openstack
/
nova
Code Issues Proposed changes

Validate pci_passthrough_whitelist when starting n-cpu 

Loading up CONF.pci_passthrough_whitelist in the Whitelist
object performs a bunch of validation and can fail in several
different ways (invalid json, invalid values, invalid combinations
of keys, devices not found, etc). This happens today when
creating the PciDevTracker in the ResourceTracker when updating
available resources. If the configuration is bad, it kills the
periodic task to update available resources on the compute node.

We should just load up the pci_passthrough_whitelist (if set)
when starting the nova-compute service so we can fail fast and
kill the service on any misconfiguration rather than run with
a broken service.

Change-Id: If50fb837b490042bb5ef20e9ad843b28f871a44e
Closes-Bug: #1603034
This commit is contained in:
Matt Riedemann 2016-07-14 13:37:05 -04:00
parent f6f4003dfd
commit 3a61ae35d4
2 changed files with 21 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
nova/compute/manager.py
View File

@ -88,6 +88,7 @@ from nova.objects import base as obj_base
from nova.objects import fields
from nova.objects import instance as obj_instance
from nova.objects import migrate_data as migrate_data_obj
from nova.pci import whitelist
from nova import rpc
from nova import safe_utils
from nova.scheduler import client as scheduler_client
@ -1119,6 +1120,17 @@ class ComputeManager(manager.Manager):
def init_host(self):
"""Initialization for a standalone compute service."""
if CONF.pci_passthrough_whitelist:
# Simply loading the PCI passthrough whitelist will do a bunch of
# validation that would otherwise wait until the PciDevTracker is
# constructed when updating available resources for the compute
# node(s) in the resource tracker, effectively killing that task.
# So load up the whitelist when starting the compute service to
# flush any invalid configuration early so we can kill the service
# if the configuration is wrong.
whitelist.Whitelist(CONF.pci_passthrough_whitelist)
self.driver.init_host(host=self.host)
context = nova.context.get_admin_context()
instances = objects.InstanceList.get_by_host(

9
nova/tests/unit/compute/test_compute_mgr.py
View File

@ -2764,6 +2764,15 @@ class ComputeManagerUnitTestCase(test.NoDBTestCase):
msg = mock_log.warning.call_args_list[0]
self.assertIn('appears to not be owned by this host', msg[0][0])
def test_init_host_pci_passthrough_whitelist_validation_failure(self):
# Tests that we fail init_host if there is a pci_passthrough_whitelist
# configured incorrectly.
self.flags(pci_passthrough_whitelist=[
# it's invalid to specify both in the same devspec
jsonutils.dumps({'address': 'foo', 'devname': 'bar'})])
self.assertRaises(exception.PciDeviceInvalidDeviceName,
self.compute.init_host)
@mock.patch('nova.compute.manager.ComputeManager._instance_update')
def test_error_out_instance_on_exception_not_implemented_err(self,
inst_update_mock):