x
/
tobiko
Code Issues Proposed changes

Merge "Fix service start/stop in neutron agent tests to support OSP13"

This commit is contained in:
Zuul 2020-10-22 16:43:59 +00:00 committed by Gerrit Code Review
parent 705164604a bac80c37fd
commit dfe03b1742
4 changed files with 107 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tobiko/openstack/topology/__init__.py
View File

@ -22,8 +22,10 @@ NoSuchOpenStackTopologyNode = _exception.NoSuchOpenStackTopologyNode
UnknowOpenStackServiceNameError = _topology.UnknowOpenStackServiceNameError
get_agent_service_name = _topology.get_agent_service_name
get_agent_container_name = _topology.get_agent_container_name
get_openstack_topology = _topology.get_openstack_topology
get_openstack_node = _topology.get_openstack_node
check_systemd_monitors_agent = _topology.check_systemd_monitors_agent
find_openstack_node = _topology.find_openstack_node
get_default_openstack_topology_class = (
_topology.get_default_openstack_topology_class)

32
tobiko/openstack/topology/_topology.py
View File

@ -96,11 +96,33 @@ def get_agent_service_name(agent_name: str) -> str:
return topology_class.get_agent_service_name(agent_name)
def get_agent_container_name(agent_name: str) -> str:
topology_class = get_default_openstack_topology_class()
return topology_class.get_agent_container_name(agent_name)
def check_systemd_monitors_agent(hostname: str, agent_name: str) -> bool:
ssh_client = get_openstack_node(hostname=hostname).ssh_client
services = sh.execute('systemctl list-units --type=service --all '
'--no-pager --no-legend',
ssh_client=ssh_client,
sudo=True).stdout.split('\n')
for service in services:
if get_agent_service_name(agent_name) in service:
return True
return False
class UnknowOpenStackServiceNameError(tobiko.TobikoException):
message = ("Unknown service name for agent name '{agent_name}' and "
"topology class '{topology_class}'")
class UnknowOpenStackContainerNameError(tobiko.TobikoException):
message = ("Unknown container name for agent name '{agent_name}' and "
"topology class '{topology_class}'")
class OpenStackTopologyNode(object):
_docker_client = None
@ -161,6 +183,7 @@ class OpenStackTopology(tobiko.SharedFixture):
neutron.OPENVSWITCH_AGENT: 'devstack@q-agt',
neutron.METADATA_AGENT: 'devstack@q-meta'
}
agent_to_container_name_mappings: typing.Dict[str, str] = {}
has_containers = False
@ -195,6 +218,15 @@ class OpenStackTopology(tobiko.SharedFixture):
raise UnknowOpenStackServiceNameError(agent_name=agent_name,
topology_class=cls)
@classmethod
def get_agent_container_name(cls, agent_name: str) -> str:
try:
return cls.agent_to_container_name_mappings[agent_name]
except KeyError:
pass
raise UnknowOpenStackContainerNameError(agent_name=agent_name,
topology_class=cls)
def discover_nodes(self):
self.discover_configured_nodes()
self.discover_controller_nodes()

105
tobiko/tests/faults/agents/test_neutron_agents.py
View File

@ -43,6 +43,7 @@ class BaseAgentTest(testtools.TestCase):
@classmethod
def setUpClass(cls):
cls.service_name: str = topology.get_agent_service_name(cls.agent_name)
cls.container_name: str = ''
cls.agents: AgentListType = \
neutron.list_networking_agents(binary=cls.agent_name)
@ -50,13 +51,14 @@ class BaseAgentTest(testtools.TestCase):
def hosts(self) -> typing.List[str]:
return [agent['host'] for agent in self.agents]
def stop_service(self, hosts: typing.Optional[typing.List[str]] = None):
'''Stop systemd service on hosts
def stop_agent(self, hosts: typing.Optional[typing.List[str]] = None):
'''Stop network agent on hosts
It ensures service service is stopped and register systemd service
restart as test case cleanup.
restart as test case cleanup. In case of systemd service is not
available it stops the container itself
:parm hosts: List of hostnames to stop service on
:parm hosts: List of hostnames to stop agent on
:type hosts: list of strings
'''
hosts = hosts or self.hosts
@ -64,23 +66,36 @@ class BaseAgentTest(testtools.TestCase):
for host in hosts:
ssh_client = topology.get_openstack_node(hostname=host).ssh_client
LOG.debug(f"Stopping service '{self.service_name}' on "
f"host '{host}'...")
sh.execute(f"systemctl stop {self.service_name}",
ssh_client=ssh_client,
sudo=True)
LOG.debug(f"Service '{self.service_name}' stopped on host "
f"'{host}'.")
is_systemd = topology.check_systemd_monitors_agent(host,
self.agent_name)
if is_systemd:
LOG.debug(f"Stopping service '{self.service_name}' on "
f"host '{host}'...")
sh.execute(f"systemctl stop {self.service_name}",
ssh_client=ssh_client,
sudo=True)
LOG.debug(f"Service '{self.service_name}' stopped on host "
f"'{host}'.")
else:
self.container_name = \
topology.get_agent_container_name(self.agent_name)
LOG.debug(f'Stopping container {self.container_name} on '
f"host '{host}'...")
sh.execute(f'docker stop {self.container_name}',
ssh_client=ssh_client,
sudo=True)
LOG.debug(f'Container {self.container_name} has been stopped '
f"on host '{host}'...")
# Schedule auto-restart of service at the end of this test case
self.addCleanup(sh.execute, f"systemctl start {self.service_name}",
ssh_client=ssh_client, sudo=True)
self.addCleanup(self.start_agent, hosts=[host, ])
def start_service(self, hosts: typing.Optional[typing.List[str]] = None):
'''Start systemd service on hosts
def start_agent(self, hosts: typing.Optional[typing.List[str]] = None):
'''Start network agent on hosts
It ensures system service is running.
It ensures system service is running. If the systemd service is not
available it starts container itself
:parm hosts: List of hostnames to start service on
:parm hosts: List of hostnames to start agent on
:type hosts: list of strings
'''
hosts = hosts or self.hosts
@ -88,10 +103,22 @@ class BaseAgentTest(testtools.TestCase):
for host in hosts:
ssh_client = topology.get_openstack_node(hostname=host).ssh_client
LOG.debug(f"Starting service '{self.service_name}' on "
f"host '{host}'...")
sh.execute(f"systemctl start {self.service_name}",
ssh_client=ssh_client, sudo=True)
is_systemd = topology.check_systemd_monitors_agent(host,
self.agent_name)
if is_systemd:
LOG.debug(f"Starting service '{self.service_name}' on "
f"host '{host}'...")
sh.execute(f"systemctl start {self.service_name}",
ssh_client=ssh_client,
sudo=True)
else:
self.container_name = \
topology.get_agent_container_name(self.agent_name)
LOG.debug(f'Starting container {self.container_name} on '
f"host '{host}'...")
sh.execute(f'docker start {self.container_name}',
ssh_client=ssh_client,
sudo=True)
def get_cmd_pids(self, process_name, command_filter, hosts=None,
timeout=120, interval=2, min_pids_per_host=1) -> \
@ -255,11 +282,11 @@ class DHCPAgentTest(BaseAgentTest):
f"'{self.stack.network}'")
pids = self.get_cmd_pids("dnsmasq", self.stack.network)
self.stop_service()
self.stop_agent()
self.assertEqual(pids, self.get_cmd_pids("dnsmasq",
self.stack.network))
self.start_service()
self.start_agent()
self.wait_processes_destroyed(self.stack.network, pids)
new_pids = self.get_cmd_pids("dnsmasq", self.stack.network)
self.assertNotEqual(pids, new_pids)
@ -279,13 +306,13 @@ class DHCPAgentTest(BaseAgentTest):
[], self.agents, "No DHCP agent found serving network "
f"'{self.stack.network}'")
pids = self.get_cmd_pids("dnsmasq", self.stack.network)
self.stop_service()
self.stop_agent()
nova.shutoff_server(self.stack.server_id)
nova.activate_server(self.stack.server_id)
ping.ping_until_received(self.stack.ip_address).assert_replied()
self.start_service()
self.start_agent()
self.wait_processes_destroyed(self.stack.network, pids)
new_pids = self.get_cmd_pids("dnsmasq", self.stack.network)
self.assertNotEqual(pids, new_pids)
@ -329,8 +356,8 @@ class L3AgentTest(BaseAgentTest):
ha_router_id = self.ha_stack.network_stack.gateway_id
pids = self.get_cmd_pids('keepalived', ha_router_id,
min_pids_per_host=2)
self.stop_service()
self.start_service()
self.stop_agent()
self.start_agent()
self.agents = [self.wait_for_active_ha_l3_agent()]
self.assertEqual(pids,
self.get_cmd_pids('keepalived',
@ -357,7 +384,7 @@ class L3AgentTest(BaseAgentTest):
for _ in tobiko.retry(timeout=120., interval=5.):
new_ka_state_pids = self.get_cmd_pids('/usr/bin/python',
ka_state_cmd,
min_pids_per_host=2)
min_pids_per_host=1)
if ka_state_pids == new_ka_state_pids:
break
else:
@ -377,13 +404,13 @@ class L3AgentTest(BaseAgentTest):
def test_metadata_haproxy_during_stop_L3_agent(self):
self.agents = neutron.list_l3_agent_hosting_routers(self.router_id)
pids = self.get_cmd_pids("haproxy", self.router_id)
self.stop_service()
self.stop_agent()
# Now check if haproxy processes are still run and have got same pids
# like before dhcp agent's stop
self.assertEqual(pids, self.get_cmd_pids("haproxy", self.router_id))
self.start_service()
self.start_agent()
# And finally check if haproxy processes are still run and have got
# same pids like at the beginning of the test
@ -416,12 +443,12 @@ class L3AgentTest(BaseAgentTest):
self.agents = neutron.list_l3_agent_hosting_routers(self.router_id)
pids = self.get_cmd_pids("radvd", self.router_id)
self.stop_service()
self.stop_agent()
# Now check if radvd processes are still run and have got same pids
# like before dhcp agent's stop
self.assertEqual(pids, self.get_cmd_pids("radvd", self.router_id))
self.start_service()
self.start_agent()
# And finally check if dnsmasq processes are still run and have got
# same pids like at the beginning of the test
@ -447,11 +474,11 @@ class OpenVSwitchAgentTest(BaseAgentTest):
def test_vm_reachability_during_stop_ovs_agent(self):
# Check if vm is reachable before stopping service
self.start_service()
self.start_agent()
ping.ping_until_received(self.stack.ip_address).assert_replied()
# Check if vm is reachable after stopping service
self.stop_service(hosts=[self.stack.hypervisor_host])
self.stop_agent(hosts=[self.stack.hypervisor_host])
ping.ping_until_received(self.stack.ip_address).assert_replied()
@ -522,22 +549,22 @@ class MetadataAgentTest(BaseAgentTest):
def test_metadata_service_restart(self):
# Ensure service is up
self.start_service()
self.start_agent()
self.wait_for_metadata_status(is_reachable=True)
# Ensure the servive gets down
self.stop_service()
self.stop_agent()
self.wait_for_metadata_status(is_reachable=False)
# Ensure service gets up again
self.start_service()
self.start_agent()
self.wait_for_metadata_status(is_reachable=True)
def test_vm_reachability_when_metadata_agent_is_down(self):
self.stop_service()
self.stop_agent()
self.wait_for_metadata_status(is_reachable=False)
ping.ping_until_received(self.stack.ip_address).assert_replied()
self.start_service()
self.start_agent()
self.wait_for_metadata_status(is_reachable=True)

7
tobiko/tripleo/_topology.py
View File

@ -36,6 +36,13 @@ class TripleoTopology(topology.OpenStackTopology):
neutron.METADATA_AGENT: 'tripleo_neutron_metadata_agent'
}
agent_to_container_name_mappings = {
neutron.DHCP_AGENT: 'neutron_dhcp',
neutron.L3_AGENT: 'neutron_l3_agent',
neutron.OPENVSWITCH_AGENT: 'neutron_ovs_agent',
neutron.METADATA_AGENT: 'neutron_metadata_agent'
}
has_containers = True
# TODO: add more known subgrups here