)]}'
{"specs/newton/approved/config-guest-nic-pci-address.rst":[{"author":{"_account_id":1779,"name":"Daniel Berrange","email":"berrange@redhat.com","username":"berrange"},"change_message_id":"63b1f21e5e9e4e051ff28b9209431ecfe0c6730f","unresolved":false,"context_lines":[{"line_number":44,"context_line":"NFV type of applications."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Nova currently doesn\u0027t have the ability to assign the PCI addresses of the"},{"line_number":47,"context_line":"virtual NIC devices.  If not specified, libvirt is responsible to allocate and"},{"line_number":48,"context_line":"assign PCI addresses for the devices of a guest (see [0]).  It\u0027s also possible"},{"line_number":49,"context_line":"with libvirt to use the same XML interface to specify the PCI addresses.  The"},{"line_number":50,"context_line":"target PCI address can found in the libvirt XML description of the instance."}],"source_content_type":"text/x-rst","patch_set":1,"id":"5a18252c_79cfa3b0","line":47,"updated":"2016-04-12 13:18:17.000000000","message":"This lack of PCI address assignment is entirely intentional by design. PCI address assignment is a very difficult thing to get right across different hypervisors, different architectures, different machine types and different software versions. The cloud users do not have anywhere near enough knowledge about the underlying hypervisor to be able to do it correctly. As such we will *not* add PCI address assignment in the Nova APis.\n\nThere is an existing approved spec which will allow Nova to expose information to the guest to let it identify devices wrt intended application usage roles:\n\n  http://specs.openstack.org/openstack/nova-specs/specs/newton/approved/virt-device-role-tagging.html","commit_id":"2f18c3df569b55a871b8c9b1909aabcc6d4fea73"},{"author":{"_account_id":17973,"name":"Ludovic Beliveau","email":"lbeliveau@gmail.com","username":"lbeliveau"},"change_message_id":"a8cae7b32a4f200b814cc3c3b2af6e93aceaf39f","unresolved":false,"context_lines":[{"line_number":44,"context_line":"NFV type of applications."},{"line_number":45,"context_line":""},{"line_number":46,"context_line":"Nova currently doesn\u0027t have the ability to assign the PCI addresses of the"},{"line_number":47,"context_line":"virtual NIC devices.  If not specified, libvirt is responsible to allocate and"},{"line_number":48,"context_line":"assign PCI addresses for the devices of a guest (see [0]).  It\u0027s also possible"},{"line_number":49,"context_line":"with libvirt to use the same XML interface to specify the PCI addresses.  The"},{"line_number":50,"context_line":"target PCI address can found in the libvirt XML description of the instance."}],"source_content_type":"text/x-rst","patch_set":1,"id":"1a122d0e_841b0f1c","line":47,"in_reply_to":"5a18252c_79cfa3b0","updated":"2016-04-22 17:50:37.000000000","message":"Daniel, based on the virt-device-role-tagging, how do you correlate the actual device created by libvirt/qemu back to the tag ?\n\nLet\u0027s say nova libvirt driver specifies two network interfaces for a given vm, it\u0027s going to specify them in the instance xml.  Then the request will be made to libvirt to launch the vm.\n\nOnce booted, the nova libvirt driver will then have access to the \"final\" instance xml representing the actual vm that is running.  This \"final\" xml document then contains the PCI addresses (in the case that interests us) for each devices that were requested.  The idea I guess is to use that to build the metadata document.\n\nIn this example with multiple devices, how do you correlate back the devices to the original tag specified in the boot command ?\n\nThere is the metadata node, but I\u0027m not sure if you can use it at any level of the xml document (I think it\u0027s only in the domain - you probably know better than me).\n\nOr tampering with the device name and embed the tag in there ?","commit_id":"2f18c3df569b55a871b8c9b1909aabcc6d4fea73"}]}