|
| 1 | +# Licensed under the Apache License, Version 2.0 (the "License"); you may |
| 2 | +# not use this file except in compliance with the License. You may obtain |
| 3 | +# a copy of the License at |
| 4 | +# |
| 5 | +# http://www.apache.org/licenses/LICENSE-2.0 |
| 6 | +# |
| 7 | +# Unless required by applicable law or agreed to in writing, software |
| 8 | +# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT |
| 9 | +# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the |
| 10 | +# License for the specific language governing permissions and limitations |
| 11 | +# under the License. |
| 12 | + |
| 13 | +import time |
| 14 | + |
| 15 | +from nova import exception |
| 16 | +from nova.tests.functional import integrated_helpers |
| 17 | +from nova.tests.unit import fake_notifier |
| 18 | +from nova.tests.unit.image import fake as fake_image |
| 19 | + |
| 20 | + |
| 21 | +class BuildRescheduleClaimFailsTestCase( |
| 22 | + integrated_helpers.ProviderUsageBaseTestCase): |
| 23 | + """Regression test case for bug 1837955 where a server build fails on the |
| 24 | + primary host and then attempting to allocate resources on the alternate |
| 25 | + host, the alternate host is full and the allocations claim in placement |
| 26 | + fails, resulting in the build failing due to MaxRetriesExceeded and the |
| 27 | + server going to ERROR status. |
| 28 | + """ |
| 29 | + compute_driver = 'fake.SmallFakeDriver' |
| 30 | + |
| 31 | + def _wait_for_unversioned_notification(self, event_type): |
| 32 | + for x in range(20): # wait up to 10 seconds |
| 33 | + for notification in fake_notifier.NOTIFICATIONS: |
| 34 | + if notification.event_type == event_type: |
| 35 | + return notification |
| 36 | + time.sleep(.5) |
| 37 | + self.fail('Timed out waiting for unversioned notification %s. Got: %s' |
| 38 | + % (event_type, fake_notifier.NOTIFICATIONS)) |
| 39 | + |
| 40 | + def test_build_reschedule_alt_host_alloc_fails(self): |
| 41 | + # Start two compute services so we have one alternate host. |
| 42 | + # Set cpu_allocation_ratio=1.0 to make placement inventory |
| 43 | + # and allocations for VCPU easier to manage. |
| 44 | + self.flags(cpu_allocation_ratio=1.0) |
| 45 | + for x in range(2): |
| 46 | + self._start_compute('host%i' % x) |
| 47 | + |
| 48 | + def fake_instance_claim(_self, _context, _inst, nodename, *a, **kw): |
| 49 | + # Before triggering the reschedule to the other host, max out the |
| 50 | + # capacity on the alternate host. |
| 51 | + alt_nodename = 'host0' if nodename == 'host1' else 'host1' |
| 52 | + rp_uuid = self._get_provider_uuid_by_host(alt_nodename) |
| 53 | + inventories = self._get_provider_inventory(rp_uuid) |
| 54 | + # Fake some other consumer taking all of the VCPU on the alt host. |
| 55 | + # Since we set cpu_allocation_ratio=1.0 the total is the total |
| 56 | + # capacity for VCPU on the host. |
| 57 | + total_vcpu = inventories['VCPU']['total'] |
| 58 | + alt_consumer = '7d32d0bc-af16-44b2-8019-a24925d76152' |
| 59 | + allocs = { |
| 60 | + 'allocations': { |
| 61 | + rp_uuid: { |
| 62 | + 'resources': { |
| 63 | + 'VCPU': total_vcpu |
| 64 | + } |
| 65 | + } |
| 66 | + }, |
| 67 | + 'project_id': self.api.project_id, |
| 68 | + 'user_id': self.api.project_id |
| 69 | + } |
| 70 | + resp = self.placement_api.put( |
| 71 | + '/allocations/%s' % alt_consumer, allocs, version='1.12') |
| 72 | + self.assertEqual(204, resp.status, resp.content) |
| 73 | + raise exception.ComputeResourcesUnavailable(reason='overhead!') |
| 74 | + |
| 75 | + # Stub out the instance claim (regardless of which host the scheduler |
| 76 | + # picks as the primary) to trigger a reschedule. |
| 77 | + self.stub_out('nova.compute.manager.resource_tracker.ResourceTracker.' |
| 78 | + 'instance_claim', fake_instance_claim) |
| 79 | + |
| 80 | + # Now that our stub is in place, try to create a server and wait for it |
| 81 | + # to go to ERROR status. |
| 82 | + server = self._build_minimal_create_server_request( |
| 83 | + self.api, 'test_build_reschedule_alt_host_alloc_fails', |
| 84 | + image_uuid=fake_image.get_valid_image_id(), |
| 85 | + networks=[{'port': self.neutron.port_1['id']}]) |
| 86 | + server = self.api.post_server({'server': server}) |
| 87 | + # FIXME(mriedem): This is bug 1837955 where the status is stuck in |
| 88 | + # BUILD rather than the vm_state being set to error and the task_state |
| 89 | + # being set to None. Uncomment this when the bug is fixed. |
| 90 | + # server = self._wait_for_state_change(self.api, server, 'ERROR') |
| 91 | + |
| 92 | + # Wait for the MaxRetriesExceeded fault to be recorded. |
| 93 | + # set_vm_state_and_notify sets the vm_state to ERROR before the fault |
| 94 | + # is recorded but after the notification is sent. So wait for the |
| 95 | + # unversioned notification to show up and then get the fault. |
| 96 | + # FIXME(mriedem): Uncomment this when bug 1837955 is fixed. |
| 97 | + # self._wait_for_unversioned_notification( |
| 98 | + # 'compute_task.build_instances') |
| 99 | + # server = self.api.get_server(server['id']) |
| 100 | + # self.assertIn('fault', server) |
| 101 | + # self.assertIn('Exceeded maximum number of retries', |
| 102 | + # server['fault']['message']) |
| 103 | + |
| 104 | + # TODO(mriedem): Remove this when the bug is fixed. We need to assert |
| 105 | + # something before the bug is fixed to show the failure so check the |
| 106 | + # logs. |
| 107 | + for x in range(20): |
| 108 | + logs = self.stdlog.logger.output |
| 109 | + if 'MaxRetriesExceeded' in logs: |
| 110 | + break |
| 111 | + time.sleep(.5) |
| 112 | + else: |
| 113 | + self.fail('Timed out waiting for MaxRetriesExceeded to show up ' |
| 114 | + 'in the logs.') |
0 commit comments