def test_reload_configuration(duthost, conn_graph_facts):
"""
@summary: This test case is to reload the configuration and check platform status
"""
interfaces = conn_graph_facts["device_conn"]
asic_type = duthost.facts["asic_type"]
logging.info("Reload configuration")
duthost.shell("sudo config reload -y &>/dev/null", executable="/bin/bash")
logging.info("Wait until all critical services are fully started")
wait_critical_processes(duthost)
logging.info("Wait some time for all the transceivers to be detected")
assert wait_until(300, 20, check_interface_information, duthost, interfaces), \
"Not all transceivers are detected in 300 seconds"
logging.info("Check transceiver status")
check_transceiver_basic(duthost, interfaces)
if asic_type in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(duthost)
logging.info("Check sysfs")
check_sysfs(duthost)
def restart_service_and_check(localhost, dut, service, interfaces):
"""
Restart specified service and check platform status
"""
logging.info("Restart the %s service" % service)
dut.command("sudo systemctl restart %s" % service)
logging.info("Wait until all critical services are fully started")
wait_critical_processes(dut)
logging.info("Wait some time for all the transceivers to be detected")
pytest_assert(
wait_until(300, 20, check_interface_information, dut, interfaces),
"Not all interface information are detected within 300 seconds")
logging.info("Check transceiver status")
check_transceiver_basic(dut, interfaces)
if dut.facts["asic_type"] in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(dut)
logging.info("Check sysfs")
check_sysfs(dut)
logging.info("Check that critical processes are healthy for 60 seconds")
check_critical_processes(dut, 60)
def check_interfaces_and_services(dut,
interfaces,
xcvr_skip_list,
reboot_type=None):
"""
Perform a further check after reboot-cause, including transceiver status, interface status
@param localhost: The Localhost object.
@param dut: The AnsibleHost object of DUT.
@param interfaces: DUT's interfaces defined by minigraph
"""
logging.info("Wait until all critical services are fully started")
wait_critical_processes(dut)
if reboot_type is not None:
logging.info("Check reboot cause")
assert wait_until(MAX_WAIT_TIME_FOR_REBOOT_CAUSE, 20, check_reboot_cause, dut, reboot_type), \
"got reboot-cause failed after rebooted by %s" % reboot_type
if reboot_ctrl_dict[reboot_type]["test_reboot_cause_only"]:
logging.info(
"Further checking skipped for %s test which intends to verify reboot-cause only"
% reboot_type)
return
if dut.is_supervisor_node():
logging.info("skipping interfaces related check for supervisor")
else:
logging.info(
"Wait {} seconds for all the transceivers to be detected".format(
MAX_WAIT_TIME_FOR_INTERFACES))
result = wait_until(MAX_WAIT_TIME_FOR_INTERFACES, 20,
check_all_interface_information, dut, interfaces,
xcvr_skip_list)
assert result, "Not all transceivers are detected or interfaces are up in {} seconds".format(
MAX_WAIT_TIME_FOR_INTERFACES)
logging.info("Check transceiver status")
for asic_index in dut.get_frontend_asic_ids():
# Get the interfaces pertaining to that asic
interface_list = get_port_map(dut, asic_index)
interfaces_per_asic = {
k: v
for k, v in interface_list.items() if k in interfaces
}
check_transceiver_basic(dut, asic_index, interfaces_per_asic,
xcvr_skip_list)
logging.info("Check pmon daemon status")
assert check_pmon_daemon_status(dut), "Not all pmon daemons running."
if dut.facts["asic_type"] in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(dut)
logging.info("Check sysfs")
check_sysfs(dut)
def restart_service_and_check(localhost, dut, enum_frontend_asic_index,
service, interfaces, xcvr_skip_list):
"""
Restart specified service and check platform status
"""
logging.info("Restart the %s service on asic %s" %
(service, enum_frontend_asic_index))
asichost = dut.asic_instance(enum_frontend_asic_index)
service_name = asichost.get_docker_name(service)
dut.command("sudo systemctl restart {}".format(service_name))
for container in dut.get_default_critical_services_list():
if is_service_hiting_start_limit(dut, container) is True:
logging.info(
"{} hits start limit and clear reset-failed flag".format(
container))
dut.shell(
"sudo systemctl reset-failed {}.service".format(container))
dut.shell("sudo systemctl start {}.service".format(container))
logging.info("Wait until all critical services are fully started")
wait_critical_processes(dut)
logging.info("Wait some time for all the transceivers to be detected")
pytest_assert(
wait_until(300, 20, check_interface_information, dut,
enum_frontend_asic_index, interfaces, xcvr_skip_list),
"Not all interface information are detected within 300 seconds")
logging.info("Check transceiver status on asic %s" %
enum_frontend_asic_index)
check_transceiver_basic(dut, enum_frontend_asic_index, interfaces,
xcvr_skip_list)
if dut.facts["asic_type"] in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(dut)
logging.info("Check sysfs")
check_sysfs(dut)
logging.info("Check that critical processes are healthy for 60 seconds")
check_critical_processes(dut, 60)
def check_interfaces_and_transceivers(dut, interfaces):
"""
Perform a check of transceivers, LAGs and interfaces status
@param dut: The AnsibleHost object of DUT.
@param interfaces: DUT's interfaces defined by minigraph
"""
logging.info("Wait %d seconds for all the transceivers to be detected" %
MAX_WAIT_TIME_FOR_INTERFACES)
pytest_assert(wait_until(MAX_WAIT_TIME_FOR_INTERFACES, 20, check_interface_information, dut, interfaces), \
"Not all transceivers are detected or interfaces are up in %d seconds" % MAX_WAIT_TIME_FOR_INTERFACES)
logging.info("Check transceiver status")
check_transceiver_basic(dut, interfaces)
logging.info("Check LAGs and interfaces status")
checks.check_interfaces(dut)
def test_reload_configuration(duthosts, rand_one_dut_hostname,
conn_graph_facts, xcvr_skip_list):
"""
@summary: This test case is to reload the configuration and check platform status
"""
duthost = duthosts[rand_one_dut_hostname]
interfaces = conn_graph_facts["device_conn"][duthost.hostname]
asic_type = duthost.facts["asic_type"]
if config_force_option_supported(duthost):
assert wait_until(300, 20, 0, config_system_checks_passed, duthost)
logging.info("Reload configuration")
duthost.shell("sudo config reload -y &>/dev/null", executable="/bin/bash")
logging.info("Wait until all critical services are fully started")
wait_critical_processes(duthost)
logging.info("Wait some time for all the transceivers to be detected")
assert wait_until(300, 20, 0, check_all_interface_information, duthost, interfaces, xcvr_skip_list), \
"Not all transceivers are detected in 300 seconds"
logging.info("Check transceiver status")
for asic_index in duthost.get_frontend_asic_ids():
# Get the interfaces pertaining to that asic
interface_list = get_port_map(duthost, asic_index)
interfaces_per_asic = {
k: v
for k, v in interface_list.items() if k in interfaces
}
check_transceiver_basic(duthost, asic_index, interfaces_per_asic,
xcvr_skip_list)
if asic_type in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(duthost)
logging.info("Check sysfs")
check_sysfs(duthost)
def check_interfaces_and_services(dut, interfaces, reboot_type=None):
"""
Perform a further check after reboot-cause, including transceiver status, interface status
@param localhost: The Localhost object.
@param dut: The AnsibleHost object of DUT.
@param interfaces: DUT's interfaces defined by minigraph
"""
logging.info("Wait until all critical services are fully started")
wait_critical_processes(dut)
if reboot_type is not None:
logging.info("Check reboot cause")
assert wait_until(MAX_WAIT_TIME_FOR_REBOOT_CAUSE, 20, check_reboot_cause, dut, reboot_type), \
"got reboot-cause failed after rebooted by %s" % reboot_type
if reboot_ctrl_dict[reboot_type]["test_reboot_cause_only"]:
logging.info(
"Further checking skipped for %s test which intends to verify reboot-cause only"
% reboot_type)
return
logging.info("Wait %d seconds for all the transceivers to be detected" %
MAX_WAIT_TIME_FOR_INTERFACES)
assert wait_until(MAX_WAIT_TIME_FOR_INTERFACES, 20, check_interface_information, dut, interfaces), \
"Not all transceivers are detected or interfaces are up in %d seconds" % MAX_WAIT_TIME_FOR_INTERFACES
logging.info("Check transceiver status")
check_transceiver_basic(dut, interfaces)
logging.info("Check pmon daemon status")
assert check_pmon_daemon_status(dut), "Not all pmon daemons running."
if dut.facts["asic_type"] in ["mellanox"]:
from .mellanox.check_hw_mgmt_service import check_hw_management_service
from .mellanox.check_sysfs import check_sysfs
logging.info("Check the hw-management service")
check_hw_management_service(dut)
logging.info("Check sysfs")
check_sysfs(dut)
