def ensure_process_is_running(duthost, container_name, critical_process):
"""Checks the running status of a critical process and starts it if it was not running.
Args:
duthost: Hostname of DUT.
container_name: A string shows name of a container.
critical_process: A string shows name of a process.
Returns:
None.
"""
logger.info(
"Checking whether process '{}' in container '{}' is running...".format(
critical_process, container_name))
program_status, program_pid = get_program_info(duthost, container_name,
critical_process)
if program_status == "RUNNING":
logger.info("Process '{}' in container '{} is running.".format(
critical_process, container_name))
else:
logger.info(
"Process '{}' in container '{}' is not running and start it...".
format(critical_process, container_name))
command_output = duthost.shell(
"docker exec {} supervisorctl start {}".format(
container_name, critical_process))
if command_output["rc"] == 0:
logger.info("Process '{}' in container '{}' is started.".format(
critical_process, container_name))
else:
pytest.fail(
"Failed to start process '{}' in container '{}'.".format(
critical_process, container_name))
def is_tunnel_packet_handler_running(duthost):
"""Check if tunnel_packet_handler is running on duthost
Args:
duthost (AnsibleHost): Device Under Test (DUT)
Returns:
bool: True if tunnel_packet_handler is running. Otherwise, return False.
"""
status, _ = get_program_info(duthost, "swss", "tunnel_packet_handler")
return status == 'RUNNING'
def stop_critical_processes(duthost, containers_in_namespaces):
"""Gets critical processes of each running container and then stops them from running.
Args:
duthost: Hostname of DUT.
containers_in_namespaces: A dictionary where keys are container names and
values are lists which contains ids of namespaces this container should reside in.
Returns:
None.
"""
for container_name in containers_in_namespaces.keys():
namespace_ids = containers_in_namespaces[container_name]
container_name_in_namespace = container_name
# If a container is only running on host, then namespace_ids is [None]
# If a container is running on multi-ASIC, then namespace_ids is [0, 1, ...]
# If a container is running on host and multi-ASICs, then namespace_ids is [None, 0, 1, ...]
if len(namespace_ids) >= 2:
container_name_in_namespace += namespace_ids[1]
critical_group_list, critical_process_list, succeeded = duthost.get_critical_group_and_process_lists(
container_name_in_namespace)
pytest_assert(
succeeded,
"Failed to get critical group and process lists of container '{}'".
format(container_name_in_namespace))
for namespace_id in namespace_ids:
container_name_in_namespace = container_name
if namespace_id != DEFAULT_ASIC_ID:
container_name_in_namespace += namespace_id
for critical_process in critical_process_list:
# Skip 'dsserve' process since it was not managed by supervisord
# TODO: Should remove the following two lines once the issue was solved in the image.
if "syncd" in container_name_in_namespace and critical_process == "dsserve":
continue
program_status, program_pid = get_program_info(
duthost, container_name_in_namespace, critical_process)
check_and_kill_process(duthost, container_name_in_namespace,
critical_process, program_status,
program_pid)
for critical_group in critical_group_list:
group_program_info = get_group_program_info(
duthost, container_name_in_namespace, critical_group)
for program_name in group_program_info:
check_and_kill_process(duthost,
container_name_in_namespace,
critical_group + ":" + program_name,
group_program_info[program_name][0],
group_program_info[program_name][1])