heat/heat_integrationtests/common/test.py
Takashi Kajinami 1828df32fa Use Block Storage API v3 instead of API v2
Block Storage API v2 was deprecated during Pike cycle and is being
removed during Xena cycle, and current v3 API should be used instead.
Unused volume_client and network_client in integration test code are
also removed by this change.

Note:
granade tests is made non-voting temporally until the same issue is
fixed in stable/wallaby.

Depends-on: https://review.opendev.org/802150/
Change-Id: I6a2b5afa13480791971bbd8bba1f43b9f2db8294
2021-08-02 14:39:30 +09:00

669 lines
27 KiB
Python

# Licensed under the Apache License, Version 2.0 (the "License"); you may
# not use this file except in compliance with the License. You may obtain
# a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
# WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
# License for the specific language governing permissions and limitations
# under the License.
import os
import random
import re
import subprocess
import time
import urllib
import fixtures
from heatclient import exc as heat_exceptions
from keystoneauth1 import exceptions as kc_exceptions
from oslo_log import log as logging
from oslo_utils import timeutils
from tempest import config
import testscenarios
import testtools
from heat_integrationtests.common import clients
from heat_integrationtests.common import exceptions
LOG = logging.getLogger(__name__)
_LOG_FORMAT = "%(levelname)8s [%(name)s] %(message)s"
def call_until_true(duration, sleep_for, func, *args, **kwargs):
"""Call the function until it returns True or the duration elapsed.
Call the given function until it returns True (and return True) or
until the specified duration (in seconds) elapses (and return
False).
:param func: A zero argument callable that returns True on success.
:param duration: The number of seconds for which to attempt a
successful call of the function.
:param sleep_for: The number of seconds to sleep after an unsuccessful
invocation of the function.
"""
now = time.time()
timeout = now + duration
while now < timeout:
if func(*args, **kwargs):
return True
LOG.debug("Sleeping for %d seconds", sleep_for)
time.sleep(sleep_for)
now = time.time()
return False
def rand_name(name=''):
randbits = str(random.randint(1, 0x7fffffff))
if name:
return name + '-' + randbits
else:
return randbits
def requires_convergence(test_method):
'''Decorator for convergence-only tests.
The decorated test will be skipped when convergence is disabled.
'''
convergence_enabled = config.CONF.heat_plugin.convergence_engine_enabled
skipper = testtools.skipUnless(convergence_enabled,
"Convergence-only tests are disabled")
return skipper(test_method)
class HeatIntegrationTest(testscenarios.WithScenarios,
testtools.TestCase):
def setUp(self):
super(HeatIntegrationTest, self).setUp()
self.conf = config.CONF.heat_plugin
self.assertIsNotNone(self.conf.auth_url,
'No auth_url configured')
self.assertIsNotNone(self.conf.username,
'No username configured')
self.assertIsNotNone(self.conf.password,
'No password configured')
self.setup_clients(self.conf)
self.useFixture(fixtures.FakeLogger(format=_LOG_FORMAT))
if self.conf.disable_ssl_certificate_validation:
self.verify_cert = False
else:
self.verify_cert = self.conf.ca_file or True
def setup_clients(self, conf, admin_credentials=False):
self.manager = clients.ClientManager(conf, admin_credentials)
self.identity_client = self.manager.identity_client
self.keystone_client = self.manager.keystone_client
self.orchestration_client = self.manager.orchestration_client
self.compute_client = self.manager.compute_client
self.object_client = self.manager.object_client
self.client = self.orchestration_client
def setup_clients_for_admin(self):
self.setup_clients(self.conf, True)
def check_connectivity(self, check_ip):
def try_connect(ip):
try:
urllib.request.urlopen('http://%s/' % ip)
return True
except IOError:
return False
timeout = self.conf.connectivity_timeout
elapsed_time = 0
while not try_connect(check_ip):
time.sleep(10)
elapsed_time += 10
if elapsed_time > timeout:
raise exceptions.TimeoutException()
def _log_console_output(self, servers=None):
if not servers:
servers = self.compute_client.servers.list()
for server in servers:
LOG.info('Console output for %s', server.id)
LOG.info(server.get_console_output())
def _load_template(self, base_file, file_name, sub_dir=None):
sub_dir = sub_dir or ''
filepath = os.path.join(os.path.dirname(os.path.realpath(base_file)),
sub_dir, file_name)
with open(filepath) as f:
return f.read()
def create_keypair(self, client=None, name=None):
if client is None:
client = self.compute_client
if name is None:
name = rand_name('heat-keypair')
keypair = client.keypairs.create(name)
self.assertEqual(keypair.name, name)
def delete_keypair():
keypair.delete()
self.addCleanup(delete_keypair)
return keypair
def assign_keypair(self):
if self.conf.keypair_name:
self.keypair = None
self.keypair_name = self.conf.keypair_name
else:
self.keypair = self.create_keypair()
self.keypair_name = self.keypair.id
def _stack_rand_name(self):
test_name = self.id()
if test_name and '.' in test_name:
name = '-'.join(test_name.split('.')[-2:])
# remove 'testname(...)' cases
name = name.split('(')[0]
else:
name = self.__name__
return rand_name(name)
def is_service_available(self, service_type):
try:
self.identity_client.get_endpoint_url(
service_type, self.conf.region)
except kc_exceptions.EndpointNotFound:
return False
else:
return True
@staticmethod
def _stack_output(stack, output_key, validate_errors=True):
"""Return a stack output value for a given key."""
value = None
for o in stack.outputs:
if validate_errors and 'output_error' in o:
# scan for errors in the stack output.
raise ValueError(
'Unexpected output errors in %s : %s' % (
output_key, o['output_error']))
if o['output_key'] == output_key:
value = o['output_value']
return value
def _ping_ip_address(self, ip_address, should_succeed=True):
cmd = ['ping', '-c1', '-w1', ip_address]
def ping():
proc = subprocess.Popen(cmd,
stdout=subprocess.PIPE,
stderr=subprocess.PIPE)
proc.wait()
return (proc.returncode == 0) == should_succeed
return call_until_true(
self.conf.build_timeout, 1, ping)
def _wait_for_all_resource_status(self, stack_identifier,
status, failure_pattern='^.*_FAILED$',
success_on_not_found=False):
for res in self.client.resources.list(stack_identifier):
self._wait_for_resource_status(
stack_identifier, res.resource_name,
status, failure_pattern=failure_pattern,
success_on_not_found=success_on_not_found)
def _wait_for_resource_status(self, stack_identifier, resource_name,
status, failure_pattern='^.*_FAILED$',
success_on_not_found=False):
"""Waits for a Resource to reach a given status."""
fail_regexp = re.compile(failure_pattern)
build_timeout = self.conf.build_timeout
build_interval = self.conf.build_interval
res = None
start = timeutils.utcnow()
while timeutils.delta_seconds(start,
timeutils.utcnow()) < build_timeout:
try:
res = self.client.resources.get(
stack_identifier, resource_name)
except heat_exceptions.HTTPNotFound:
if success_on_not_found:
return
# ignore this, as the resource may not have
# been created yet
else:
if res.resource_status == status:
return
wait_for_action = status.split('_')[0]
resource_action = res.resource_status.split('_')[0]
if (resource_action == wait_for_action and
fail_regexp.search(res.resource_status)):
raise exceptions.StackResourceBuildErrorException(
resource_name=res.resource_name,
stack_identifier=stack_identifier,
resource_status=res.resource_status,
resource_status_reason=res.resource_status_reason)
time.sleep(build_interval)
message = ('Resource %s from stack %s failed to reach %s status '
'within the required time (%s s). Current resource '
'status: %s.' %
(resource_name, stack_identifier, status, build_timeout,
res.resource_status))
raise exceptions.TimeoutException(message)
def verify_resource_status(self, stack_identifier, resource_name,
status='CREATE_COMPLETE'):
try:
res = self.client.resources.get(stack_identifier, resource_name)
except heat_exceptions.HTTPNotFound:
return False
return res.resource_status == status
def _verify_status(self, stack, stack_identifier, status,
fail_regexp, is_action_cancelled=False):
if stack.stack_status == status:
if status == 'DELETE_COMPLETE' and stack.deletion_time is None:
# Wait for deleted_time to be filled, so that we have more
# confidence the operation is finished.
return False
else:
return True
wait_for_action = status.split('_')[0]
if (stack.action == wait_for_action and
fail_regexp.search(stack.stack_status)):
raise exceptions.StackBuildErrorException(
stack_identifier=stack_identifier,
stack_status=stack.stack_status,
stack_status_reason=stack.stack_status_reason)
return False
def _wait_for_stack_status(self, stack_identifier, status,
failure_pattern=None,
success_on_not_found=False,
signal_required=False,
resources_to_signal=None,
is_action_cancelled=False):
"""Waits for a Stack to reach a given status.
Note this compares the full $action_$status, e.g
CREATE_COMPLETE, not just COMPLETE which is exposed
via the status property of Stack in heatclient
"""
if failure_pattern:
fail_regexp = re.compile(failure_pattern)
elif 'FAILED' in status:
# If we're looking for e.g CREATE_FAILED, COMPLETE is unexpected.
fail_regexp = re.compile('^.*_COMPLETE$')
else:
fail_regexp = re.compile('^.*_FAILED$')
build_timeout = self.conf.build_timeout
build_interval = self.conf.build_interval
start = timeutils.utcnow()
while timeutils.delta_seconds(start,
timeutils.utcnow()) < build_timeout:
try:
stack = self.client.stacks.get(stack_identifier,
resolve_outputs=False)
except heat_exceptions.HTTPNotFound:
if success_on_not_found:
return
elif not any(s in status for s in ['CREATE', 'ADOPT']):
# raise exception when stack not found, if it's not
# in create or adopt (which should be the only two possible
# reason that stack may not have been created yet)
raise
# ignore this, as the resource may not have
# been created yet
else:
if self._verify_status(stack, stack_identifier, status,
fail_regexp, is_action_cancelled):
return
if signal_required:
self.signal_resources(resources_to_signal)
time.sleep(build_interval)
message = ('Stack %s failed to reach %s status within '
'the required time (%s s). Current stack state: %s.' %
(stack_identifier, status, build_timeout,
stack.stack_status))
raise exceptions.TimeoutException(message)
def _stack_delete(self, stack_identifier):
try:
self._handle_in_progress(self.client.stacks.delete,
stack_identifier)
except heat_exceptions.HTTPNotFound:
pass
self._wait_for_stack_status(
stack_identifier, 'DELETE_COMPLETE',
success_on_not_found=True)
def _handle_in_progress(self, fn, *args, **kwargs):
build_timeout = self.conf.build_timeout
build_interval = self.conf.build_interval
start = timeutils.utcnow()
while timeutils.delta_seconds(start,
timeutils.utcnow()) < build_timeout:
try:
fn(*args, **kwargs)
except heat_exceptions.HTTPConflict as ex:
# FIXME(sirushtim): Wait a little for the stack lock to be
# released and hopefully, the stack should be usable again.
if ex.error['error']['type'] != 'ActionInProgress':
raise ex
time.sleep(build_interval)
else:
break
def update_stack(self, stack_identifier, template=None, environment=None,
files=None, parameters=None, tags=None,
expected_status='UPDATE_COMPLETE',
disable_rollback=True,
existing=False):
env = environment or {}
env_files = files or {}
parameters = parameters or {}
self._handle_in_progress(
self.client.stacks.update,
stack_id=stack_identifier,
template=template,
files=env_files,
disable_rollback=disable_rollback,
parameters=parameters,
environment=env,
tags=tags,
existing=existing)
kwargs = {'stack_identifier': stack_identifier,
'status': expected_status}
if expected_status in ['ROLLBACK_COMPLETE']:
# To trigger rollback you would intentionally fail the stack
# Hence check for rollback failures
kwargs['failure_pattern'] = '^ROLLBACK_FAILED$'
self._wait_for_stack_status(**kwargs)
def cancel_update_stack(self, stack_identifier, rollback=True,
expected_status='ROLLBACK_COMPLETE'):
stack_name = stack_identifier.split('/')[0]
if rollback:
self.client.actions.cancel_update(stack_name)
else:
self.client.actions.cancel_without_rollback(stack_name)
kwargs = {'stack_identifier': stack_identifier,
'status': expected_status}
if expected_status == 'UPDATE_FAILED':
kwargs['is_action_cancelled'] = True
if expected_status in ['ROLLBACK_COMPLETE']:
# To trigger rollback you would intentionally fail the stack
# Hence check for rollback failures
kwargs['failure_pattern'] = '^ROLLBACK_FAILED$'
self._wait_for_stack_status(**kwargs)
def preview_update_stack(self, stack_identifier, template,
environment=None, files=None, parameters=None,
tags=None, disable_rollback=True,
show_nested=False):
env = environment or {}
env_files = files or {}
parameters = parameters or {}
return self.client.stacks.preview_update(
stack_id=stack_identifier,
template=template,
files=env_files,
disable_rollback=disable_rollback,
parameters=parameters,
environment=env,
tags=tags,
show_nested=show_nested
)
def assert_resource_is_a_stack(self, stack_identifier, res_name,
wait=False):
build_timeout = self.conf.build_timeout
build_interval = self.conf.build_interval
start = timeutils.utcnow()
while timeutils.delta_seconds(start,
timeutils.utcnow()) < build_timeout:
time.sleep(build_interval)
try:
nested_identifier = self._get_nested_identifier(
stack_identifier, res_name)
except Exception:
# We may have to wait, if the create is in-progress
if wait:
time.sleep(build_interval)
else:
raise
else:
return nested_identifier
def _get_nested_identifier(self, stack_identifier, res_name):
rsrc = self.client.resources.get(stack_identifier, res_name)
nested_link = [lk for lk in rsrc.links if lk['rel'] == 'nested']
nested_href = nested_link[0]['href']
nested_id = nested_href.split('/')[-1]
nested_identifier = '/'.join(nested_href.split('/')[-2:])
self.assertEqual(rsrc.physical_resource_id, nested_id)
nested_stack = self.client.stacks.get(nested_id, resolve_outputs=False)
nested_identifier2 = '%s/%s' % (nested_stack.stack_name,
nested_stack.id)
self.assertEqual(nested_identifier, nested_identifier2)
parent_id = stack_identifier.split("/")[-1]
self.assertEqual(parent_id, nested_stack.parent)
return nested_identifier
def group_nested_identifier(self, stack_identifier,
group_name):
# Get the nested stack identifier from a group resource
rsrc = self.client.resources.get(stack_identifier, group_name)
physical_resource_id = rsrc.physical_resource_id
nested_stack = self.client.stacks.get(physical_resource_id,
resolve_outputs=False)
nested_identifier = '%s/%s' % (nested_stack.stack_name,
nested_stack.id)
parent_id = stack_identifier.split("/")[-1]
self.assertEqual(parent_id, nested_stack.parent)
return nested_identifier
def list_group_resources(self, stack_identifier,
group_name, minimal=True):
nested_identifier = self.group_nested_identifier(stack_identifier,
group_name)
if minimal:
return self.list_resources(nested_identifier)
return self.client.resources.list(nested_identifier)
def list_resources(self, stack_identifier, filter_func=None):
resources = self.client.resources.list(stack_identifier)
return dict((r.resource_name, r.resource_type) for r in resources
if (filter_func(r) if callable(filter_func) else True))
def get_resource_stack_id(self, r):
stack_link = [lk for lk in r.links if lk.get('rel') == 'stack'][0]
return stack_link['href'].split("/")[-1]
def get_physical_resource_id(self, stack_identifier, resource_name):
try:
resource = self.client.resources.get(
stack_identifier, resource_name)
return resource.physical_resource_id
except Exception:
raise Exception('Resource (%s) not found in stack (%s)!' %
(stack_identifier, resource_name))
def get_stack_output(self, stack_identifier, output_key,
validate_errors=True):
stack = self.client.stacks.get(stack_identifier)
return self._stack_output(stack, output_key, validate_errors)
def check_input_values(self, group_resources, key, value):
# Check inputs for deployment and derived config
for r in group_resources:
d = self.client.software_deployments.get(
r.physical_resource_id)
self.assertEqual({key: value}, d.input_values)
c = self.client.software_configs.get(
d.config_id)
foo_input_c = [i for i in c.inputs if i.get('name') == key][0]
self.assertEqual(value, foo_input_c.get('value'))
def signal_resources(self, resources):
# Signal all IN_PROGRESS resources
for r in resources:
if 'IN_PROGRESS' in r.resource_status:
stack_id = self.get_resource_stack_id(r)
self.client.resources.signal(stack_id, r.resource_name)
def stack_create(self, stack_name=None, template=None, files=None,
parameters=None, environment=None, tags=None,
expected_status='CREATE_COMPLETE',
disable_rollback=True, enable_cleanup=True,
environment_files=None, timeout=None):
name = stack_name or self._stack_rand_name()
templ = template or self.template
templ_files = files or {}
params = parameters or {}
env = environment or {}
timeout_mins = timeout or self.conf.build_timeout
self.client.stacks.create(
stack_name=name,
template=templ,
files=templ_files,
disable_rollback=disable_rollback,
parameters=params,
environment=env,
tags=tags,
environment_files=environment_files,
timeout_mins=timeout_mins
)
if enable_cleanup:
self.addCleanup(self._stack_delete, name)
stack = self.client.stacks.get(name, resolve_outputs=False)
stack_identifier = '%s/%s' % (name, stack.id)
kwargs = {'stack_identifier': stack_identifier,
'status': expected_status}
if expected_status:
if expected_status in ['ROLLBACK_COMPLETE']:
# To trigger rollback you would intentionally fail the stack
# Hence check for rollback failures
kwargs['failure_pattern'] = '^ROLLBACK_FAILED$'
self._wait_for_stack_status(**kwargs)
return stack_identifier
def stack_adopt(self, stack_name=None, files=None,
parameters=None, environment=None, adopt_data=None,
wait_for_status='ADOPT_COMPLETE'):
if (self.conf.skip_test_stack_action_list and
'ADOPT' in self.conf.skip_test_stack_action_list):
self.skipTest('Testing Stack adopt disabled in conf, skipping')
name = stack_name or self._stack_rand_name()
templ_files = files or {}
params = parameters or {}
env = environment or {}
self.client.stacks.create(
stack_name=name,
files=templ_files,
disable_rollback=True,
parameters=params,
environment=env,
adopt_stack_data=adopt_data,
)
self.addCleanup(self._stack_delete, name)
stack = self.client.stacks.get(name, resolve_outputs=False)
stack_identifier = '%s/%s' % (name, stack.id)
self._wait_for_stack_status(stack_identifier, wait_for_status)
return stack_identifier
def stack_abandon(self, stack_id):
if (self.conf.skip_test_stack_action_list and
'ABANDON' in self.conf.skip_test_stack_action_list):
self.addCleanup(self._stack_delete, stack_id)
self.skipTest('Testing Stack abandon disabled in conf, skipping')
info = self.client.stacks.abandon(stack_id=stack_id)
return info
def stack_snapshot(self, stack_id,
wait_for_status='SNAPSHOT_COMPLETE'):
snapshot = self.client.stacks.snapshot(stack_id=stack_id)
self._wait_for_stack_status(stack_id, wait_for_status)
return snapshot['id']
def stack_restore(self, stack_id, snapshot_id,
wait_for_status='RESTORE_COMPLETE'):
self.client.stacks.restore(stack_id, snapshot_id)
self._wait_for_stack_status(stack_id, wait_for_status)
def stack_suspend(self, stack_identifier):
if (self.conf.skip_test_stack_action_list and
'SUSPEND' in self.conf.skip_test_stack_action_list):
self.addCleanup(self._stack_delete, stack_identifier)
self.skipTest('Testing Stack suspend disabled in conf, skipping')
self._handle_in_progress(self.client.actions.suspend, stack_identifier)
# improve debugging by first checking the resource's state.
self._wait_for_all_resource_status(stack_identifier,
'SUSPEND_COMPLETE')
self._wait_for_stack_status(stack_identifier, 'SUSPEND_COMPLETE')
def stack_resume(self, stack_identifier):
if (self.conf.skip_test_stack_action_list and
'RESUME' in self.conf.skip_test_stack_action_list):
self.addCleanup(self._stack_delete, stack_identifier)
self.skipTest('Testing Stack resume disabled in conf, skipping')
self._handle_in_progress(self.client.actions.resume, stack_identifier)
# improve debugging by first checking the resource's state.
self._wait_for_all_resource_status(stack_identifier,
'RESUME_COMPLETE')
self._wait_for_stack_status(stack_identifier, 'RESUME_COMPLETE')
def wait_for_event_with_reason(self, stack_identifier, reason,
rsrc_name=None, num_expected=1):
build_timeout = self.conf.build_timeout
build_interval = self.conf.build_interval
start = timeutils.utcnow()
while timeutils.delta_seconds(start,
timeutils.utcnow()) < build_timeout:
try:
rsrc_events = self.client.events.list(stack_identifier,
resource_name=rsrc_name)
except heat_exceptions.HTTPNotFound:
LOG.debug("No events yet found for %s", rsrc_name)
else:
matched = [e for e in rsrc_events
if e.resource_status_reason == reason]
if len(matched) == num_expected:
return matched
time.sleep(build_interval)
def check_autoscale_complete(self, stack_id, expected_num, parent_stack,
group_name):
res_list = self.client.resources.list(stack_id)
all_res_complete = all(res.resource_status in ('UPDATE_COMPLETE',
'CREATE_COMPLETE')
for res in res_list)
all_res = len(res_list) == expected_num
if all_res and all_res_complete:
metadata = self.client.resources.metadata(parent_stack, group_name)
return not metadata.get('scaling_in_progress')
return False