def _event_record(run_id,
solid_name,
timestamp,
event_type,
event_specific_data=None):
pipeline_name = "pipeline_name"
solid_handle = SolidHandle(solid_name, None)
step_handle = StepHandle(solid_handle)
return EventRecord(
None,
"",
"debug",
"",
run_id,
timestamp,
step_key=step_handle.to_key(),
pipeline_name=pipeline_name,
dagster_event=DagsterEvent(
event_type.value,
pipeline_name,
solid_handle=solid_handle,
step_handle=step_handle,
event_specific_data=event_specific_data,
),
)
def create_compute_step(pipeline_name, environment_config, solid, step_inputs,
handle):
check.str_param(pipeline_name, "pipeline_name")
check.inst_param(solid, "solid", Solid)
check.list_param(step_inputs, "step_inputs", of_type=StepInput)
check.opt_inst_param(handle, "handle", SolidHandle)
# the environment config has the solid output name configured
config_output_names = set()
current_handle = handle
while current_handle:
solid_config = environment_config.solids.get(
current_handle.to_string())
current_handle = current_handle.parent
config_output_names = config_output_names.union(
solid_config.outputs.output_names)
return ExecutionStep(
handle=StepHandle(solid_handle=handle),
pipeline_name=pipeline_name,
step_inputs=step_inputs,
step_outputs=[
StepOutput(output_def=output_def,
should_materialize=name in config_output_names)
for name, output_def in solid.definition.output_dict.items()
],
compute_fn=lambda step_context, inputs: _execute_core_compute(
step_context.for_compute(), inputs, solid.definition.compute_fn),
solid=solid,
)
def create_compute_step(solid, solid_handle, step_inputs, pipeline_name,
environment_config):
check.inst_param(solid, "solid", Solid)
check.inst_param(solid_handle, "solid_handle", SolidHandle)
check.list_param(step_inputs, "step_inputs", of_type=StepInput)
check.str_param(pipeline_name, "pipeline_name")
return ExecutionStep(
handle=StepHandle(solid_handle=solid_handle),
pipeline_name=pipeline_name,
step_inputs=step_inputs,
step_outputs=_create_step_outputs(solid, solid_handle,
environment_config),
compute_fn=lambda step_context, inputs: _execute_core_compute(
step_context.for_compute(), inputs, solid.definition.compute_fn),
solid=solid,
)
def __new__(
cls,
event_type_value,
pipeline_name,
step_handle=None,
solid_handle=None,
step_kind_value=None,
logging_tags=None,
event_specific_data=None,
message=None,
pid=None,
# legacy
step_key=None,
):
event_type_value, event_specific_data = _handle_back_compat(
event_type_value, event_specific_data
)
# old events may contain solid_handle but not step_handle
if solid_handle is not None and step_handle is None:
step_handle = StepHandle(solid_handle)
# Legacy events may have step_key set directly, preserve those to stay in sync
# with legacy execution plan snapshots.
if step_handle is not None and step_key is None:
step_key = step_handle.to_key()
return super(DagsterEvent, cls).__new__(
cls,
check.str_param(event_type_value, "event_type_value"),
check.str_param(pipeline_name, "pipeline_name"),
check.opt_inst_param(
step_handle, "step_handle", (StepHandle, ResolvedFromDynamicStepHandle)
),
check.opt_inst_param(solid_handle, "solid_handle", SolidHandle),
check.opt_str_param(step_kind_value, "step_kind_value"),
check.opt_dict_param(logging_tags, "logging_tags"),
_validate_event_specific_data(DagsterEventType(event_type_value), event_specific_data),
check.opt_str_param(message, "message"),
check.opt_int_param(pid, "pid"),
check.opt_str_param(step_key, "step_key"),
)
def _build_from_sorted_solids(
self,
solids: List[Solid],
dependency_structure: DependencyStructure,
parent_handle: Optional[SolidHandle] = None,
parent_step_inputs: Optional[List[
Union[StepInput, UnresolvedMappedStepInput,
UnresolvedCollectStepInput]]] = None,
):
for solid in solids:
handle = SolidHandle(solid.name, parent_handle)
### 1. INPUTS
# Create and add execution plan steps for solid inputs
has_unresolved_input = False
has_pending_input = False
step_inputs: List[Union[StepInput, UnresolvedMappedStepInput,
UnresolvedCollectStepInput]] = []
for input_name, input_def in solid.definition.input_dict.items():
step_input_source = get_step_input_source(
self,
solid,
input_name,
input_def,
dependency_structure,
handle,
parent_step_inputs,
)
# If an input with dagster_type "Nothing" doesn't have a value
# we don't create a StepInput
if step_input_source is None:
continue
if isinstance(
step_input_source,
(FromPendingDynamicStepOutput, FromUnresolvedStepOutput),
):
has_unresolved_input = True
step_inputs.append(
UnresolvedMappedStepInput(
name=input_name,
dagster_type_key=input_def.dagster_type.key,
source=step_input_source,
))
elif isinstance(step_input_source, FromDynamicCollect):
has_pending_input = True
step_inputs.append(
UnresolvedCollectStepInput(
name=input_name,
dagster_type_key=input_def.dagster_type.key,
source=step_input_source,
))
else:
check.inst_param(step_input_source, "step_input_source",
StepInputSource)
step_inputs.append(
StepInput(
name=input_name,
dagster_type_key=input_def.dagster_type.key,
source=step_input_source,
))
### 2a. COMPUTE FUNCTION
# Create and add execution plan step for the solid compute function
if isinstance(solid.definition, SolidDefinition):
step_outputs = create_step_outputs(solid, handle,
self.environment_config)
if has_pending_input and has_unresolved_input:
check.failed(
"Can not have pending and unresolved step inputs")
elif has_unresolved_input:
new_step: IExecutionStep = UnresolvedMappedExecutionStep(
handle=UnresolvedStepHandle(solid_handle=handle),
pipeline_name=self.pipeline_name,
step_inputs=cast(
List[Union[StepInput, UnresolvedMappedStepInput]],
step_inputs),
step_outputs=step_outputs,
tags=solid.tags,
)
elif has_pending_input:
new_step = UnresolvedCollectExecutionStep(
handle=StepHandle(solid_handle=handle),
pipeline_name=self.pipeline_name,
step_inputs=cast(
List[Union[StepInput, UnresolvedCollectStepInput]],
step_inputs),
step_outputs=step_outputs,
tags=solid.tags,
)
else:
new_step = ExecutionStep(
handle=StepHandle(solid_handle=handle),
pipeline_name=self.pipeline_name,
step_inputs=cast(List[StepInput], step_inputs),
step_outputs=step_outputs,
tags=solid.tags,
)
self.add_step(new_step)
### 2b. RECURSE
# Recurse over the solids contained in an instance of GraphDefinition
elif isinstance(solid.definition, GraphDefinition):
self._build_from_sorted_solids(
solid.definition.solids_in_topological_order,
solid.definition.dependency_structure,
parent_handle=handle,
parent_step_inputs=step_inputs,
)
else:
check.invariant(
False,
"Unexpected solid type {type} encountered during execution planning"
.format(type=type(solid.definition)),
)
### 3. OUTPUTS
# Create output handles for solid outputs
for name, output_def in solid.definition.output_dict.items():
output_handle = solid.output_handle(name)
# Punch through layers of composition scope to map to the output of the
# actual compute step
resolved_output_def, resolved_handle = solid.definition.resolve_output_to_origin(
output_def.name, handle)
step = self.get_step_by_solid_handle(resolved_handle)
if isinstance(step,
(ExecutionStep, UnresolvedCollectExecutionStep)):
step_output_handle: Union[
StepOutputHandle,
UnresolvedStepOutputHandle] = StepOutputHandle(
step.key, resolved_output_def.name)
elif isinstance(step, UnresolvedMappedExecutionStep):
step_output_handle = UnresolvedStepOutputHandle(
step.handle,
resolved_output_def.name,
step.resolved_by_step_key,
step.resolved_by_output_name,
)
else:
check.failed(f"Unexpected step type {step}")
self.set_output_handle(output_handle, step_output_handle)