def test_azure_kernel_upgrade_case_insensitive_uuid(
session_cloud: IntegrationCloud
):
cfg_image_spec = ImageSpecification.from_os_image()
if (cfg_image_spec.os, cfg_image_spec.release) != ("ubuntu", "bionic"):
pytest.skip(
"Test only supports ubuntu:bionic not {0.os}:{0.release}".format(
cfg_image_spec
)
)
source = get_validated_source(session_cloud)
if not source.installs_new_version():
pytest.skip(
"Provide CLOUD_INIT_SOURCE to install expected working cloud-init"
)
image_id = IMG_AZURE_UBUNTU_PRO_FIPS_BIONIC
with session_cloud.launch(
launch_kwargs={"image_id": image_id}
) as instance:
# We can't use setup_image fixture here because we want to avoid
# taking a snapshot or cleaning the booted machine after cloud-init
# upgrade.
instance.install_new_cloud_init(
source, take_snapshot=False, clean=False
)
_check_iid_insensitive_across_kernel_upgrade(instance)
def test_upgrade_package(session_cloud: IntegrationCloud):
if get_validated_source(session_cloud) != CloudInitSource.DEB_PACKAGE:
not_run_message = 'Test only supports upgrading to build deb'
if os.environ.get('TRAVIS'):
# If this isn't running on CI, we should know
pytest.fail(not_run_message)
else:
pytest.skip(not_run_message)
launch_kwargs = {'image_id': session_cloud.released_image_id}
with session_cloud.launch(launch_kwargs=launch_kwargs) as instance:
instance.install_deb()
instance.restart()
assert instance.execute('cloud-init status --wait --long').ok
def _setup_custom_image(session_cloud: IntegrationCloud):
"""Like `setup_image` in conftest.py, but with customized content."""
source = get_validated_source(session_cloud)
if not source.installs_new_version():
return
client = session_cloud.launch()
# Insert our "disable_network_activation" file here
client.write_to_file(
'/etc/cloud/cloud.cfg.d/99-disable-network-activation.cfg',
'disable_network_activation: true\n',
)
client.install_new_cloud_init(source)
# Even if we're keeping instances, we don't want to keep this
# one around as it was just for image creation
client.destroy()
def test_subsequent_boot_of_upgraded_package(session_cloud: IntegrationCloud):
source = get_validated_source(session_cloud)
if not source.installs_new_version():
if os.environ.get('TRAVIS'):
# If this isn't running on CI, we should know
pytest.fail(UNSUPPORTED_INSTALL_METHOD_MSG.format(source))
else:
pytest.skip(UNSUPPORTED_INSTALL_METHOD_MSG.format(source))
return # type checking doesn't understand that skip raises
launch_kwargs = {'image_id': session_cloud.released_image_id}
with session_cloud.launch(launch_kwargs=launch_kwargs) as instance:
instance.install_new_cloud_init(
source, take_snapshot=False, clean=False
)
instance.restart()
assert instance.execute('cloud-init status --wait --long').ok
def test_upgrade(session_cloud: IntegrationCloud):
source = get_validated_source(session_cloud)
if not source.installs_new_version():
pytest.skip(
"Install method '{}' not supported for this test".format(source))
return # type checking doesn't understand that skip raises
launch_kwargs = {
'image_id': session_cloud._get_initial_image(),
}
image = ImageSpecification.from_os_image()
# Get the paths to write test logs
output_dir = Path(session_cloud.settings.LOCAL_LOG_PATH)
output_dir.mkdir(parents=True, exist_ok=True)
base_filename = 'test_upgrade_{platform}_{os}_{{stage}}_{time}.log'.format(
platform=session_cloud.settings.PLATFORM,
os=image.release,
time=session_start_time,
)
before_path = output_dir / base_filename.format(stage='before')
after_path = output_dir / base_filename.format(stage='after')
# Get the network cfg file
netcfg_path = '/dev/null'
if image.os == 'ubuntu':
netcfg_path = '/etc/netplan/50-cloud-init.yaml'
if image.release == 'xenial':
netcfg_path = '/etc/network/interfaces.d/50-cloud-init.cfg'
with session_cloud.launch(
launch_kwargs=launch_kwargs,
user_data=USER_DATA,
wait=True,
) as instance:
_output_to_compare(instance, before_path, netcfg_path)
instance.install_new_cloud_init(source, take_snapshot=False)
instance.execute('hostname something-else')
_restart(instance)
_output_to_compare(instance, after_path, netcfg_path)
log.info('Wrote upgrade test logs to %s and %s', before_path, after_path)
def test_clean_boot_of_upgraded_package(session_cloud: IntegrationCloud):
source = get_validated_source(session_cloud)
if not source.installs_new_version():
pytest.skip(UNSUPPORTED_INSTALL_METHOD_MSG.format(source))
return # type checking doesn't understand that skip raises
if (ImageSpecification.from_os_image().release == 'bionic' and
session_cloud.settings.PLATFORM == 'lxd_vm'):
# The issues that we see on Bionic VMs don't appear anywhere
# else, including when calling KVM directly. It likely has to
# do with the extra lxd-agent setup happening on bionic.
# Given that we still have Bionic covered on all other platforms,
# the risk of skipping bionic here seems low enough.
pytest.skip("Upgrade test doesn't run on LXD VMs and bionic")
return
launch_kwargs = {
'image_id': session_cloud.released_image_id,
}
with session_cloud.launch(
launch_kwargs=launch_kwargs, user_data=USER_DATA,
) as instance:
# get pre values
pre_hostname = instance.execute('hostname')
pre_cloud_id = instance.execute('cloud-id')
pre_result = instance.execute('cat /run/cloud-init/result.json')
pre_network = instance.execute('cat /etc/netplan/50-cloud-init.yaml')
pre_systemd_analyze = instance.execute('systemd-analyze')
pre_systemd_blame = instance.execute('systemd-analyze blame')
pre_cloud_analyze = instance.execute('cloud-init analyze show')
pre_cloud_blame = instance.execute('cloud-init analyze blame')
# Ensure no issues pre-upgrade
assert not json.loads(pre_result)['v1']['errors']
log = instance.read_from_file('/var/log/cloud-init.log')
assert 'Traceback' not in log
assert 'WARN' not in log
# Upgrade and reboot
instance.install_new_cloud_init(source, take_snapshot=False)
instance.execute('hostname something-else')
instance.restart()
assert instance.execute('cloud-init status --wait --long').ok
# 'cloud-init init' helps us understand if our pickling upgrade paths
# have broken across re-constitution of a cached datasource. Some
# platforms invalidate their datasource cache on reboot, so we run
# it here to ensure we get a dirty run.
assert instance.execute('cloud-init init').ok
# get post values
post_hostname = instance.execute('hostname')
post_cloud_id = instance.execute('cloud-id')
post_result = instance.execute('cat /run/cloud-init/result.json')
post_network = instance.execute('cat /etc/netplan/50-cloud-init.yaml')
post_systemd_analyze = instance.execute('systemd-analyze')
post_systemd_blame = instance.execute('systemd-analyze blame')
post_cloud_analyze = instance.execute('cloud-init analyze show')
post_cloud_blame = instance.execute('cloud-init analyze blame')
# Ensure no issues post-upgrade
assert not json.loads(pre_result)['v1']['errors']
log = instance.read_from_file('/var/log/cloud-init.log')
assert 'Traceback' not in log
assert 'WARN' not in log
# Ensure important things stayed the same
assert pre_hostname == post_hostname
assert pre_cloud_id == post_cloud_id
assert pre_result == post_result
assert pre_network == post_network
# Calculate and log all the boot numbers
pre_analyze_totals = [
x for x in pre_cloud_analyze.splitlines()
if x.startswith('Finished stage') or x.startswith('Total Time')
]
post_analyze_totals = [
x for x in post_cloud_analyze.splitlines()
if x.startswith('Finished stage') or x.startswith('Total Time')
]
# pylint: disable=logging-format-interpolation
LOG.info(LOG_TEMPLATE.format(
pre_systemd_analyze=pre_systemd_analyze,
post_systemd_analyze=post_systemd_analyze,
pre_systemd_blame='\n'.join(pre_systemd_blame.splitlines()[:10]),
post_systemd_blame='\n'.join(post_systemd_blame.splitlines()[:10]),
pre_analyze_totals='\n'.join(pre_analyze_totals),
post_analyze_totals='\n'.join(post_analyze_totals),
pre_cloud_blame='\n'.join(pre_cloud_blame.splitlines()[:10]),
post_cloud_blame='\n'.join(post_cloud_blame.splitlines()[:10]),
))
def test_clean_boot_of_upgraded_package(session_cloud: IntegrationCloud):
source = get_validated_source(session_cloud)
if not source.installs_new_version():
pytest.skip(UNSUPPORTED_INSTALL_METHOD_MSG.format(source))
return # type checking doesn't understand that skip raises
if (ImageSpecification.from_os_image().release == "bionic"
and session_cloud.settings.PLATFORM == "lxd_vm"):
# The issues that we see on Bionic VMs don't appear anywhere
# else, including when calling KVM directly. It likely has to
# do with the extra lxd-agent setup happening on bionic.
# Given that we still have Bionic covered on all other platforms,
# the risk of skipping bionic here seems low enough.
pytest.skip("Upgrade test doesn't run on LXD VMs and bionic")
return
launch_kwargs = {
"image_id": session_cloud.released_image_id,
}
with session_cloud.launch(
launch_kwargs=launch_kwargs,
user_data=USER_DATA,
) as instance:
# get pre values
pre_hostname = instance.execute("hostname")
pre_cloud_id = instance.execute("cloud-id")
pre_result = instance.execute("cat /run/cloud-init/result.json")
pre_network = instance.execute("cat /etc/netplan/50-cloud-init.yaml")
pre_systemd_analyze = instance.execute("systemd-analyze")
pre_systemd_blame = instance.execute("systemd-analyze blame")
pre_cloud_analyze = instance.execute("cloud-init analyze show")
pre_cloud_blame = instance.execute("cloud-init analyze blame")
# Ensure no issues pre-upgrade
log = instance.read_from_file("/var/log/cloud-init.log")
assert not json.loads(pre_result)["v1"]["errors"]
try:
verify_clean_log(log)
except AssertionError:
LOG.warning("There were errors/warnings/tracebacks pre-upgrade. "
"Any failures may be due to pre-upgrade problem")
# Upgrade
instance.install_new_cloud_init(source, take_snapshot=False)
# 'cloud-init init' helps us understand if our pickling upgrade paths
# have broken across re-constitution of a cached datasource. Some
# platforms invalidate their datasource cache on reboot, so we run
# it here to ensure we get a dirty run.
assert instance.execute("cloud-init init").ok
# Reboot
instance.execute("hostname something-else")
instance.restart()
assert instance.execute("cloud-init status --wait --long").ok
# get post values
post_hostname = instance.execute("hostname")
post_cloud_id = instance.execute("cloud-id")
post_result = instance.execute("cat /run/cloud-init/result.json")
post_network = instance.execute("cat /etc/netplan/50-cloud-init.yaml")
post_systemd_analyze = instance.execute("systemd-analyze")
post_systemd_blame = instance.execute("systemd-analyze blame")
post_cloud_analyze = instance.execute("cloud-init analyze show")
post_cloud_blame = instance.execute("cloud-init analyze blame")
# Ensure no issues post-upgrade
assert not json.loads(pre_result)["v1"]["errors"]
log = instance.read_from_file("/var/log/cloud-init.log")
verify_clean_log(log)
# Ensure important things stayed the same
assert pre_hostname == post_hostname
assert pre_cloud_id == post_cloud_id
try:
assert pre_result == post_result
except AssertionError:
if instance.settings.PLATFORM == "azure":
pre_json = json.loads(pre_result)
post_json = json.loads(post_result)
assert pre_json["v1"]["datasource"].startswith(
"DataSourceAzure")
assert post_json["v1"]["datasource"].startswith(
"DataSourceAzure")
assert pre_network == post_network
# Calculate and log all the boot numbers
pre_analyze_totals = [
x for x in pre_cloud_analyze.splitlines()
if x.startswith("Finished stage") or x.startswith("Total Time")
]
post_analyze_totals = [
x for x in post_cloud_analyze.splitlines()
if x.startswith("Finished stage") or x.startswith("Total Time")
]
# pylint: disable=logging-format-interpolation
LOG.info(
LOG_TEMPLATE.format(
pre_systemd_analyze=pre_systemd_analyze,
post_systemd_analyze=post_systemd_analyze,
pre_systemd_blame="\n".join(
pre_systemd_blame.splitlines()[:10]),
post_systemd_blame="\n".join(
post_systemd_blame.splitlines()[:10]),
pre_analyze_totals="\n".join(pre_analyze_totals),
post_analyze_totals="\n".join(post_analyze_totals),
pre_cloud_blame="\n".join(pre_cloud_blame.splitlines()[:10]),
post_cloud_blame="\n".join(post_cloud_blame.splitlines()[:10]),
))
