def queue_task_instance(self,
task_instance: TaskInstance,
mark_success: bool = False,
pickle_id: Optional[str] = None,
ignore_all_deps: bool = False,
ignore_depends_on_past: bool = False,
ignore_task_deps: bool = False,
ignore_ti_state: bool = False,
pool: Optional[str] = None,
cfg_path: Optional[str] = None) -> None:
"""Queues task instance."""
pool = pool or task_instance.pool
# TODO (edgarRd): AIRFLOW-1985:
# cfg_path is needed to propagate the config values if using impersonation
# (run_as_user), given that there are different code paths running tasks.
# For a long term solution we need to address AIRFLOW-1986
command_list_to_run = task_instance.command_as_list(
local=True,
mark_success=mark_success,
ignore_all_deps=ignore_all_deps,
ignore_depends_on_past=ignore_depends_on_past,
ignore_task_deps=ignore_task_deps,
ignore_ti_state=ignore_ti_state,
pool=pool,
pickle_id=pickle_id,
cfg_path=cfg_path)
self.queue_command(SimpleTaskInstance(task_instance),
command_list_to_run,
priority=task_instance.task.priority_weight_total,
queue=task_instance.task.queue)
def generate_pod_yaml(args):
"""Generates yaml files for each task in the DAG. Used for testing output of KubernetesExecutor"""
execution_date = args.execution_date
dag = get_dag(subdir=args.subdir, dag_id=args.dag_id)
yaml_output_path = args.output_path
kube_config = KubeConfig()
for task in dag.tasks:
ti = TaskInstance(task, execution_date)
pod = PodGenerator.construct_pod(
dag_id=args.dag_id,
task_id=ti.task_id,
pod_id=create_pod_id(args.dag_id, ti.task_id),
try_number=ti.try_number,
kube_image=kube_config.kube_image,
date=ti.execution_date,
args=ti.command_as_list(),
pod_override_object=PodGenerator.from_obj(ti.executor_config),
scheduler_job_id="worker-config",
namespace=kube_config.executor_namespace,
base_worker_pod=PodGenerator.deserialize_model_file(kube_config.pod_template_file),
)
pod_mutation_hook(pod)
api_client = ApiClient()
date_string = pod_generator.datetime_to_label_safe_datestring(execution_date)
yaml_file_name = f"{args.dag_id}_{ti.task_id}_{date_string}.yml"
os.makedirs(os.path.dirname(yaml_output_path + "/airflow_yaml_output/"), exist_ok=True)
with open(yaml_output_path + "/airflow_yaml_output/" + yaml_file_name, "w") as output:
sanitized_pod = api_client.sanitize_for_serialization(pod)
output.write(yaml.dump(sanitized_pod))
print(f"YAML output can be found at {yaml_output_path}/airflow_yaml_output/")
def generate_pod_yaml(args):
"""Generates yaml files for each task in the DAG. Used for testing output of KubernetesExecutor"""
from kubernetes.client.api_client import ApiClient
from airflow.executors.kubernetes_executor import AirflowKubernetesScheduler, KubeConfig
from airflow.kubernetes import pod_generator
from airflow.kubernetes.pod_generator import PodGenerator
from airflow.kubernetes.worker_configuration import WorkerConfiguration
from airflow.settings import pod_mutation_hook
execution_date = args.execution_date
dag = get_dag(subdir=args.subdir, dag_id=args.dag_id)
yaml_output_path = args.output_path
kube_config = KubeConfig()
for task in dag.tasks:
ti = TaskInstance(task, execution_date)
pod = PodGenerator.construct_pod(
dag_id=args.dag_id,
task_id=ti.task_id,
pod_id=AirflowKubernetesScheduler._create_pod_id( # pylint: disable=W0212
args.dag_id, ti.task_id),
try_number=ti.try_number,
kube_image=kube_config.kube_image,
date=ti.execution_date,
command=ti.command_as_list(),
pod_override_object=PodGenerator.from_obj(ti.executor_config),
worker_uuid="worker-config",
namespace=kube_config.executor_namespace,
base_worker_pod=WorkerConfiguration(
kube_config=kube_config).as_pod())
pod_mutation_hook(pod)
api_client = ApiClient()
date_string = pod_generator.datetime_to_label_safe_datestring(
execution_date)
yaml_file_name = f"{args.dag_id}_{ti.task_id}_{date_string}.yml"
os.makedirs(os.path.dirname(yaml_output_path +
"/airflow_yaml_output/"),
exist_ok=True)
with open(yaml_output_path + "/airflow_yaml_output/" + yaml_file_name,
"w") as output:
sanitized_pod = api_client.sanitize_for_serialization(pod)
output.write(yaml.dump(sanitized_pod))
print(
f"YAML output can be found at {yaml_output_path}/airflow_yaml_output/")
def _op_to_task(self, dag_id, image, op, node_map):
"""
Generate task given an operator inherited from dsl.ContainerOp.
:param airflow.models.BaseOperator op:
:param dict(Text, SdkNode) node_map:
:rtype: Tuple(base_tasks.SdkTask, SdkNode)
"""
interface_inputs = {}
interface_outputs = {}
input_mappings = {}
processed_args = None
# for key, val in six.iteritems(op.params):
# interface_inputs[key] = interface_model.Variable(
# _type_helpers.python_std_to_sdk_type(Types.String).to_flyte_literal_type(),
# ''
# )
#
# if param.op_name == '':
# binding = promise_common.Input(sdk_type=Types.String, name=param.name)
# else:
# binding = promise_common.NodeOutput(
# sdk_node=node_map[param.op_name],
# sdk_type=Types.String,
# var=param.name)
# input_mappings[param.name] = binding
#
# for param in op.outputs.values():
# interface_outputs[param.name] = interface_model.Variable(
# _type_helpers.python_std_to_sdk_type(Types.String).to_flyte_literal_type(),
# ''
# )
requests = []
if op.resources:
requests.append(
task_model.Resources.ResourceEntry(
task_model.Resources.ResourceName.Cpu, op.resources.cpus))
requests.append(
task_model.Resources.ResourceEntry(
task_model.Resources.ResourceName.Memory,
op.resources.ram))
requests.append(
task_model.Resources.ResourceEntry(
task_model.Resources.ResourceName.Gpu, op.resources.gpus))
requests.append(
task_model.Resources.ResourceEntry(
task_model.Resources.ResourceName.Storage,
op.resources.disk))
task_instance = TaskInstance(op, datetime.datetime.now())
command = task_instance.command_as_list(local=True,
mark_success=False,
ignore_all_deps=True,
ignore_depends_on_past=True,
ignore_task_deps=True,
ignore_ti_state=True,
pool=task_instance.pool,
pickle_id=dag_id,
cfg_path=None)
task = base_tasks.SdkTask(
op.task_id,
SingleStepTask,
"airflow_op",
task_model.TaskMetadata(
False,
task_model.RuntimeMetadata(
type=task_model.RuntimeMetadata.RuntimeType.Other,
version=airflow.version.version,
flavor='airflow'),
datetime.timedelta(seconds=0),
literals_model.RetryStrategy(0),
'1',
None,
),
interface_common.TypedInterface(inputs=interface_inputs,
outputs=interface_outputs),
custom=None,
container=task_model.Container(
image=image,
command=command,
args=[],
resources=task_model.Resources(limits=[], requests=requests),
env={},
config={},
))
return task, task(**input_mappings).assign_id_and_return(op.task_id)