def value_from_storage_dict(storage_dict, klass):
# Handle case where daemon_type used to be an enum (e.g. DaemonType.SCHEDULER)
return DaemonHeartbeat(
timestamp=storage_dict.get("timestamp"),
daemon_type=(storage_dict.get("daemon_type").value if isinstance(
storage_dict.get("daemon_type"), DaemonType) else
storage_dict.get("daemon_type")),
daemon_id=storage_dict.get("daemon_id"),
errors=[unpack_value(storage_dict.get("error"))
] # error was replaced with errors
if storage_dict.get("error") else unpack_value(
storage_dict.get("errors")),
)
def _execute_plan(self, execute_step_args_packed, executable_dict):
execute_step_args = unpack_value(
check.dict_param(
execute_step_args_packed,
"execute_step_args_packed",
))
check.inst_param(execute_step_args, "execute_step_args",
ExecuteStepArgs)
check.dict_param(executable_dict, "executable_dict")
instance = DagsterInstance.from_ref(execute_step_args.instance_ref)
pipeline = ReconstructablePipeline.from_dict(executable_dict)
retries = Retries.from_config(execute_step_args.retries_dict)
pipeline_run = instance.get_run_by_id(
execute_step_args.pipeline_run_id)
check.invariant(
pipeline_run,
"Could not load run {}".format(execute_step_args.pipeline_run_id))
step_keys_str = ", ".join(execute_step_args.step_keys_to_execute)
execution_plan = create_execution_plan(
pipeline,
pipeline_run.run_config,
mode=pipeline_run.mode,
step_keys_to_execute=execute_step_args.step_keys_to_execute,
)
engine_event = instance.report_engine_event(
"Executing steps {} in celery worker".format(step_keys_str),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_keys_str, "step_keys"),
EventMetadataEntry.text(self.request.hostname,
"Celery worker"),
],
marker_end=DELEGATE_MARKER,
),
CeleryExecutor,
step_key=execution_plan.step_key_for_single_step_plans(),
)
events = [engine_event]
for step_event in execute_plan_iterator(
execution_plan,
pipeline_run=pipeline_run,
run_config=pipeline_run.run_config,
instance=instance,
retries=retries,
):
events.append(step_event)
serialized_events = [
serialize_dagster_namedtuple(event) for event in events
]
return serialized_events
def from_dict(val):
check.dict_param(val, "val")
inst = unpack_value(val)
check.invariant(
isinstance(inst, ReconstructablePipeline),
"Deserialized object is not instance of ReconstructablePipeline, got {type}"
.format(type=type(inst)),
)
return inst
def from_dict(val):
check.dict_param(val, 'val')
inst = unpack_value(val)
check.invariant(
isinstance(inst, InterProcessExecutablePipeline),
'Deserialized object is not instance of InterProcessExecutablePipeline, got {type}'
.format(type=type(inst)),
)
return inst
def _execute_step_k8s_job(
self,
execute_step_args_packed,
job_config_dict,
job_namespace,
load_incluster_config,
user_defined_k8s_config_dict=None,
kubeconfig_file=None,
):
"""Run step execution in a K8s job pod."""
execute_step_args = unpack_value(
check.dict_param(
execute_step_args_packed,
"execute_step_args_packed",
))
check.inst_param(execute_step_args, "execute_step_args",
ExecuteStepArgs)
check.invariant(
len(execute_step_args.step_keys_to_execute) == 1,
"Celery K8s task executor can only execute 1 step at a time",
)
# Celery will serialize this as a list
job_config = DagsterK8sJobConfig.from_dict(job_config_dict)
check.inst_param(job_config, "job_config", DagsterK8sJobConfig)
check.str_param(job_namespace, "job_namespace")
check.bool_param(load_incluster_config, "load_incluster_config")
user_defined_k8s_config = UserDefinedDagsterK8sConfig.from_dict(
user_defined_k8s_config_dict)
check.opt_inst_param(
user_defined_k8s_config,
"user_defined_k8s_config",
UserDefinedDagsterK8sConfig,
)
check.opt_str_param(kubeconfig_file, "kubeconfig_file")
# For when launched via DinD or running the cluster
if load_incluster_config:
kubernetes.config.load_incluster_config()
else:
kubernetes.config.load_kube_config(kubeconfig_file)
instance = DagsterInstance.from_ref(execute_step_args.instance_ref)
pipeline_run = instance.get_run_by_id(
execute_step_args.pipeline_run_id)
check.inst(
pipeline_run,
PipelineRun,
"Could not load run {}".format(execute_step_args.pipeline_run_id),
)
step_key = execute_step_args.step_keys_to_execute[0]
celery_worker_name = self.request.hostname
celery_pod_name = os.environ.get("HOSTNAME")
instance.report_engine_event(
"Task for step {step_key} picked up by Celery".format(
step_key=step_key),
pipeline_run,
EngineEventData([
EventMetadataEntry.text(celery_worker_name,
"Celery worker name"),
EventMetadataEntry.text(celery_pod_name,
"Celery worker Kubernetes Pod name"),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
if pipeline_run.status != PipelineRunStatus.STARTED:
instance.report_engine_event(
"Not scheduling step because pipeline run status is not STARTED",
pipeline_run,
EngineEventData([
EventMetadataEntry.text(step_key, "Step key"),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
return []
# Ensure we stay below k8s name length limits
k8s_name_key = get_k8s_job_name(execute_step_args.pipeline_run_id,
step_key)
retry_state = execute_step_args.known_state.get_retry_state()
if retry_state.get_attempt_count(step_key):
attempt_number = retry_state.get_attempt_count(step_key)
job_name = "dagster-job-%s-%d" % (k8s_name_key, attempt_number)
pod_name = "dagster-job-%s-%d" % (k8s_name_key, attempt_number)
else:
job_name = "dagster-job-%s" % (k8s_name_key)
pod_name = "dagster-job-%s" % (k8s_name_key)
input_json = serialize_dagster_namedtuple(execute_step_args)
args = ["dagster", "api", "execute_step", input_json]
job = construct_dagster_k8s_job(job_config, args, job_name,
user_defined_k8s_config, pod_name)
# Running list of events generated from this task execution
events = []
# Post event for starting execution
job_name = job.metadata.name
engine_event = instance.report_engine_event(
"Executing step {} in Kubernetes job {}".format(
step_key, job_name),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
EventMetadataEntry.text(job_name, "Kubernetes Job name"),
EventMetadataEntry.text(job_config.job_image, "Job image"),
EventMetadataEntry.text(job_config.image_pull_policy,
"Image pull policy"),
EventMetadataEntry.text(str(job_config.image_pull_secrets),
"Image pull secrets"),
EventMetadataEntry.text(
str(job_config.service_account_name),
"Service account name"),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
# validated above that step_keys is length 1, and it is not possible to use ETH or
# execution plan in this function (Celery K8s workers should not access to user code)
step_key=step_key,
)
events.append(engine_event)
try:
kubernetes.client.BatchV1Api().create_namespaced_job(
body=job, namespace=job_namespace)
except kubernetes.client.rest.ApiException as e:
if e.reason == "Conflict":
# There is an existing job with the same name so proceed and see if the existing job succeeded
instance.report_engine_event(
"Did not create Kubernetes job {} for step {} since job name already "
"exists, proceeding with existing job.".format(
job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
EventMetadataEntry.text(job_name,
"Kubernetes Job name"),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
step_key=step_key,
)
else:
instance.report_engine_event(
"Encountered unexpected error while creating Kubernetes job {} for step {}, "
"exiting.".format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
],
error=serializable_error_info_from_exc_info(
sys.exc_info()),
),
CeleryK8sJobExecutor,
step_key=step_key,
)
return []
try:
wait_for_job_success(
job_name=job_name,
namespace=job_namespace,
instance=instance,
run_id=execute_step_args.pipeline_run_id,
)
except (DagsterK8sError, DagsterK8sTimeoutError) as err:
step_failure_event = construct_step_failure_event_and_handle(
pipeline_run, step_key, err, instance=instance)
events.append(step_failure_event)
except DagsterK8sPipelineStatusException:
instance.report_engine_event(
"Terminating Kubernetes Job because pipeline run status is not STARTED",
pipeline_run,
EngineEventData([
EventMetadataEntry.text(step_key, "Step key"),
EventMetadataEntry.text(job_name, "Kubernetes Job name"),
EventMetadataEntry.text(job_namespace,
"Kubernetes Job namespace"),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
delete_job(job_name=job_name, namespace=job_namespace)
return []
except (
DagsterK8sUnrecoverableAPIError,
DagsterK8sAPIRetryLimitExceeded,
# We shouldn't see unwrapped APIExceptions anymore, as they should all be wrapped in
# a retry boundary. We still catch it here just in case we missed one so that we can
# report it to the event log
kubernetes.client.rest.ApiException,
) as err:
instance.report_engine_event(
"Encountered unexpected error while waiting on Kubernetes job {} for step {}, "
"exiting.".format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
],
error=serializable_error_info_from_exc_info(
sys.exc_info()),
),
CeleryK8sJobExecutor,
step_key=step_key,
)
return []
try:
pod_names = get_pod_names_in_job(job_name, namespace=job_namespace)
except kubernetes.client.rest.ApiException as e:
instance.report_engine_event(
"Encountered unexpected error retreiving Pods for Kubernetes job {} for step {}, "
"exiting.".format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
],
error=serializable_error_info_from_exc_info(
sys.exc_info()),
),
CeleryK8sJobExecutor,
step_key=step_key,
)
return []
# Post engine event for log retrieval
engine_event = instance.report_engine_event(
"Retrieving logs from Kubernetes Job pods",
pipeline_run,
EngineEventData(
[EventMetadataEntry.text("\n".join(pod_names), "Pod names")]),
CeleryK8sJobExecutor,
step_key=step_key,
)
events.append(engine_event)
logs = []
for pod_name in pod_names:
try:
raw_logs = retrieve_pod_logs(pod_name, namespace=job_namespace)
logs += raw_logs.split("\n")
except kubernetes.client.rest.ApiException as e:
instance.report_engine_event(
"Encountered unexpected error while fetching pod logs for Kubernetes job {}, "
"Pod name {} for step {}. Will attempt to continue with other pods."
.format(job_name, pod_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step key"),
],
error=serializable_error_info_from_exc_info(
sys.exc_info()),
),
CeleryK8sJobExecutor,
step_key=step_key,
)
events += filter_dagster_events_from_pod_logs(logs)
serialized_events = [
serialize_dagster_namedtuple(event) for event in events
]
return serialized_events
def _execute_step_k8s_job(
self,
instance_ref_dict,
step_keys,
run_config,
mode,
repo_name,
repo_location_name,
run_id,
job_config_dict,
job_namespace,
load_incluster_config,
retries_dict,
pipeline_origin_packed,
user_defined_k8s_config_dict=None,
kubeconfig_file=None,
):
"""Run step execution in a K8s job pod.
"""
check.dict_param(instance_ref_dict, "instance_ref_dict")
check.list_param(step_keys, "step_keys", of_type=str)
check.invariant(
len(step_keys) == 1, "Celery K8s task executor can only execute 1 step at a time"
)
check.dict_param(run_config, "run_config")
check.str_param(mode, "mode")
check.str_param(repo_name, "repo_name")
check.str_param(repo_location_name, "repo_location_name")
check.str_param(run_id, "run_id")
# Celery will serialize this as a list
job_config = DagsterK8sJobConfig.from_dict(job_config_dict)
check.inst_param(job_config, "job_config", DagsterK8sJobConfig)
check.str_param(job_namespace, "job_namespace")
check.bool_param(load_incluster_config, "load_incluster_config")
check.dict_param(retries_dict, "retries_dict")
pipeline_origin = unpack_value(
check.dict_param(
pipeline_origin_packed, "pipeline_origin_packed"
) # TODO: make part of args
)
check.inst(pipeline_origin, PipelineOrigin)
user_defined_k8s_config = UserDefinedDagsterK8sConfig.from_dict(
user_defined_k8s_config_dict
)
check.opt_inst_param(
user_defined_k8s_config, "user_defined_k8s_config", UserDefinedDagsterK8sConfig,
)
check.opt_str_param(kubeconfig_file, "kubeconfig_file")
# For when launched via DinD or running the cluster
if load_incluster_config:
kubernetes.config.load_incluster_config()
else:
kubernetes.config.load_kube_config(kubeconfig_file)
instance_ref = InstanceRef.from_dict(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
pipeline_run = instance.get_run_by_id(run_id)
check.invariant(pipeline_run, "Could not load run {}".format(run_id))
step_key = step_keys[0]
celery_worker_name = self.request.hostname
celery_pod_name = os.environ.get("HOSTNAME")
instance.report_engine_event(
"Task for step {step_key} picked up by Celery".format(step_key=step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(celery_worker_name, "Celery worker name"),
EventMetadataEntry.text(celery_pod_name, "Celery worker Kubernetes Pod name"),
]
),
CeleryK8sJobExecutor,
step_key=step_key,
)
if pipeline_run.status != PipelineRunStatus.STARTED:
instance.report_engine_event(
"Not scheduling step because pipeline run status is not STARTED",
pipeline_run,
EngineEventData([EventMetadataEntry.text(step_key, "Step keys"),]),
CeleryK8sJobExecutor,
step_key=step_key,
)
return
# Ensure we stay below k8s name length limits
k8s_name_key = get_k8s_job_name(run_id, step_key)
retries = Retries.from_config(retries_dict)
if retries.get_attempt_count(step_key):
attempt_number = retries.get_attempt_count(step_key)
job_name = "dagster-job-%s-%d" % (k8s_name_key, attempt_number)
pod_name = "dagster-job-%s-%d" % (k8s_name_key, attempt_number)
else:
job_name = "dagster-job-%s" % (k8s_name_key)
pod_name = "dagster-job-%s" % (k8s_name_key)
input_json = serialize_dagster_namedtuple(
ExecuteStepArgs(
pipeline_origin=pipeline_origin,
pipeline_run_id=run_id,
instance_ref=None,
mode=mode,
step_keys_to_execute=step_keys,
run_config=run_config,
retries_dict=retries_dict,
)
)
command = ["dagster"]
args = ["api", "execute_step_with_structured_logs", input_json]
job = construct_dagster_k8s_job(
job_config, command, args, job_name, user_defined_k8s_config, pod_name
)
# Running list of events generated from this task execution
events = []
# Post event for starting execution
job_name = job.metadata.name
engine_event = instance.report_engine_event(
"Executing step {} in Kubernetes job {}".format(step_key, job_name),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step keys"),
EventMetadataEntry.text(job_name, "Kubernetes Job name"),
EventMetadataEntry.text(pod_name, "Kubernetes Pod name"),
EventMetadataEntry.text(job_config.job_image, "Job image"),
EventMetadataEntry.text(job_config.image_pull_policy, "Image pull policy"),
EventMetadataEntry.text(
str(job_config.image_pull_secrets), "Image pull secrets"
),
EventMetadataEntry.text(
str(job_config.service_account_name), "Service account name"
),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
# validated above that step_keys is length 1, and it is not possible to use ETH or
# execution plan in this function (Celery K8s workers should not access to user code)
step_key=step_key,
)
events.append(engine_event)
try:
kubernetes.client.BatchV1Api().create_namespaced_job(body=job, namespace=job_namespace)
except kubernetes.client.rest.ApiException as e:
if e.reason == "Conflict":
# There is an existing job with the same name so do not procede.
instance.report_engine_event(
"Did not create Kubernetes job {} for step {} since job name already "
"exists, exiting.".format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step keys"),
EventMetadataEntry.text(job_name, "Kubernetes Job name"),
EventMetadataEntry.text(pod_name, "Kubernetes Pod name"),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
step_key=step_key,
)
else:
instance.report_engine_event(
"Encountered unexpected error while creating Kubernetes job {} for step {}, "
"exiting.".format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step keys"),
EventMetadataEntry.text(e, "Error"),
]
),
CeleryK8sJobExecutor,
step_key=step_key,
)
return
try:
wait_for_job_success(
job_name=job_name, namespace=job_namespace, instance=instance, run_id=run_id,
)
except DagsterK8sPipelineStatusException:
instance.report_engine_event(
"Terminating Kubernetes Job because pipeline run status is not STARTED",
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, "Step keys"),
EventMetadataEntry.text(job_name, "Kubernetes Job name"),
EventMetadataEntry.text(job_namespace, "Kubernetes Job namespace"),
]
),
CeleryK8sJobExecutor,
step_key=step_key,
)
delete_job(job_name=job_name, namespace=job_namespace)
return
pod_names = get_pod_names_in_job(job_name, namespace=job_namespace)
# Post engine event for log retrieval
engine_event = instance.report_engine_event(
"Retrieving logs from Kubernetes Job pods",
pipeline_run,
EngineEventData([EventMetadataEntry.text("\n".join(pod_names), "Pod names")]),
CeleryK8sJobExecutor,
step_key=step_key,
)
events.append(engine_event)
logs = []
for pod_name in pod_names:
raw_logs = retrieve_pod_logs(pod_name, namespace=job_namespace)
logs += raw_logs.split("\n")
events += filter_dagster_events_from_pod_logs(logs)
serialized_events = [serialize_dagster_namedtuple(event) for event in events]
return serialized_events
def reconstitute_pipeline_context(
self,
output_log_path=None,
marshal_dir=None,
run_config=None,
executable_dict=None,
pipeline_run_dict=None,
solid_handle_kwargs=None,
instance_ref_dict=None,
):
"""Reconstitutes a context for dagstermill-managed execution.
You'll see this function called to reconstruct a pipeline context within the ``injected
parameters`` cell of a dagstermill output notebook. Users should not call this function
interactively except when debugging output notebooks.
Use :func:`dagstermill.get_context` in the ``parameters`` cell of your notebook to define a
context for interactive exploration and development. This call will be replaced by one to
:func:`dagstermill.reconstitute_pipeline_context` when the notebook is executed by
dagstermill.
"""
check.opt_str_param(output_log_path, "output_log_path")
check.opt_str_param(marshal_dir, "marshal_dir")
run_config = check.opt_dict_param(run_config,
"run_config",
key_type=str)
check.dict_param(pipeline_run_dict, "pipeline_run_dict")
check.dict_param(executable_dict, "executable_dict")
check.dict_param(solid_handle_kwargs, "solid_handle_kwargs")
check.dict_param(instance_ref_dict, "instance_ref_dict")
pipeline = ReconstructablePipeline.from_dict(executable_dict)
pipeline_def = pipeline.get_definition()
try:
instance_ref = unpack_value(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
except Exception as err: # pylint: disable=broad-except
raise DagstermillError(
"Error when attempting to resolve DagsterInstance from serialized InstanceRef"
) from err
pipeline_run = unpack_value(pipeline_run_dict)
solid_handle = SolidHandle.from_dict(solid_handle_kwargs)
solid_def = pipeline_def.get_solid(solid_handle).definition
self.marshal_dir = marshal_dir
self.in_pipeline = True
self.solid_def = solid_def
self.pipeline = pipeline
environment_config = EnvironmentConfig.build(pipeline_def,
run_config,
mode=pipeline_run.mode)
execution_plan = ExecutionPlan.build(
self.pipeline,
environment_config,
step_keys_to_execute=pipeline_run.step_keys_to_execute,
)
with scoped_pipeline_context(
execution_plan,
pipeline,
run_config,
pipeline_run,
instance,
scoped_resources_builder_cm=self._setup_resources,
# Set this flag even though we're not in test for clearer error reporting
raise_on_error=True,
) as pipeline_context:
self.context = DagstermillRuntimeExecutionContext(
pipeline_context=pipeline_context,
pipeline_def=pipeline_def,
solid_config=run_config.get("solids",
{}).get(solid_def.name,
{}).get("config"),
resource_keys_to_init=get_required_resource_keys_to_init(
execution_plan,
pipeline_def,
environment_config,
pipeline_context.intermediate_storage_def,
),
solid_name=solid_def.name,
)
return self.context
def _execute_step_k8s_job(
_self,
instance_ref_dict,
step_keys,
run_config,
mode,
repo_name,
repo_location_name,
run_id,
job_config_dict,
job_namespace,
load_incluster_config,
retries_dict,
pipeline_origin_packed,
resources=None,
kubeconfig_file=None,
):
'''Run step execution in a K8s job pod.
'''
check.dict_param(instance_ref_dict, 'instance_ref_dict')
check.list_param(step_keys, 'step_keys', of_type=str)
check.invariant(
len(step_keys) == 1,
'Celery K8s task executor can only execute 1 step at a time')
check.dict_param(run_config, 'run_config')
check.str_param(mode, 'mode')
check.str_param(repo_name, 'repo_name')
check.str_param(repo_location_name, 'repo_location_name')
check.str_param(run_id, 'run_id')
# Celery will serialize this as a list
job_config = DagsterK8sJobConfig.from_dict(job_config_dict)
check.inst_param(job_config, 'job_config', DagsterK8sJobConfig)
check.str_param(job_namespace, 'job_namespace')
check.bool_param(load_incluster_config, 'load_incluster_config')
check.dict_param(retries_dict, 'retries_dict')
pipeline_origin = unpack_value(
check.dict_param(
pipeline_origin_packed,
'pipeline_origin_packed') # TODO: make part of args
)
check.inst(pipeline_origin, PipelineOrigin)
check.opt_dict_param(resources,
'resources',
key_type=str,
value_type=dict)
check.opt_str_param(kubeconfig_file, 'kubeconfig_file')
# For when launched via DinD or running the cluster
if load_incluster_config:
kubernetes.config.load_incluster_config()
else:
kubernetes.config.load_kube_config(kubeconfig_file)
instance_ref = InstanceRef.from_dict(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
pipeline_run = instance.get_run_by_id(run_id)
check.invariant(pipeline_run, 'Could not load run {}'.format(run_id))
step_key = step_keys[0]
if pipeline_run.status != PipelineRunStatus.STARTED:
instance.report_engine_event(
'Not scheduling step because pipeline run status is not STARTED',
pipeline_run,
EngineEventData([
EventMetadataEntry.text(step_key, 'Step keys'),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
return
# Ensure we stay below k8s name length limits
k8s_name_key = get_k8s_job_name(run_id, step_key)
retries = Retries.from_config(retries_dict)
if retries.get_attempt_count(step_key):
attempt_number = retries.get_attempt_count(step_key)
job_name = 'dagster-job-%s-%d' % (k8s_name_key, attempt_number)
pod_name = 'dagster-job-%s-%d' % (k8s_name_key, attempt_number)
else:
job_name = 'dagster-job-%s' % (k8s_name_key)
pod_name = 'dagster-job-%s' % (k8s_name_key)
input_json = serialize_dagster_namedtuple(
ExecuteStepArgs(
pipeline_origin=pipeline_origin,
pipeline_run_id=run_id,
instance_ref=None,
mode=mode,
step_keys_to_execute=step_keys,
run_config=run_config,
retries_dict=retries_dict,
))
command = ['dagster']
args = ['api', 'execute_step_with_structured_logs', input_json]
job = construct_dagster_k8s_job(job_config, command, args, job_name,
resources, pod_name)
# Running list of events generated from this task execution
events = []
# Post event for starting execution
job_name = job.metadata.name
engine_event = instance.report_engine_event(
'Executing step {} in Kubernetes job {}'.format(
step_key, job_name),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, 'Step keys'),
EventMetadataEntry.text(job_name, 'Kubernetes Job name'),
EventMetadataEntry.text(pod_name, 'Kubernetes Pod name'),
EventMetadataEntry.text(job_config.job_image, 'Job image'),
EventMetadataEntry.text(job_config.image_pull_policy,
'Image pull policy'),
EventMetadataEntry.text(str(job_config.image_pull_secrets),
'Image pull secrets'),
EventMetadataEntry.text(
str(job_config.service_account_name),
'Service account name'),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
# validated above that step_keys is length 1, and it is not possible to use ETH or
# execution plan in this function (Celery K8s workers should not access to user code)
step_key=step_key,
)
events.append(engine_event)
try:
kubernetes.client.BatchV1Api().create_namespaced_job(
body=job, namespace=job_namespace)
except kubernetes.client.rest.ApiException as e:
if e.reason == 'Conflict':
# There is an existing job with the same name so do not procede.
instance.report_engine_event(
'Did not create Kubernetes job {} for step {} since job name already '
'exists, exiting.'.format(job_name, step_key),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_key, 'Step keys'),
EventMetadataEntry.text(job_name,
'Kubernetes Job name'),
EventMetadataEntry.text(pod_name,
'Kubernetes Pod name'),
],
marker_end=DELEGATE_MARKER,
),
CeleryK8sJobExecutor,
step_key=step_key,
)
else:
instance.report_engine_event(
'Encountered unexpected error while creating Kubernetes job {} for step {}, '
'exiting.'.format(job_name, step_key),
pipeline_run,
EngineEventData([
EventMetadataEntry.text(step_key, 'Step keys'),
EventMetadataEntry.text(e, 'Error'),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
return
try:
wait_for_job_success(
job_name=job_name,
namespace=job_namespace,
instance=instance,
run_id=run_id,
)
except DagsterK8sPipelineStatusException:
instance.report_engine_event(
'Terminating Kubernetes Job because pipeline run status is not STARTED',
pipeline_run,
EngineEventData([
EventMetadataEntry.text(step_key, 'Step keys'),
EventMetadataEntry.text(job_name, 'Kubernetes Job name'),
EventMetadataEntry.text(job_namespace,
'Kubernetes Job namespace'),
]),
CeleryK8sJobExecutor,
step_key=step_key,
)
delete_job(job_name=job_name, namespace=job_namespace)
return
pod_names = get_pod_names_in_job(job_name, namespace=job_namespace)
# Post engine event for log retrieval
engine_event = instance.report_engine_event(
'Retrieving logs from Kubernetes Job pods',
pipeline_run,
EngineEventData(
[EventMetadataEntry.text('\n'.join(pod_names), 'Pod names')]),
CeleryK8sJobExecutor,
step_key=step_key,
)
events.append(engine_event)
logs = []
for pod_name in pod_names:
raw_logs = retrieve_pod_logs(pod_name, namespace=job_namespace)
logs += raw_logs.split('\n')
events += filter_dagster_events_from_pod_logs(logs)
serialized_events = [
serialize_dagster_namedtuple(event) for event in events
]
return serialized_events
def _execute_step_docker(
self,
execute_step_args_packed,
docker_config,
):
"""Run step execution in a Docker container."""
execute_step_args = unpack_value(
check.dict_param(
execute_step_args_packed,
"execute_step_args_packed",
))
check.inst_param(execute_step_args, "execute_step_args",
ExecuteStepArgs)
check.dict_param(docker_config, "docker_config")
instance = DagsterInstance.from_ref(execute_step_args.instance_ref)
pipeline_run = instance.get_run_by_id(
execute_step_args.pipeline_run_id)
check.inst(
pipeline_run,
PipelineRun,
"Could not load run {}".format(execute_step_args.pipeline_run_id),
)
step_keys_str = ", ".join(execute_step_args.step_keys_to_execute)
input_json = serialize_dagster_namedtuple(execute_step_args)
command = "dagster api execute_step {}".format(json.dumps(input_json))
docker_image = (docker_config["image"]
if docker_config.get("image") else execute_step_args.
pipeline_origin.repository_origin.container_image)
if not docker_image:
raise Exception(
"No docker image specified by either the job or the repository"
)
client = docker.client.from_env()
if docker_config.get("registry"):
client.login(
registry=docker_config["registry"]["url"],
username=docker_config["registry"]["username"],
password=docker_config["registry"]["password"],
)
# Post event for starting execution
engine_event = instance.report_engine_event(
"Executing steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData(
[
MetadataEntry.text(step_keys_str, "Step keys"),
MetadataEntry.text(docker_image, "Image"),
MetadataEntry.text(self.request.hostname, "Celery worker"),
],
marker_end=DELEGATE_MARKER,
),
CeleryDockerExecutor,
step_key=execute_step_args.step_keys_to_execute[0],
)
serialized_events = [serialize_dagster_namedtuple(engine_event)]
docker_env = {}
if docker_config.get("env_vars"):
docker_env = {
env_name: os.getenv(env_name)
for env_name in docker_config["env_vars"]
}
try:
docker_response = client.containers.run(
docker_image,
command=command,
detach=False,
auto_remove=True,
# pass through this worker's environment for things like AWS creds etc.
environment=docker_env,
network=docker_config.get("network", None),
)
res = docker_response.decode("utf-8")
except docker.errors.ContainerError as err:
instance.report_engine_event(
"Failed to run steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData([
MetadataEntry.text(docker_image, "Job image"),
MetadataEntry.text(err.stderr, "Docker stderr"),
], ),
CeleryDockerExecutor,
step_key=execute_step_args.step_keys_to_execute[0],
)
raise
else:
if res is None:
raise Exception(
"No response from execute_step in CeleryDockerExecutor")
serialized_events += [event for event in res.split("\n") if event]
return serialized_events
def reconstitute_pipeline_context(
self,
output_log_path=None,
marshal_dir=None,
environment_dict=None,
handle_kwargs=None,
pipeline_run_dict=None,
solid_subset=None,
solid_handle_kwargs=None,
instance_ref_dict=None,
):
'''Reconstitutes a context for dagstermill-managed execution.
You'll see this function called to reconstruct a pipeline context within the ``injected
parameters`` cell of a dagstermill output notebook. Users should not call this function
interactively except when debugging output notebooks.
Use :func:`dagstermill.get_context` in the ``parameters`` cell of your notebook to define a
context for interactive exploration and development. This call will be replaced by one to
:func:`dagstermill.reconstitute_pipeline_context` when the notebook is executed by
dagstermill.
'''
check.opt_str_param(output_log_path, 'output_log_path')
check.opt_str_param(marshal_dir, 'marshal_dir')
environment_dict = check.opt_dict_param(environment_dict, 'environment_dict', key_type=str)
check.dict_param(pipeline_run_dict, 'pipeline_run_dict')
check.dict_param(handle_kwargs, 'handle_kwargs')
check.opt_list_param(solid_subset, 'solid_subset', of_type=str)
check.dict_param(solid_handle_kwargs, 'solid_handle_kwargs')
check.dict_param(instance_ref_dict, 'instance_ref_dict')
try:
handle = load_handle.handle_for_pipeline_cli_args(
handle_kwargs, use_default_repository_yaml=False
)
except (check.CheckError, load_handle.UsageError) as err:
six.raise_from(
DagstermillError(
'Cannot invoke a dagstermill solid from an in-memory pipeline that was not loaded '
'from an ExecutionTargetHandle. Run this pipeline using dagit, the dagster CLI, '
'through dagster-graphql, or in-memory after loading it through an '
'ExecutionTargetHandle.'
),
err,
)
try:
instance_ref = unpack_value(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
except Exception as err: # pylint: disable=broad-except
six.raise_from(
DagstermillError(
'Error when attempting to resolve DagsterInstance from serialized InstanceRef'
),
err,
)
pipeline_def = check.inst_param(
handle.build_pipeline_definition(),
'pipeline_def (from handle {handle_dict})'.format(handle_dict=handle.data._asdict()),
PipelineDefinition,
).build_sub_pipeline(solid_subset)
solid_handle = SolidHandle.from_dict(solid_handle_kwargs)
solid_def = pipeline_def.get_solid(solid_handle).definition
pipeline_run = unpack_value(pipeline_run_dict)
self.marshal_dir = marshal_dir
self.in_pipeline = True
self.solid_def = solid_def
self.pipeline_def = pipeline_def
execution_plan = create_execution_plan(self.pipeline_def, environment_dict, pipeline_run)
with scoped_pipeline_context(
self.pipeline_def,
environment_dict,
pipeline_run,
instance,
execution_plan,
scoped_resources_builder_cm=self._setup_resources,
) as pipeline_context:
self.context = DagstermillRuntimeExecutionContext(
pipeline_context=pipeline_context,
solid_config=None,
resource_keys_to_init=get_required_resource_keys_to_init(
execution_plan, pipeline_context.system_storage_def
),
)
return self.context
def _execute_step_docker(
self,
instance_ref_dict,
step_keys,
run_config,
mode,
repo_name,
run_id,
docker_config,
pipeline_origin_packed,
retries_dict,
):
"""Run step execution in a Docker container.
"""
check.dict_param(instance_ref_dict, "instance_ref_dict")
check.list_param(step_keys, "step_keys", of_type=str)
check.dict_param(run_config, "run_config")
check.str_param(mode, "mode")
check.str_param(repo_name, "repo_name")
check.str_param(run_id, "run_id")
check.dict_param(docker_config, "docker_config")
pipeline_origin = unpack_value(
check.dict_param(pipeline_origin_packed, "pipeline_origin_packed"))
check.dict_param(retries_dict, "retries_dict")
instance_ref = InstanceRef.from_dict(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
pipeline_run = instance.get_run_by_id(run_id)
check.invariant(pipeline_run, "Could not load run {}".format(run_id))
step_keys_str = ", ".join(step_keys)
input_json = serialize_dagster_namedtuple(
ExecuteStepArgs(
pipeline_origin=pipeline_origin,
pipeline_run_id=run_id,
instance_ref=instance_ref,
mode=mode,
step_keys_to_execute=step_keys,
run_config=run_config,
retries_dict=retries_dict,
))
command = "dagster api execute_step_with_structured_logs {}".format(
json.dumps(input_json))
docker_image = docker_config["image"]
client = docker.client.from_env()
if docker_config.get("registry"):
client.login(
registry=docker_config["registry"]["url"],
username=docker_config["registry"]["username"],
password=docker_config["registry"]["password"],
)
# Post event for starting execution
engine_event = instance.report_engine_event(
"Executing steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData(
[
EventMetadataEntry.text(step_keys_str, "Step keys"),
EventMetadataEntry.text(docker_image, "Image"),
EventMetadataEntry.text(self.request.hostname,
"Celery worker"),
],
marker_end=DELEGATE_MARKER,
),
CeleryDockerExecutor,
step_key=step_keys[0],
)
serialized_events = [serialize_dagster_namedtuple(engine_event)]
docker_env = {}
if docker_config.get("env_vars"):
docker_env = {
env_name: os.getenv(env_name)
for env_name in docker_config["env_vars"]
}
try:
docker_response = client.containers.run(
docker_image,
command=command,
detach=False,
auto_remove=True,
# pass through this worker's environment for things like AWS creds etc.
environment=docker_env,
network=docker_config.get("network", None),
)
res = docker_response.decode("utf-8")
except docker.errors.ContainerError as err:
instance.report_engine_event(
"Failed to run steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData([
EventMetadataEntry.text(docker_image, "Job image"),
EventMetadataEntry.text(err.stderr, "Docker stderr"),
], ),
CeleryDockerExecutor,
step_key=step_keys[0],
)
raise
else:
if res is None:
raise Exception(
"No response from execute_step_with_structured_logs in CeleryDockerExecutor"
)
serialized_events += [event for event in res.split("\n") if event]
return serialized_events
def reconstitute_pipeline_context(
self,
output_log_path=None,
marshal_dir=None,
environment_dict=None,
executable_dict=None,
pipeline_run_dict=None,
solid_handle_kwargs=None,
instance_ref_dict=None,
):
'''Reconstitutes a context for dagstermill-managed execution.
You'll see this function called to reconstruct a pipeline context within the ``injected
parameters`` cell of a dagstermill output notebook. Users should not call this function
interactively except when debugging output notebooks.
Use :func:`dagstermill.get_context` in the ``parameters`` cell of your notebook to define a
context for interactive exploration and development. This call will be replaced by one to
:func:`dagstermill.reconstitute_pipeline_context` when the notebook is executed by
dagstermill.
'''
check.opt_str_param(output_log_path, 'output_log_path')
check.opt_str_param(marshal_dir, 'marshal_dir')
environment_dict = check.opt_dict_param(environment_dict,
'environment_dict',
key_type=str)
check.dict_param(pipeline_run_dict, 'pipeline_run_dict')
check.dict_param(executable_dict, 'executable_dict')
check.dict_param(solid_handle_kwargs, 'solid_handle_kwargs')
check.dict_param(instance_ref_dict, 'instance_ref_dict')
pipeline = InterProcessExecutablePipeline.from_dict(executable_dict)
pipeline_def = pipeline.get_definition()
try:
instance_ref = unpack_value(instance_ref_dict)
instance = DagsterInstance.from_ref(instance_ref)
except Exception as err: # pylint: disable=broad-except
six.raise_from(
DagstermillError(
'Error when attempting to resolve DagsterInstance from serialized InstanceRef'
),
err,
)
pipeline_run = unpack_value(pipeline_run_dict)
solid_handle = SolidHandle.from_dict(solid_handle_kwargs)
solid_def = pipeline_def.get_solid(solid_handle).definition
self.marshal_dir = marshal_dir
self.in_pipeline = True
self.solid_def = solid_def
self.pipeline = pipeline
execution_plan = create_execution_plan(
self.pipeline,
environment_dict,
mode=pipeline_run.mode,
step_keys_to_execute=pipeline_run.step_keys_to_execute,
)
with scoped_pipeline_context(
execution_plan,
environment_dict,
pipeline_run,
instance,
scoped_resources_builder_cm=self._setup_resources,
# Set this flag even though we're not in test for clearer error reporting
raise_on_error=True,
) as pipeline_context:
self.context = DagstermillRuntimeExecutionContext(
pipeline_context=pipeline_context,
solid_config=None,
resource_keys_to_init=get_required_resource_keys_to_init(
execution_plan, pipeline_context.system_storage_def),
)
return self.context
def _execute_step_docker(
self,
execute_step_args_packed,
docker_config,
):
"""Run step execution in a Docker container."""
execute_step_args = unpack_value(
check.dict_param(
execute_step_args_packed,
"execute_step_args_packed",
))
check.inst_param(execute_step_args, "execute_step_args",
ExecuteStepArgs)
check.dict_param(docker_config, "docker_config")
instance = DagsterInstance.from_ref(execute_step_args.instance_ref)
pipeline_run = instance.get_run_by_id(
execute_step_args.pipeline_run_id)
check.inst(
pipeline_run,
PipelineRun,
"Could not load run {}".format(execute_step_args.pipeline_run_id),
)
step_keys_str = ", ".join(execute_step_args.step_keys_to_execute)
input_json = serialize_dagster_namedtuple(execute_step_args)
command = "dagster api execute_step {}".format(json.dumps(input_json))
docker_image = (docker_config["image"]
if docker_config.get("image") else execute_step_args.
pipeline_origin.repository_origin.container_image)
if not docker_image:
raise Exception(
"No docker image specified by either the job or the repository"
)
client = docker.client.from_env()
if docker_config.get("registry"):
client.login(
registry=docker_config["registry"]["url"],
username=docker_config["registry"]["username"],
password=docker_config["registry"]["password"],
)
# Post event for starting execution
engine_event = instance.report_engine_event(
"Executing steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData(
[
MetadataEntry("Step keys", value=step_keys_str),
MetadataEntry("Image", value=docker_image),
MetadataEntry("Celery worker",
value=self.request.hostname),
],
marker_end=DELEGATE_MARKER,
),
CeleryDockerExecutor,
step_key=execute_step_args.step_keys_to_execute[0],
)
serialized_events = [serialize_dagster_namedtuple(engine_event)]
docker_env = {}
if docker_config.get("env_vars"):
docker_env = {
env_name: os.getenv(env_name)
for env_name in docker_config["env_vars"]
}
container_kwargs = check.opt_dict_param(
docker_config.get("container_kwargs"),
"container_kwargs",
key_type=str)
# set defaults for detach and auto_remove
container_kwargs["detach"] = container_kwargs.get("detach", False)
container_kwargs["auto_remove"] = container_kwargs.get(
"auto_remove", True)
# if environment variables are provided via container_kwargs, merge with env_vars
if container_kwargs.get("environment") is not None:
e_vars = container_kwargs.get("environment")
if isinstance(e_vars, dict):
docker_env.update(e_vars)
else:
for v in e_vars:
key, val = v.split("=")
docker_env[key] = val
del container_kwargs["environment"]
try:
docker_response = client.containers.run(
docker_image,
command=command,
# pass through this worker's environment for things like AWS creds etc.
environment=docker_env,
network=docker_config.get("network", None),
**container_kwargs,
)
res = docker_response.decode("utf-8")
except docker.errors.ContainerError as err:
entries = [MetadataEntry("Job image", value=docker_image)]
if err.stderr is not None:
entries.append(MetadataEntry("Docker stderr",
value=err.stderr))
instance.report_engine_event(
"Failed to run steps {} in Docker container {}".format(
step_keys_str, docker_image),
pipeline_run,
EngineEventData(entries),
CeleryDockerExecutor,
step_key=execute_step_args.step_keys_to_execute[0],
)
raise
else:
if res is None:
raise Exception(
"No response from execute_step in CeleryDockerExecutor")
serialized_events += [event for event in res.split("\n") if event]
return serialized_events