def test_agent_restart_with_checkpointing_enabled():
(driver_task_id, driver_task, executor_task) = _submit_job_and_get_tasks(
extra_args=["--conf spark.mesos.checkpoint=true"])
driver_ip = sdk_networks.get_task_host(driver_task)
executor_ip = sdk_networks.get_task_host(executor_task)
utils.restart_task_agent_and_verify_state(executor_ip, executor_task,
"TASK_RUNNING")
utils.restart_task_agent_and_verify_state(driver_ip, driver_task,
"TASK_RUNNING")
_kill_driver_task(driver_task_id)
def test_agent_restart_with_checkpointing_disabled():
(driver_task_id, driver_task, executor_task) = _submit_job_and_get_tasks()
driver_ip = sdk_networks.get_task_host(driver_task)
executor_ip = sdk_networks.get_task_host(executor_task)
# Dispatcher starts Driver Tasks with checkpointing enabled so Driver is expected to be in RUNNING state
utils.restart_task_agent_and_verify_state(driver_ip, driver_task,
"TASK_RUNNING")
utils.restart_task_agent_and_verify_state(executor_ip, executor_task,
"TASK_LOST")
_kill_driver_task(driver_task_id)
def test_offers_suppressed_with_lost_task():
app_name = "MockTaskRunner"
driver_task_id = _launch_test_task(app_name)
executor_task = shakedown.get_service_tasks(app_name)[0]
executor_ip = sdk_networks.get_task_host(executor_task)
utils.restart_task_agent_and_verify_state(executor_ip, executor_task, "TASK_LOST")
shakedown.wait_for_task_completion(driver_task_id, timeout_sec=utils.JOB_WAIT_TIMEOUT_SECONDS)
_check_logged_offers(driver_task_id, 1)
def docker_inspect(task, format_options=None):
host_ip = sdk_networks.get_task_host(task)
task_id = dcos_utils.get_task_container_id(task)
inspect_cmd = "sudo docker inspect "
if format_options is not None:
inspect_cmd = inspect_cmd + format_options + " "
inspect_cmd = inspect_cmd + "mesos-" + task_id
return dcos_utils.agent_ssh(host_ip, inspect_cmd)
def _check_task_network(task, is_ucr=True):
host_ip = sdk_networks.get_task_host(task)
task_ip = sdk_networks.get_task_ip(task)
subnet = sdk_networks.get_overlay_subnet()
_verify_task_ip(task_ip, host_ip, subnet)
_verify_task_network_name(task)
if is_ucr:
_verify_ucr_task_inet_address(task, subnet)
else:
_check_docker_network(task, host_ip, subnet)
def docker_exec(task, cmd):
host_ip = sdk_networks.get_task_host(task)
task_id = dcos_utils.get_task_container_id(task)
exec_command = "sudo docker exec mesos-{} {}".format(task_id, cmd)
return dcos_utils.agent_ssh(host_ip, exec_command)