def create_schedule_definition(
self,
schedule_name,
cron_schedule,
partition_selector,
should_execute=None,
environment_vars=None,
execution_timezone=None,
):
"""Create a ScheduleDefinition from a PartitionSetDefinition.
Arguments:
schedule_name (str): The name of the schedule.
cron_schedule (str): A valid cron string for the schedule
partition_selector (Callable[ScheduleExecutionContext, PartitionSetDefinition],
Partition): Function that determines the partition to use at a given execution time.
For time-based partition sets, will likely be either `identity_partition_selector` or a
selector returned by `create_offset_partition_selector`.
should_execute (Optional[function]): Function that runs at schedule execution time that
determines whether a schedule should execute. Defaults to a function that always returns
``True``.
environment_vars (Optional[dict]): The environment variables to set for the schedule.
execution_timezone (Optional[str]): Timezone in which the schedule should run. Only works
with DagsterDaemonScheduler, and must be set when using that scheduler.
Returns:
ScheduleDefinition: The generated ScheduleDefinition for the partition selector
"""
check.str_param(schedule_name, "schedule_name")
check.str_param(cron_schedule, "cron_schedule")
check.opt_callable_param(should_execute, "should_execute")
check.opt_dict_param(environment_vars, "environment_vars", key_type=str, value_type=str)
check.callable_param(partition_selector, "partition_selector")
check.opt_str_param(execution_timezone, "execution_timezone")
def _should_execute_wrapper(context):
check.inst_param(context, "context", ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition or not selected_partition.name in self.get_partition_names():
return False
elif not should_execute:
return True
else:
return should_execute(context)
def _run_config_fn_wrapper(context):
check.inst_param(context, "context", ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition or not selected_partition.name in self.get_partition_names():
raise DagsterInvariantViolationError(
"The partition selection function `{selector}` did not return "
"a partition from PartitionSet {partition_set}".format(
selector=getattr(partition_selector, "__name__", repr(partition_selector)),
partition_set=self.name,
)
)
return self.run_config_for_partition(selected_partition)
def _tags_fn_wrapper(context):
check.inst_param(context, "context", ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition:
raise DagsterInvariantViolationError(
"The partition selection function `{selector}` did not return "
"a partition from PartitionSet {partition_set}".format(
selector=getattr(partition_selector, "__name__", repr(partition_selector)),
partition_set=self.name,
)
)
return self.tags_for_partition(selected_partition)
return PartitionScheduleDefinition(
name=schedule_name,
cron_schedule=cron_schedule,
pipeline_name=self.pipeline_name,
run_config_fn=_run_config_fn_wrapper,
tags_fn=_tags_fn_wrapper,
solid_selection=self.solid_selection,
mode=self.mode,
should_execute=_should_execute_wrapper,
environment_vars=environment_vars,
partition_set=self,
execution_timezone=execution_timezone,
)
def _create_solid_compute_wrapper(fn, input_defs, output_defs):
check.callable_param(fn, 'fn')
check.list_param(input_defs, 'input_defs', of_type=InputDefinition)
check.list_param(output_defs, 'output_defs', of_type=OutputDefinition)
input_names = [
input_def.name for input_def in input_defs if not input_def.runtime_type.is_nothing
]
@wraps(fn)
def compute(context, input_defs):
kwargs = {}
for input_name in input_names:
kwargs[input_name] = input_defs[input_name]
result = fn(context, **kwargs)
if inspect.isgenerator(result):
for item in result:
yield item
else:
if isinstance(result, (Materialization, ExpectationResult)):
raise DagsterInvariantViolationError(
(
'Error in solid {solid_name}: If you are returning a Materialization '
'or an ExpectationResult from solid you must yield them to avoid '
'ambiguity with an implied result from returning a value.'.format(
solid_name=context.solid.name
)
)
)
if isinstance(result, Output):
yield result
elif len(output_defs) == 1:
yield Output(value=result, output_name=output_defs[0].name)
elif result is not None:
if not output_defs:
raise DagsterInvariantViolationError(
(
'Error in solid {solid_name}: Unexpectedly returned output {result} '
'of type {type_}. Solid is explicitly defined to return no '
'results.'
).format(solid_name=context.solid.name, result=result, type_=type(result))
)
raise DagsterInvariantViolationError(
(
'Error in solid {solid_name}: Solid unexpectedly returned '
'output {result} of type {type_}. Should '
'be a generator, containing or yielding '
'{n_results} results: {{{expected_results}}}.'
).format(
solid_name=context.solid.name,
result=result,
type_=type(result),
n_results=len(output_defs),
expected_results=', '.join(
[
'\'{result_name}\': {runtime_type}'.format(
result_name=output_def.name,
runtime_type=output_def.runtime_type,
)
for output_def in output_defs
]
),
)
)
return compute
def __call__(self, fn):
check.callable_param(fn, 'fn')
if not self.name:
self.name = fn.__name__
input_defs = (
self.input_defs
if self.input_defs is not None
else infer_input_definitions_for_composite_solid(self.name, fn)
)
explicit_outputs = False
if self.output_defs is not None:
explicit_outputs = True
output_defs = self.output_defs
else:
explicit_outputs = has_explicit_return_type(fn)
output_defs = infer_output_definitions('@composite_solid', self.name, fn)
positional_inputs = validate_solid_fn(
'@composite_solid', self.name, fn, input_defs, exclude_nothing=False
)
kwargs = {input_def.name: InputMappingNode(input_def) for input_def in input_defs}
output = None
mapping = None
enter_composition(self.name, '@composite_solid')
try:
output = fn(**kwargs)
mapping = composite_mapping_from_output(output, output_defs, self.name)
finally:
context = exit_composition(mapping)
check.invariant(
context.name == self.name,
'Composition context stack desync: received context for '
'"{context.name}" expected "{self.name}"'.format(context=context, self=self),
)
# line up mappings in definition order
input_mappings = []
for defn in input_defs:
mappings = [
mapping
for mapping in context.input_mappings
if mapping.definition.name == defn.name
]
if len(mappings) == 0:
raise DagsterInvalidDefinitionError(
"@composite_solid '{solid_name}' has unmapped input '{input_name}'. "
"Remove it or pass it to the appropriate solid invocation.".format(
solid_name=self.name, input_name=defn.name
)
)
input_mappings += mappings
output_mappings = []
for defn in output_defs:
mapping = context.output_mapping_dict.get(defn.name)
if mapping is None:
# if we inferred output_defs we will be flexible and either take a mapping or not
if not explicit_outputs:
continue
raise DagsterInvalidDefinitionError(
"@composite_solid '{solid_name}' has unmapped output '{output_name}'. "
"Remove it or return a value from the appropriate solid invocation.".format(
solid_name=self.name, output_name=defn.name
)
)
output_mappings.append(mapping)
config_mapping = _get_validated_config_mapping(self.name, self.config, self.config_fn)
return CompositeSolidDefinition(
name=self.name,
input_mappings=input_mappings,
output_mappings=output_mappings,
dependencies=context.dependencies,
solid_defs=context.solid_defs,
description=self.description,
config_mapping=config_mapping,
positional_inputs=positional_inputs,
)
def __init__(self, resource_fn, config_field=None, description=None):
self._resource_fn = check.callable_param(resource_fn, 'resource_fn')
self._config_field = check_user_facing_opt_field_param(
config_field, 'config_field', 'of a ResourceDefinition or @resource'
)
self._description = check.opt_str_param(description, 'description')
def __init__(self, manager_fn, marks):
self.manager_fn = check.callable_param(manager_fn, "manager_fn")
self.marks = check.list_param(marks, "marks")
def create_offset_partition_selector(execution_time_to_partition_fn):
""" Utility function for supplying a partition selector when creating a schedule from a
partition set made of `datetime`s that assumes a fixed time offset between the partition
time and the time at which the schedule executes.
It's important to keep the cron string that's supplied to
`PartitionSetDefinition.create_schedule_definition` in sync with the offset that's
supplied to this function. For example, a schedule created from a partition set with
partitions for each day at midnight that fills in the partition for day N at day N+1 at
10:00AM would create the partition selector as follows:
.. code-block:: python
partition_set = PartitionSetDefinition(
name='hello_world_partition_set',
pipeline_name='hello_world_pipeline',
partition_fn= date_partition_range(
start=datetime.datetime(2021, 1, 1),
delta_range="days",
timezone="US/Central",
)
run_config_fn_for_partition=my_run_config_fn,
)
schedule_definition = partition_set.create_schedule_definition(
"daily_10am_schedule",
"0 10 * * *",
partition_selector=create_offset_partition_selector(lambda d: d.subtract(hours=10, days=1))
execution_timezone="US/Central",
)
Args:
execution_time_to_partition_fn (Callable[[datetime.datetime], datetime.datetime]): A
function that maps the execution time of the schedule to the partition time.
"""
check.callable_param(execution_time_to_partition_fn,
"execution_time_to_partition_fn")
def offset_partition_selector(context, partition_set_def):
check.inst_param(context, "context", ScheduleExecutionContext)
check.inst_param(partition_set_def, "partition_set_def",
PartitionSetDefinition)
if not context.scheduled_execution_time:
partitions = partition_set_def.get_partitions()
if not partitions:
return None
return partitions[-1]
partition_time = execution_time_to_partition_fn(
context.scheduled_execution_time)
for partition in reversed(
partition_set_def.get_partitions(
context.scheduled_execution_time)):
if partition.value.isoformat() == partition_time.isoformat():
return partition
if partition.value < partition_time:
break
return None
return offset_partition_selector
def create_schedule_definition(
self,
schedule_name,
cron_schedule,
should_execute=None,
partition_selector=last_partition,
environment_vars=None,
):
'''Create a ScheduleDefinition from a PartitionSetDefinition.
Arguments:
schedule_name (str): The name of the schedule.
cron_schedule (str): A valid cron string for the schedule
should_execute (Optional[function]): Function that runs at schedule execution time that
determines whether a schedule should execute. Defaults to a function that always returns
``True``.
partition_selector (Callable[PartitionSet], Partition): A partition selector for the
schedule
environment_vars (Optional[dict]): The environment variables to set for the schedule
Returns:
ScheduleDefinition -- The generated ScheduleDefinition for the partition selector
'''
check.str_param(schedule_name, 'schedule_name')
check.str_param(cron_schedule, 'cron_schedule')
check.opt_callable_param(should_execute, 'should_execute')
check.opt_dict_param(environment_vars, 'environment_vars', key_type=str, value_type=str)
check.callable_param(partition_selector, 'partition_selector')
def _should_execute_wrapper(context):
check.inst_param(context, 'context', ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition:
return False
elif not should_execute:
return True
else:
return should_execute(context)
def _environment_dict_fn_wrapper(context):
check.inst_param(context, 'context', ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition:
raise DagsterInvariantViolationError(
"The partition selection function `{selector}` did not return "
"a partition from PartitionSet {partition_set}".format(
selector=getattr(partition_selector, '__name__', repr(partition_selector)),
partition_set=self.name,
)
)
return self.environment_dict_for_partition(selected_partition)
def _tags_fn_wrapper(context):
check.inst_param(context, 'context', ScheduleExecutionContext)
selected_partition = partition_selector(context, self)
if not selected_partition:
raise DagsterInvariantViolationError(
"The partition selection function `{selector}` did not return "
"a partition from PartitionSet {partition_set}".format(
selector=getattr(partition_selector, '__name__', repr(partition_selector)),
partition_set=self.name,
)
)
return self.tags_for_partition(selected_partition)
return PartitionScheduleDefinition(
name=schedule_name,
cron_schedule=cron_schedule,
pipeline_name=self.pipeline_name,
environment_dict_fn=_environment_dict_fn_wrapper,
tags_fn=_tags_fn_wrapper,
solid_subset=self.solid_subset,
mode=self.mode,
should_execute=_should_execute_wrapper,
environment_vars=environment_vars,
partition_set=self,
)
def __call__(self, fn: Callable[[], Any]) -> RepositoryDefinition:
from dagster.core.asset_defs import AssetGroup
check.callable_param(fn, "fn")
if not self.name:
self.name = fn.__name__
repository_definitions = fn()
if not (isinstance(repository_definitions, list)
or isinstance(repository_definitions, dict)
or isinstance(repository_definitions, RepositoryData)):
raise DagsterInvalidDefinitionError(
"Bad return value of type {type_} from repository construction function: must "
"return list, dict, or RepositoryData. See the @repository decorator docstring for "
"details and examples".format(
type_=type(repository_definitions)), )
if isinstance(repository_definitions, list):
bad_definitions = []
for i, definition in enumerate(repository_definitions):
if not (isinstance(definition, PipelineDefinition)
or isinstance(definition, PartitionSetDefinition)
or isinstance(definition, ScheduleDefinition)
or isinstance(definition, SensorDefinition)
or isinstance(definition, GraphDefinition)
or isinstance(definition, AssetGroup)):
bad_definitions.append((i, type(definition)))
if bad_definitions:
bad_definitions_str = ", ".join([
"value of type {type_} at index {i}".format(type_=type_,
i=i)
for i, type_ in bad_definitions
])
raise DagsterInvalidDefinitionError(
"Bad return value from repository construction function: all elements of list "
"must be of type JobDefinition, GraphDefinition, PipelineDefinition, "
"PartitionSetDefinition, ScheduleDefinition, or SensorDefinition. "
f"Got {bad_definitions_str}.")
repository_data = CachingRepositoryData.from_list(
repository_definitions)
elif isinstance(repository_definitions, dict):
if not set(repository_definitions.keys()).issubset(
VALID_REPOSITORY_DATA_DICT_KEYS):
raise DagsterInvalidDefinitionError(
"Bad return value from repository construction function: dict must not contain "
"keys other than {{'pipelines', 'partition_sets', 'schedules', 'jobs'}}: found "
"{bad_keys}".format(bad_keys=", ".join([
"'{key}'".format(key=key)
for key in repository_definitions.keys()
if key not in VALID_REPOSITORY_DATA_DICT_KEYS
])))
repository_data = CachingRepositoryData.from_dict(
repository_definitions)
elif isinstance(repository_definitions, RepositoryData):
repository_data = repository_definitions
repository_def = RepositoryDefinition(name=self.name,
description=self.description,
repository_data=repository_data)
update_wrapper(repository_def, fn)
return repository_def
def split_function_parameters(fn, expected_positionals):
check.callable_param(fn, 'fn')
check.list_param(expected_positionals, 'expected_positionals', str)
fn_params = list(funcsigs.signature(fn).parameters.values())
return fn_params[0:len(expected_positionals
)], fn_params[len(expected_positionals):]
def __init__(
self,
name: str,
pipeline_run_status: PipelineRunStatus,
run_status_sensor_fn: Callable[[RunStatusSensorContext],
Union[SkipReason, PipelineRunReaction]],
pipeline_selection: Optional[List[str]] = None,
minimum_interval_seconds: Optional[int] = None,
description: Optional[str] = None,
job_selection: Optional[List[Union[PipelineDefinition,
GraphDefinition]]] = None,
):
from dagster.core.storage.event_log.base import RunShardedEventsCursor, EventRecordsFilter
check.str_param(name, "name")
check.inst_param(pipeline_run_status, "pipeline_run_status",
PipelineRunStatus)
check.callable_param(run_status_sensor_fn, "run_status_sensor_fn")
check.opt_list_param(pipeline_selection, "pipeline_selection", str)
check.opt_int_param(minimum_interval_seconds,
"minimum_interval_seconds")
check.opt_str_param(description, "description")
check.opt_list_param(job_selection, "job_selection",
(PipelineDefinition, GraphDefinition))
def _wrapped_fn(context: SensorEvaluationContext):
# initiate the cursor to (most recent event id, current timestamp) when:
# * it's the first time starting the sensor
# * or, the cursor isn't in valid format (backcompt)
if context.cursor is None or not RunStatusSensorCursor.is_valid(
context.cursor):
most_recent_event_records = list(
context.instance.get_event_records(ascending=False,
limit=1))
most_recent_event_id = (most_recent_event_records[0].storage_id
if len(most_recent_event_records) == 1
else -1)
new_cursor = RunStatusSensorCursor(
update_timestamp=pendulum.now("UTC").isoformat(),
record_id=most_recent_event_id,
)
context.update_cursor(new_cursor.to_json())
yield SkipReason(
f"Initiating {name}. Set cursor to {new_cursor}")
return
record_id, update_timestamp = RunStatusSensorCursor.from_json(
context.cursor)
# Fetch events after the cursor id
# * we move the cursor forward to the latest visited event's id to avoid revisits
# * when the daemon is down, bc we persist the cursor info, we can go back to where we
# left and backfill alerts for the qualified events (up to 5 at a time) during the downtime
# Note: this is a cross-run query which requires extra handling in sqlite, see details in SqliteEventLogStorage.
event_records = context.instance.get_event_records(
EventRecordsFilter(
after_cursor=RunShardedEventsCursor(
id=record_id,
run_updated_after=cast(
datetime, pendulum.parse(update_timestamp)),
),
event_type=PIPELINE_RUN_STATUS_TO_EVENT_TYPE[
pipeline_run_status],
),
ascending=True,
limit=5,
)
for event_record in event_records:
event_log_entry = event_record.event_log_entry
storage_id = event_record.storage_id
# get run info
run_records = context.instance.get_run_records(
filters=PipelineRunsFilter(
run_ids=[event_log_entry.run_id]))
# skip if we couldn't find the right run
if len(run_records) != 1:
# bc we couldn't find the run, we use the event timestamp as the approximate
# run update timestamp
approximate_update_timestamp = utc_datetime_from_timestamp(
event_log_entry.timestamp)
context.update_cursor(
RunStatusSensorCursor(
record_id=storage_id,
update_timestamp=approximate_update_timestamp.
isoformat(),
).to_json())
continue
pipeline_run = run_records[0].pipeline_run
update_timestamp = run_records[0].update_timestamp
# skip if any of of the followings happens:
if (
# the pipeline does not have a repository (manually executed)
not pipeline_run.external_pipeline_origin or
# the pipeline does not belong to the current repository
pipeline_run.external_pipeline_origin.
external_repository_origin.repository_name !=
context.repository_name or
# if pipeline is not selected
(pipeline_selection
and pipeline_run.pipeline_name not in pipeline_selection)
or
# if job not selected
(job_selection and pipeline_run.pipeline_name not in map(
lambda x: x.name, job_selection))):
context.update_cursor(
RunStatusSensorCursor(
record_id=storage_id,
update_timestamp=update_timestamp.isoformat()).
to_json())
continue
serializable_error = None
try:
with user_code_error_boundary(
RunStatusSensorExecutionError,
lambda:
f'Error occurred during the execution sensor "{name}".',
):
# one user code invocation maps to one failure event
run_status_sensor_fn(
RunStatusSensorContext(
sensor_name=name,
dagster_run=pipeline_run,
dagster_event=event_log_entry.dagster_event,
instance=context.instance,
))
except RunStatusSensorExecutionError as run_status_sensor_execution_error:
# When the user code errors, we report error to the sensor tick not the original run.
serializable_error = serializable_error_info_from_exc_info(
run_status_sensor_execution_error.original_exc_info)
context.update_cursor(
RunStatusSensorCursor(record_id=storage_id,
update_timestamp=update_timestamp.
isoformat()).to_json())
# Yield PipelineRunReaction to indicate the execution success/failure.
# The sensor machinery would
# * report back to the original run if success
# * update cursor and job state
yield PipelineRunReaction(
pipeline_run=pipeline_run,
error=serializable_error,
)
super(RunStatusSensorDefinition, self).__init__(
name=name,
evaluation_fn=_wrapped_fn,
minimum_interval_seconds=minimum_interval_seconds,
description=description,
)
def create_schedule_definition(
self,
schedule_name,
cron_schedule,
partition_selector,
should_execute=None,
environment_vars=None,
execution_timezone=None,
description=None,
):
"""Create a ScheduleDefinition from a PartitionSetDefinition.
Arguments:
schedule_name (str): The name of the schedule.
cron_schedule (str): A valid cron string for the schedule
partition_selector (Callable[ScheduleExecutionContext, PartitionSetDefinition],
Partition): Function that determines the partition to use at a given execution time.
For time-based partition sets, will likely be either `identity_partition_selector` or a
selector returned by `create_offset_partition_selector`.
should_execute (Optional[function]): Function that runs at schedule execution time that
determines whether a schedule should execute. Defaults to a function that always returns
``True``.
environment_vars (Optional[dict]): The environment variables to set for the schedule.
execution_timezone (Optional[str]): Timezone in which the schedule should run. Only works
with DagsterDaemonScheduler, and must be set when using that scheduler.
description (Optional[str]): A human-readable description of the schedule.
Returns:
ScheduleDefinition: The generated ScheduleDefinition for the partition selector
"""
check.str_param(schedule_name, "schedule_name")
check.str_param(cron_schedule, "cron_schedule")
check.opt_callable_param(should_execute, "should_execute")
check.opt_dict_param(environment_vars,
"environment_vars",
key_type=str,
value_type=str)
check.callable_param(partition_selector, "partition_selector")
check.opt_str_param(execution_timezone, "execution_timezone")
check.opt_str_param(description, "description")
def _execution_fn(context):
check.inst_param(context, "context", ScheduleExecutionContext)
with user_code_error_boundary(
ScheduleExecutionError,
lambda:
f"Error occurred during the execution of partition_selector for schedule {schedule_name}",
):
selected_partition = partition_selector(context, self)
if not selected_partition:
yield SkipReason(
"Partition selector did not return a partition. Make sure that the timezone "
"on your partition set matches your execution timezone.")
return
if selected_partition.name not in self.get_partition_names(
context.scheduled_execution_time):
yield SkipReason(
f"Partition selector returned a partition {selected_partition.name} not in the partition set."
)
return
with user_code_error_boundary(
ScheduleExecutionError,
lambda:
f"Error occurred during the execution of should_execute for schedule {schedule_name}",
):
if should_execute and not should_execute(context):
yield SkipReason(
"should_execute function for {schedule_name} returned false."
.format(schedule_name=schedule_name))
return
with user_code_error_boundary(
ScheduleExecutionError,
lambda:
f"Error occurred during the execution of run_config_fn for schedule {schedule_name}",
):
run_config = self.run_config_for_partition(selected_partition)
with user_code_error_boundary(
ScheduleExecutionError,
lambda:
f"Error occurred during the execution of tags_fn for schedule {schedule_name}",
):
tags = self.tags_for_partition(selected_partition)
yield RunRequest(
run_key=None,
run_config=run_config,
tags=tags,
)
return PartitionScheduleDefinition(
name=schedule_name,
cron_schedule=cron_schedule,
pipeline_name=self.pipeline_name,
tags_fn=None,
solid_selection=self.solid_selection,
mode=self.mode,
should_execute=None,
environment_vars=environment_vars,
partition_set=self,
execution_timezone=execution_timezone,
execution_fn=_execution_fn,
description=description,
)
def validate_solid_fn(
decorator_name: str,
fn_name: str,
compute_fn: Callable[..., Any],
input_defs: List[InputDefinition],
expected_positionals: Optional[List[str]] = None,
exclude_nothing: Optional[bool] = True,
) -> List[str]:
check.str_param(decorator_name, "decorator_name")
check.str_param(fn_name, "fn_name")
check.callable_param(compute_fn, "compute_fn")
check.list_param(input_defs, "input_defs", of_type=InputDefinition)
expected_positionals = check.opt_list_param(expected_positionals,
"expected_positionals",
of_type=str)
if exclude_nothing:
names = set(inp.name for inp in input_defs
if not inp.dagster_type.kind == DagsterTypeKind.NOTHING)
nothing_names = set(
inp.name for inp in input_defs
if inp.dagster_type.kind == DagsterTypeKind.NOTHING)
else:
names = set(inp.name for inp in input_defs)
nothing_names = set()
# Currently being super strict about naming. Might be a good idea to relax. Starting strict.
fn_positionals, input_args = split_function_parameters(
compute_fn, expected_positionals)
# Validate Positional Parameters
missing_positional = validate_decorated_fn_positionals(
fn_positionals, expected_positionals)
if missing_positional:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function does not have required positional "
"parameter '{missing_param}'. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'."
.format(decorator_name=decorator_name,
solid_name=fn_name,
missing_param=missing_positional))
# Validate non positional parameters
invalid_function_info = validate_decorated_fn_input_args(names, input_args)
if invalid_function_info:
if invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES[
"vararg"]:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has positional vararg parameter "
"'{param}'. Solid functions should only have keyword arguments that match "
"input names and a first positional parameter named 'context'."
.format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
))
elif invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES[
"missing_name"]:
if invalid_function_info.param in nothing_names:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has parameter '{param}' that is "
"one of the solid input_defs of type 'Nothing' which should not be included since "
"no data will be passed for it. ".format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
))
else:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has parameter '{param}' that is not "
"one of the solid input_defs. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'."
.format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
))
elif invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES[
"extra"]:
undeclared_inputs_printed = ", '".join(
invalid_function_info.missing_names)
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function does not have parameter(s) "
"'{undeclared_inputs_printed}', which are in solid's input_defs. Solid functions "
"should only have keyword arguments that match input names and a first positional "
"parameter named 'context'.".format(
decorator_name=decorator_name,
solid_name=fn_name,
undeclared_inputs_printed=undeclared_inputs_printed,
))
return positional_arg_name_list(input_args)
def solid_execution_error_boundary(error_cls, msg_fn, step_context, **kwargs):
"""
A specialization of user_code_error_boundary for the steps involved in executing a solid.
This variant supports the control flow exceptions RetryRequested and Failure as well
as respecting the RetryPolicy if present.
"""
from dagster.core.execution.context.system import StepExecutionContext
check.callable_param(msg_fn, "msg_fn")
check.class_param(error_cls,
"error_cls",
superclass=DagsterUserCodeExecutionError)
check.inst_param(step_context, "step_context", StepExecutionContext)
with raise_execution_interrupts():
step_context.log.begin_python_log_capture()
retry_policy = step_context.solid_retry_policy
try:
yield
except DagsterError as de:
# The system has thrown an error that is part of the user-framework contract
raise de
except Exception as e: # pylint: disable=W0703
# An exception has been thrown by user code and computation should cease
# with the error reported further up the stack
# Directly thrown RetryRequested escalate before evaluating the retry policy.
if isinstance(e, RetryRequested):
raise e
if retry_policy:
raise RetryRequested(
max_retries=retry_policy.max_retries,
seconds_to_wait=retry_policy.calculate_delay(
step_context.previous_attempt_count + 1),
) from e
# Failure exceptions get re-throw without wrapping
if isinstance(e, Failure):
raise e
# Otherwise wrap the user exception with context
raise error_cls(
msg_fn(),
user_exception=e,
original_exc_info=sys.exc_info(),
**kwargs,
) from e
except (DagsterExecutionInterruptedError, KeyboardInterrupt) as ie:
# respect retry policy when interrupts occur
if retry_policy:
raise RetryRequested(
max_retries=retry_policy.max_retries,
seconds_to_wait=retry_policy.calculate_delay(
step_context.previous_attempt_count + 1),
) from ie
else:
raise ie
finally:
step_context.log.end_python_log_capture()
def __init__(self, config_type, func, required_resource_keys):
self._config_type = check.inst_param(config_type, 'config_type',
ConfigType)
self._func = check.callable_param(func, 'func')
self._required_resource_keys = check.opt_set_param(
required_resource_keys, 'required_resource_keys', of_type=str)
def __init__(
self,
type_check_fn,
key=None,
name=None,
is_builtin=False,
description=None,
loader=None,
materializer=None,
serialization_strategy=None,
auto_plugins=None,
required_resource_keys=None,
kind=DagsterTypeKind.REGULAR,
):
check.opt_str_param(key, "key")
check.opt_str_param(name, "name")
check.invariant(not (name is None and key is None), "Must set key or name")
if name is None:
check.param_invariant(
bool(key), "key", "If name is not provided, must provide key.",
)
self.key, self._name = key, None
elif key is None:
check.param_invariant(
bool(name), "name", "If key is not provided, must provide name.",
)
self.key, self._name = name, name
else:
check.invariant(key and name)
self.key, self._name = key, name
self.description = check.opt_str_param(description, "description")
self.loader = check.opt_inst_param(loader, "loader", DagsterTypeLoader)
self.materializer = check.opt_inst_param(
materializer, "materializer", DagsterTypeMaterializer
)
self.serialization_strategy = check.opt_inst_param(
serialization_strategy,
"serialization_strategy",
SerializationStrategy,
PickleSerializationStrategy(),
)
self.required_resource_keys = check.opt_set_param(
required_resource_keys, "required_resource_keys",
)
self._type_check_fn = check.callable_param(type_check_fn, "type_check_fn")
_validate_type_check_fn(self._type_check_fn, self._name)
auto_plugins = check.opt_list_param(auto_plugins, "auto_plugins", of_type=type)
check.param_invariant(
all(
issubclass(auto_plugin_type, TypeStoragePlugin) for auto_plugin_type in auto_plugins
),
"auto_plugins",
)
self.auto_plugins = auto_plugins
self.is_builtin = check.bool_param(is_builtin, "is_builtin")
check.invariant(
self.display_name is not None,
"All types must have a valid display name, got None for key {}".format(key),
)
self.kind = check.inst_param(kind, "kind", DagsterTypeKind)
def inner(
fn: Callable[
...,
Union[RunRequest, SkipReason, RunConfig, RunRequestGenerator],
]
) -> ScheduleDefinition:
check.callable_param(fn, "fn")
schedule_name = name or fn.__name__
# perform upfront validation of schedule tags
_tags_fn: Optional[Callable[["ScheduleEvaluationContext"], Dict[str, str]]] = None
if tags_fn and tags:
raise DagsterInvalidDefinitionError(
"Attempted to provide both tags_fn and tags as arguments"
" to ScheduleDefinition. Must provide only one of the two."
)
elif tags:
check_tags(tags, "tags")
_tags_fn = cast(Callable[["ScheduleEvaluationContext"], Dict[str, str]], lambda _: tags)
elif tags_fn:
_tags_fn = cast(
Callable[["ScheduleEvaluationContext"], Dict[str, str]],
lambda context: tags_fn(context) or {},
)
def _wrapped_fn(context: "ScheduleEvaluationContext"):
if should_execute:
with user_code_error_boundary(
ScheduleExecutionError,
lambda: f"Error occurred during the execution of should_execute for schedule {schedule_name}",
):
if not should_execute(context):
yield SkipReason(
f"should_execute function for {schedule_name} returned false."
)
return
with user_code_error_boundary(
ScheduleExecutionError,
lambda: f"Error occurred during the evaluation of schedule {schedule_name}",
):
result = fn(context) if has_context_arg else fn()
if isinstance(result, dict):
# this is the run-config based decorated function, wrap the evaluated run config
# and tags in a RunRequest
evaluated_run_config = copy.deepcopy(result)
evaluated_tags = _tags_fn(context) if _tags_fn else None
yield RunRequest(
run_key=None,
run_config=evaluated_run_config,
tags=evaluated_tags,
)
else:
# this is a run-request based decorated function
yield from ensure_gen(result)
has_context_arg = is_context_provided(get_function_params(fn))
evaluation_fn = DecoratedScheduleFunction(
decorated_fn=fn,
wrapped_fn=_wrapped_fn,
has_context_arg=has_context_arg,
)
schedule_def = ScheduleDefinition(
name=schedule_name,
cron_schedule=cron_schedule,
pipeline_name=pipeline_name,
solid_selection=solid_selection,
mode=mode,
environment_vars=environment_vars,
execution_timezone=execution_timezone,
description=description,
execution_fn=evaluation_fn,
job=job,
default_status=default_status,
)
update_wrapper(schedule_def, wrapped=fn)
return schedule_def
def schedule_partition_range(
start,
end,
cron_schedule,
fmt,
timezone,
execution_time_to_partition_fn,
):
check.inst_param(start, "start", datetime.datetime)
check.opt_inst_param(end, "end", datetime.datetime)
check.str_param(cron_schedule, "cron_schedule")
check.str_param(fmt, "fmt")
check.opt_str_param(timezone, "timezone")
check.callable_param(execution_time_to_partition_fn,
"execution_time_to_partition_fn")
if end and start > end:
raise DagsterInvariantViolationError(
'Selected date range start "{start}" is after date range end "{end}'
.format(
start=start.strftime(fmt),
end=end.strftime(fmt),
))
def get_schedule_range_partitions(current_time=None):
check.opt_inst_param(current_time, "current_time", datetime.datetime)
tz = timezone if timezone else pendulum.now().timezone.name
_start = (start.in_tz(tz) if isinstance(start, pendulum.Pendulum) else
pendulum.instance(start, tz=tz))
if end:
_end = end
elif current_time:
_end = current_time
else:
_end = pendulum.now(tz)
# coerce to the definition timezone
if isinstance(_end, pendulum.Pendulum):
_end = _end.in_tz(tz)
else:
_end = pendulum.instance(_end, tz=tz)
end_timestamp = _end.timestamp()
partitions = []
for next_time in schedule_execution_time_iterator(
_start.timestamp(), cron_schedule, tz):
partition_time = execution_time_to_partition_fn(next_time)
if partition_time.timestamp() > end_timestamp:
break
if partition_time.timestamp() < _start.timestamp():
continue
partitions.append(
Partition(value=partition_time,
name=partition_time.strftime(fmt)))
return partitions[:-1]
return get_schedule_range_partitions
def event_generator(
self,
execution_plan,
run_config,
pipeline_run,
instance,
scoped_resources_builder_cm,
intermediate_storage=None,
raise_on_error=False,
resource_instances_to_override=None,
):
execution_plan = check.inst_param(execution_plan, "execution_plan",
ExecutionPlan)
pipeline_def = execution_plan.pipeline.get_definition()
run_config = check.dict_param(run_config, "run_config", key_type=str)
pipeline_run = check.inst_param(pipeline_run, "pipeline_run",
PipelineRun)
instance = check.inst_param(instance, "instance", DagsterInstance)
scoped_resources_builder_cm = check.callable_param(
scoped_resources_builder_cm, "scoped_resources_builder_cm")
intermediate_storage = check.opt_inst_param(
intermediate_storage, "intermediate_storage_data",
IntermediateStorage)
raise_on_error = check.bool_param(raise_on_error, "raise_on_error")
resource_instances_to_override = check.opt_dict_param(
resource_instances_to_override, "resource_instances_to_override")
execution_context = None
resources_manager = None
try:
context_creation_data = create_context_creation_data(
execution_plan,
run_config,
pipeline_run,
instance,
)
log_manager = create_log_manager(context_creation_data)
resources_manager = scoped_resources_builder_cm(
execution_plan,
context_creation_data.environment_config,
context_creation_data.pipeline_run,
log_manager,
context_creation_data.resource_keys_to_init,
instance,
resource_instances_to_override,
)
yield from resources_manager.generate_setup_events()
scoped_resources_builder = check.inst(
resources_manager.get_object(), ScopedResourcesBuilder)
intermediate_storage = create_intermediate_storage(
context_creation_data,
intermediate_storage,
scoped_resources_builder,
)
execution_context = self.construct_context(
context_creation_data=context_creation_data,
scoped_resources_builder=scoped_resources_builder,
log_manager=log_manager,
intermediate_storage=intermediate_storage,
raise_on_error=raise_on_error,
)
_validate_plan_with_context(execution_context, execution_plan)
yield execution_context
yield from resources_manager.generate_teardown_events()
except DagsterError as dagster_error:
if execution_context is None:
user_facing_exc_info = (
# pylint does not know original_exc_info exists is is_user_code_error is true
# pylint: disable=no-member
dagster_error.original_exc_info
if dagster_error.is_user_code_error else sys.exc_info())
error_info = serializable_error_info_from_exc_info(
user_facing_exc_info)
yield DagsterEvent.pipeline_init_failure(
pipeline_name=pipeline_def.name,
failure_data=PipelineInitFailureData(error=error_info),
log_manager=_create_context_free_log_manager(
instance, pipeline_run, pipeline_def),
)
if resources_manager:
yield from resources_manager.generate_teardown_events()
else:
# pipeline teardown failure
raise dagster_error
if raise_on_error:
raise dagster_error
def _core_celery_execution_loop(pipeline_context, execution_plan,
step_execution_fn):
from .tasks import make_app
check.inst_param(pipeline_context, 'pipeline_context',
SystemPipelineExecutionContext)
check.inst_param(execution_plan, 'execution_plan', ExecutionPlan)
check.callable_param(step_execution_fn, 'step_execution_fn')
check.param_invariant(
isinstance(pipeline_context.executor_config,
(CeleryConfig, CeleryK8sJobConfig)),
'pipeline_context',
'Expected executor_config to be Celery config got {}'.format(
pipeline_context.executor_config),
)
celery_config = pipeline_context.executor_config
storage = pipeline_context.environment_dict.get('storage')
# https://github.com/dagster-io/dagster/issues/2440
check.invariant(
pipeline_context.system_storage_def.is_persistent,
'Cannot use in-memory storage with Celery, use filesystem (on top of NFS or '
'similar system that allows files to be available to all nodes), S3, or GCS',
)
app = make_app(celery_config)
priority_for_step = lambda step: (-1 * int(
step.tags.get(DAGSTER_CELERY_STEP_PRIORITY_TAG, task_default_priority)
) + -1 * _get_run_priority(pipeline_context))
priority_for_key = lambda step_key: (priority_for_step(
execution_plan.get_step_by_key(step_key)))
_warn_on_priority_misuse(pipeline_context, execution_plan)
step_results = {} # Dict[ExecutionStep, celery.AsyncResult]
step_errors = {}
completed_steps = set({}) # Set[step_key]
active_execution = execution_plan.start(
retries=pipeline_context.executor_config.retries,
sort_key_fn=priority_for_step)
stopping = False
while (not active_execution.is_complete and not stopping) or step_results:
results_to_pop = []
for step_key, result in sorted(step_results.items(),
key=lambda x: priority_for_key(x[0])):
if result.ready():
try:
step_events = result.get()
except Exception as e: # pylint: disable=broad-except
# We will want to do more to handle the exception here.. maybe subclass Task
# Certainly yield an engine or pipeline event
step_events = []
step_errors[
step_key] = serializable_error_info_from_exc_info(
sys.exc_info())
stopping = True
for step_event in step_events:
event = deserialize_json_to_dagster_namedtuple(step_event)
yield event
active_execution.handle_event(event)
results_to_pop.append(step_key)
completed_steps.add(step_key)
for step_key in results_to_pop:
if step_key in step_results:
del step_results[step_key]
active_execution.verify_complete(pipeline_context, step_key)
# process skips from failures or uncovered inputs
for event in active_execution.skipped_step_events_iterator(
pipeline_context):
yield event
# don't add any new steps if we are stopping
if stopping:
continue
# This is a slight refinement. If we have n workers idle and schedule m > n steps for
# execution, the first n steps will be picked up by the idle workers in the order in
# which they are scheduled (and the following m-n steps will be executed in priority
# order, provided that it takes longer to execute a step than to schedule it). The test
# case has m >> n to exhibit this behavior in the absence of this sort step.
for step in active_execution.get_steps_to_execute():
try:
queue = step.tags.get(DAGSTER_CELERY_QUEUE_TAG,
task_default_queue)
yield DagsterEvent.engine_event(
pipeline_context,
'Submitting celery task for step "{step_key}" to queue "{queue}".'
.format(step_key=step.key, queue=queue),
EngineEventData(marker_start=DELEGATE_MARKER),
step_key=step.key,
)
# Get the Celery priority for this step
priority = _get_step_priority(pipeline_context, step)
# Submit the Celery tasks
step_results[step.key] = step_execution_fn(
app, pipeline_context, step, queue, priority)
except Exception:
yield DagsterEvent.engine_event(
pipeline_context,
'Encountered error during celery task submission.'.format(
),
event_specific_data=EngineEventData.engine_error(
serializable_error_info_from_exc_info(
sys.exc_info()), ),
)
raise
time.sleep(TICK_SECONDS)
if step_errors:
raise DagsterSubprocessError(
'During celery execution errors occurred in workers:\n{error_list}'
.format(error_list='\n'.join([
'[{step}]: {err}'.format(step=key, err=err.to_string())
for key, err in step_errors.items()
])),
subprocess_error_infos=list(step_errors.values()),
)
def _create_solid_compute_wrapper(fn, input_defs, output_defs):
check.callable_param(fn, "fn")
check.list_param(input_defs, "input_defs", of_type=InputDefinition)
check.list_param(output_defs, "output_defs", of_type=OutputDefinition)
input_names = [
input_def.name for input_def in input_defs
if not input_def.dagster_type.kind == DagsterTypeKind.NOTHING
]
@wraps(fn)
def compute(context, input_defs):
kwargs = {}
for input_name in input_names:
kwargs[input_name] = input_defs[input_name]
result = fn(context, **kwargs)
if inspect.isgenerator(result):
for item in result:
yield item
else:
if isinstance(
result,
(AssetMaterialization, Materialization, ExpectationResult)):
raise DagsterInvariantViolationError((
"Error in solid {solid_name}: If you are returning an AssetMaterialization "
"or an ExpectationResult from solid you must yield them to avoid "
"ambiguity with an implied result from returning a value.".
format(solid_name=context.solid.name)))
if isinstance(result, Output):
yield result
elif len(output_defs) == 1:
yield Output(value=result, output_name=output_defs[0].name)
elif result is not None:
if not output_defs:
raise DagsterInvariantViolationError((
"Error in solid {solid_name}: Unexpectedly returned output {result} "
"of type {type_}. Solid is explicitly defined to return no "
"results.").format(solid_name=context.solid.name,
result=result,
type_=type(result)))
raise DagsterInvariantViolationError((
"Error in solid {solid_name}: Solid unexpectedly returned "
"output {result} of type {type_}. Should "
"be a generator, containing or yielding "
"{n_results} results: {{{expected_results}}}.").format(
solid_name=context.solid.name,
result=result,
type_=type(result),
n_results=len(output_defs),
expected_results=", ".join([
"'{result_name}': {dagster_type}".format(
result_name=output_def.name,
dagster_type=output_def.dagster_type,
) for output_def in output_defs
]),
))
return compute
def pipeline_initialization_event_generator(
execution_plan,
run_config,
pipeline_run,
instance,
scoped_resources_builder_cm,
system_storage_data=None,
intermediate_storage=None,
raise_on_error=False,
):
execution_plan = check.inst_param(execution_plan, 'execution_plan',
ExecutionPlan)
pipeline_def = execution_plan.pipeline.get_definition()
run_config = check.dict_param(run_config, 'run_config', key_type=str)
pipeline_run = check.inst_param(pipeline_run, 'pipeline_run', PipelineRun)
instance = check.inst_param(instance, 'instance', DagsterInstance)
scoped_resources_builder_cm = check.callable_param(
scoped_resources_builder_cm, 'scoped_resources_builder_cm')
system_storage_data = check.opt_inst_param(system_storage_data,
'system_storage_data',
SystemStorageData)
intermediate_storage = check.opt_inst_param(intermediate_storage,
'intermediate_storage_data',
IntermediateStorage)
raise_on_error = check.bool_param(raise_on_error, 'raise_on_error')
pipeline_context = None
resources_manager = None
try:
context_creation_data = create_context_creation_data(
execution_plan,
run_config,
pipeline_run,
instance,
)
executor = check.inst(create_executor(context_creation_data), Executor,
'Must return an Executor')
log_manager = create_log_manager(context_creation_data)
resources_manager = scoped_resources_builder_cm(
execution_plan,
context_creation_data.environment_config,
context_creation_data.pipeline_run,
log_manager,
context_creation_data.resource_keys_to_init,
)
for event in resources_manager.generate_setup_events():
yield event
scoped_resources_builder = check.inst(resources_manager.get_object(),
ScopedResourcesBuilder)
system_storage_data = create_system_storage_data(
context_creation_data, system_storage_data,
scoped_resources_builder)
if intermediate_storage or context_creation_data.intermediate_storage_def:
intermediate_storage = create_intermediate_storage(
context_creation_data,
intermediate_storage,
scoped_resources_builder,
)
else:
# remove this as part of https://github.com/dagster-io/dagster/issues/2705
intermediate_storage = system_storage_data.intermediates_manager
pipeline_context = construct_pipeline_execution_context(
context_creation_data=context_creation_data,
scoped_resources_builder=scoped_resources_builder,
system_storage_data=system_storage_data,
intermediate_storage=intermediate_storage,
log_manager=log_manager,
executor=executor,
raise_on_error=raise_on_error,
)
_validate_plan_with_context(pipeline_context, execution_plan)
yield pipeline_context
for event in resources_manager.generate_teardown_events():
yield event
except DagsterError as dagster_error:
if pipeline_context is None:
user_facing_exc_info = (
# pylint does not know original_exc_info exists is is_user_code_error is true
# pylint: disable=no-member
dagster_error.original_exc_info
if dagster_error.is_user_code_error else sys.exc_info())
error_info = serializable_error_info_from_exc_info(
user_facing_exc_info)
yield DagsterEvent.pipeline_init_failure(
pipeline_name=pipeline_def.name,
failure_data=PipelineInitFailureData(error=error_info),
log_manager=_create_context_free_log_manager(
instance, pipeline_run, pipeline_def),
)
if resources_manager:
for event in resources_manager.generate_teardown_events():
yield event
else:
# pipeline teardown failure
raise dagster_error
if raise_on_error:
raise dagster_error
def __new__(cls, config_fn, config_schema=None):
return super(ConfigMapping, cls).__new__(
cls,
config_fn=check.callable_param(config_fn, 'config_fn'),
config_schema=check_user_facing_opt_config_param(config_schema, 'config_schema'),
)
def __init__(
self,
type_check_fn: TypeCheckFn,
key: t.Optional[str] = None,
name: t.Optional[str] = None,
is_builtin: bool = False,
description: t.Optional[str] = None,
loader: t.Optional[DagsterTypeLoader] = None,
materializer: t.Optional[DagsterTypeMaterializer] = None,
required_resource_keys: t.Optional[t.Set[str]] = None,
kind: DagsterTypeKind = DagsterTypeKind.REGULAR,
typing_type: t.Any = None,
metadata_entries: t.Optional[t.List[MetadataEntry]] = None,
metadata: t.Optional[t.Dict[str, RawMetadataValue]] = None,
):
check.opt_str_param(key, "key")
check.opt_str_param(name, "name")
check.invariant(not (name is None and key is None),
"Must set key or name")
if name is None:
key = check.not_none(
key,
"If name is not provided, must provide key.",
)
self.key, self._name = key, None
elif key is None:
name = check.not_none(
name,
"If key is not provided, must provide name.",
)
self.key, self._name = name, name
else:
check.invariant(key and name)
self.key, self._name = key, name
self.description = check.opt_str_param(description, "description")
self.loader = check.opt_inst_param(loader, "loader", DagsterTypeLoader)
self.materializer = check.opt_inst_param(materializer, "materializer",
DagsterTypeMaterializer)
self.required_resource_keys = check.opt_set_param(
required_resource_keys,
"required_resource_keys",
)
self._type_check_fn = check.callable_param(type_check_fn,
"type_check_fn")
_validate_type_check_fn(self._type_check_fn, self._name)
self.is_builtin = check.bool_param(is_builtin, "is_builtin")
check.invariant(
self.display_name is not None,
"All types must have a valid display name, got None for key {}".
format(key),
)
self.kind = check.inst_param(kind, "kind", DagsterTypeKind)
self.typing_type = typing_type
metadata_entries = check.opt_list_param(metadata_entries,
"metadata_entries",
of_type=MetadataEntry)
metadata = check.opt_dict_param(metadata, "metadata", key_type=str)
self._metadata_entries = normalize_metadata(metadata, metadata_entries)
def __init__(self, indent_level=2, printer=print):
self.current_indent = 0
self.indent_level = check.int_param(indent_level, 'indent_level')
self.printer = check.callable_param(printer, 'printer')
self._line_so_far = ''
def core_celery_execution_loop(pipeline_context, execution_plan,
step_execution_fn):
check.inst_param(pipeline_context, "pipeline_context",
SystemPipelineExecutionContext)
check.inst_param(execution_plan, "execution_plan", ExecutionPlan)
check.callable_param(step_execution_fn, "step_execution_fn")
executor = pipeline_context.executor
# https://github.com/dagster-io/dagster/issues/2440
check.invariant(
execution_plan.artifacts_persisted,
"Cannot use in-memory storage with Celery, use filesystem (on top of NFS or "
"similar system that allows files to be available to all nodes), S3, or GCS",
)
app = make_app(executor.app_args())
priority_for_step = lambda step: (-1 * int(
step.tags.get(DAGSTER_CELERY_STEP_PRIORITY_TAG, task_default_priority)
) + -1 * _get_run_priority(pipeline_context))
priority_for_key = lambda step_key: (priority_for_step(
execution_plan.get_step_by_key(step_key)))
_warn_on_priority_misuse(pipeline_context, execution_plan)
step_results = {} # Dict[ExecutionStep, celery.AsyncResult]
step_errors = {}
with execution_plan.start(
retries=pipeline_context.executor.retries,
sort_key_fn=priority_for_step,
) as active_execution:
stopping = False
while (not active_execution.is_complete
and not stopping) or step_results:
if active_execution.check_for_interrupts():
yield DagsterEvent.engine_event(
pipeline_context,
"Celery executor: received termination signal - revoking active tasks from workers",
EngineEventData.interrupted(list(step_results.keys())),
)
stopping = True
for key, result in step_results.items():
result.revoke()
active_execution.mark_interrupted(key)
results_to_pop = []
for step_key, result in sorted(
step_results.items(),
key=lambda x: priority_for_key(x[0])):
if result.ready():
try:
step_events = result.get()
except TaskRevokedError:
step_events = []
yield DagsterEvent.engine_event(
pipeline_context,
'celery task for running step "{step_key}" was revoked.'
.format(step_key=step_key, ),
EngineEventData(marker_end=DELEGATE_MARKER),
step_key=step_key,
)
except Exception: # pylint: disable=broad-except
# We will want to do more to handle the exception here.. maybe subclass Task
# Certainly yield an engine or pipeline event
step_events = []
step_errors[
step_key] = serializable_error_info_from_exc_info(
sys.exc_info())
for step_event in step_events:
event = deserialize_json_to_dagster_namedtuple(
step_event)
yield event
active_execution.handle_event(event)
results_to_pop.append(step_key)
for step_key in results_to_pop:
if step_key in step_results:
del step_results[step_key]
active_execution.verify_complete(pipeline_context,
step_key)
# process skips from failures or uncovered inputs
for event in active_execution.plan_events_iterator(
pipeline_context):
yield event
# don't add any new steps if we are stopping
if stopping or step_errors:
continue
# This is a slight refinement. If we have n workers idle and schedule m > n steps for
# execution, the first n steps will be picked up by the idle workers in the order in
# which they are scheduled (and the following m-n steps will be executed in priority
# order, provided that it takes longer to execute a step than to schedule it). The test
# case has m >> n to exhibit this behavior in the absence of this sort step.
for step in active_execution.get_steps_to_execute():
try:
queue = step.tags.get(DAGSTER_CELERY_QUEUE_TAG,
task_default_queue)
yield DagsterEvent.engine_event(
pipeline_context,
'Submitting celery task for step "{step_key}" to queue "{queue}".'
.format(step_key=step.key, queue=queue),
EngineEventData(marker_start=DELEGATE_MARKER),
step_key=step.key,
)
# Get the Celery priority for this step
priority = _get_step_priority(pipeline_context, step)
# Submit the Celery tasks
step_results[step.key] = step_execution_fn(
app, pipeline_context, step, queue, priority)
except Exception:
yield DagsterEvent.engine_event(
pipeline_context,
"Encountered error during celery task submission.".
format(),
event_specific_data=EngineEventData.engine_error(
serializable_error_info_from_exc_info(
sys.exc_info()), ),
)
raise
time.sleep(TICK_SECONDS)
if step_errors:
raise DagsterSubprocessError(
"During celery execution errors occurred in workers:\n{error_list}"
.format(error_list="\n".join([
"[{step}]: {err}".format(step=key, err=err.to_string())
for key, err in step_errors.items()
])),
subprocess_error_infos=list(step_errors.values()),
)
def __init__(
self,
type_check_fn,
key=None,
name=None,
is_builtin=False,
description=None,
input_hydration_config=None,
output_materialization_config=None,
serialization_strategy=None,
auto_plugins=None,
required_resource_keys=None,
kind=DagsterTypeKind.REGULAR,
):
check.opt_str_param(key, 'key')
check.opt_str_param(name, 'name')
check.invariant(not (name is None and key is None),
'Must set key or name')
if name is None:
check.param_invariant(
bool(key),
'key',
'If name is not provided, must provide key.',
)
self.key, self.name = key, None
elif key is None:
check.param_invariant(
bool(name),
'name',
'If key is not provided, must provide name.',
)
self.key, self.name = name, name
else:
check.invariant(key and name)
self.key, self.name = key, name
self.description = check.opt_str_param(description, 'description')
self.input_hydration_config = check.opt_inst_param(
input_hydration_config, 'input_hydration_config',
InputHydrationConfig)
self.output_materialization_config = check.opt_inst_param(
output_materialization_config,
'output_materialization_config',
OutputMaterializationConfig,
)
self.serialization_strategy = check.opt_inst_param(
serialization_strategy,
'serialization_strategy',
SerializationStrategy,
PickleSerializationStrategy(),
)
self.required_resource_keys = check.opt_set_param(
required_resource_keys,
'required_resource_keys',
)
self._type_check_fn = check.callable_param(type_check_fn,
'type_check_fn')
_validate_type_check_fn(self._type_check_fn, self.name)
auto_plugins = check.opt_list_param(auto_plugins,
'auto_plugins',
of_type=type)
check.param_invariant(
all(
issubclass(auto_plugin_type, TypeStoragePlugin)
for auto_plugin_type in auto_plugins),
'auto_plugins',
)
self.auto_plugins = auto_plugins
self.is_builtin = check.bool_param(is_builtin, 'is_builtin')
check.invariant(
self.display_name is not None,
'All types must have a valid display name, got None for key {}'.
format(key),
)
self.kind = check.inst_param(kind, 'kind', DagsterTypeKind)
def validate_solid_fn(
decorator_name, fn_name, compute_fn, input_defs, expected_positionals=None, exclude_nothing=True
):
check.str_param(decorator_name, 'decorator_name')
check.str_param(fn_name, 'fn_name')
check.callable_param(compute_fn, 'compute_fn')
check.list_param(input_defs, 'input_defs', of_type=InputDefinition)
expected_positionals = check.opt_list_param(
expected_positionals, 'expected_positionals', of_type=str
)
if exclude_nothing:
names = set(inp.name for inp in input_defs if not inp.runtime_type.is_nothing)
nothing_names = set(inp.name for inp in input_defs if inp.runtime_type.is_nothing)
else:
names = set(inp.name for inp in input_defs)
nothing_names = set()
# Currently being super strict about naming. Might be a good idea to relax. Starting strict.
fn_positionals, input_args = split_function_parameters(compute_fn, expected_positionals)
# Validate Positional Parameters
missing_positional = validate_decorated_fn_positionals(fn_positionals, expected_positionals)
if missing_positional:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function does not have required positional "
"parameter '{missing_param}'. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'.".format(
decorator_name=decorator_name, solid_name=fn_name, missing_param=missing_positional
)
)
# Validate non positional parameters
invalid_function_info = validate_decorated_fn_input_args(names, input_args)
if invalid_function_info:
if invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES['vararg']:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has positional vararg parameter "
"'{param}'. Solid functions should only have keyword arguments that match "
"input names and a first positional parameter named 'context'.".format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
)
)
elif invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES['missing_name']:
if invalid_function_info.param in nothing_names:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has parameter '{param}' that is "
"one of the solid input_defs of type 'Nothing' which should not be included since "
"no data will be passed for it. ".format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
)
)
else:
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function has parameter '{param}' that is not "
"one of the solid input_defs. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'.".format(
decorator_name=decorator_name,
solid_name=fn_name,
param=invalid_function_info.param,
)
)
elif invalid_function_info.error_type == InvalidDecoratedFunctionInfo.TYPES['extra']:
undeclared_inputs_printed = ", '".join(invalid_function_info.missing_names)
raise DagsterInvalidDefinitionError(
"{decorator_name} '{solid_name}' decorated function does not have parameter(s) "
"'{undeclared_inputs_printed}', which are in solid's input_defs. Solid functions "
"should only have keyword arguments that match input names and a first positional "
"parameter named 'context'.".format(
decorator_name=decorator_name,
solid_name=fn_name,
undeclared_inputs_printed=undeclared_inputs_printed,
)
)
return positional_arg_name_list(input_args)
def __init__(self, config_type, func, required_resource_keys):
self._config_type = check.inst_param(config_type, "config_type",
ConfigType)
self._func = check.callable_param(func, "func")
self._required_resource_keys = check.opt_set_param(
required_resource_keys, "required_resource_keys", of_type=str)
def _validate_solid_fn(solid_name,
compute_fn,
input_defs,
expected_positionals=None,
exclude_nothing=True):
check.str_param(solid_name, 'solid_name')
check.callable_param(compute_fn, 'compute_fn')
check.list_param(input_defs, 'input_defs', of_type=InputDefinition)
expected_positionals = check.opt_list_param(expected_positionals,
'expected_positionals',
of_type=(str, tuple))
if exclude_nothing:
names = set(inp.name for inp in input_defs
if not inp.runtime_type.is_nothing)
nothing_names = set(inp.name for inp in input_defs
if inp.runtime_type.is_nothing)
else:
names = set(inp.name for inp in input_defs)
nothing_names = set()
# Currently being super strict about naming. Might be a good idea to relax. Starting strict.
try:
_validate_decorated_fn(compute_fn, names, expected_positionals)
except FunctionValidationError as e:
if e.error_type == FunctionValidationError.TYPES['vararg']:
raise DagsterInvalidDefinitionError(
"solid '{solid_name}' decorated function has positional vararg parameter "
"'{e.param}'. Solid functions should only have keyword arguments that match "
"input names and a first positional parameter named 'context'."
.format(solid_name=solid_name, e=e))
elif e.error_type == FunctionValidationError.TYPES['missing_name']:
if e.param in nothing_names:
raise DagsterInvalidDefinitionError(
"solid '{solid_name}' decorated function has parameter '{e.param}' that is "
"one of the solid input_defs of type 'Nothing' which should not be included since "
"no data will be passed for it. ".format(
solid_name=solid_name, e=e))
else:
raise DagsterInvalidDefinitionError(
"solid '{solid_name}' decorated function has parameter '{e.param}' that is not "
"one of the solid input_defs. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'."
.format(solid_name=solid_name, e=e))
elif e.error_type == FunctionValidationError.TYPES[
'missing_positional']:
raise DagsterInvalidDefinitionError(
"solid '{solid_name}' decorated function does not have required positional "
"parameter '{e.param}'. Solid functions should only have keyword arguments "
"that match input names and a first positional parameter named 'context'."
.format(solid_name=solid_name, e=e))
elif e.error_type == FunctionValidationError.TYPES['extra']:
undeclared_inputs_printed = ", '".join(e.missing_names)
raise DagsterInvalidDefinitionError(
"solid '{solid_name}' decorated function does not have parameter(s) "
"'{undeclared_inputs_printed}', which are in solid's input_defs. Solid functions "
"should only have keyword arguments that match input names and a first positional "
"parameter named 'context'.".format(
solid_name=solid_name,
undeclared_inputs_printed=undeclared_inputs_printed))
else:
raise e
def __init__(self, resource_fn, config_field=None, description=None):
self.resource_fn = check.callable_param(resource_fn, 'resource_fn')
self.config_field = check_opt_field_param(config_field, 'config_field')
self.description = check.opt_str_param(description, 'description')
