8d9170181e1ee3a78d2f29028e37979e"},{"author":{"_account_id":37469,"name":"DongHun, Kim","display_name":"DongHun, Kim","email":"bigheadzacky@gmail.com","username":"donghunkimzack"},"change_message_id":"c1db45e347b26a2570b8d5203ce7c5350bf95de6","unresolved":true,"context_lines":[{"line_number":3963,"context_line":"                self._notify_instance_rebuild_error(context, instance, e, bdms)"},{"line_number":3964,"context_line":"                raise"},{"line_number":3965,"context_line":"            else:"},{"line_number":3966,"context_line":"                # NOTE: During evacuate, explicitly clean up PCI devices on the"},{"line_number":3967,"context_line":"                # source host since they aren\u0027t automatically released. This is"},{"line_number":3968,"context_line":"                # critical in non-Placement environments where PCI tracking"},{"line_number":3969,"context_line":"                # relies solely on Nova\u0027s DB."}],"source_content_type":"text/x-python","patch_set":3,"id":"89d13e37_ee6a02c0","line":3966,"in_reply_to":"5d5df74d_46c90732","updated":"2025-05-29 09:15:22.000000000","message":"During evacuation, PCI devices (e.g., GPUs) are not properly released on the source host, resulting in duplicate allocation records in the pci_devices table for both the source and destination hosts. When attempting to restart the VM, Nova detects conflicting PCI allocations, leading to startup failure.\n\n---- nova-compute log\nDEBUG nova.objects.instance ... Lazy-loading \u0027pci_devices\u0027 on Instance uuid\u003d\u003cvm-uuid\u003e  # PCI allocation check\nINFO nova.compute.manager ... Successfully reverted task state from powering-on on failure for instance.  # Rollback due to PCI conflict","commit_id":"381afadd8d9170181e1ee3a78d2f29028e37979e"}]}