def test_step_handles():
plain = StepHandle.parse_from_key("foo")
assert isinstance(plain, StepHandle)
unresolved = StepHandle.parse_from_key("foo[?]")
assert isinstance(unresolved, UnresolvedStepHandle)
resolved = StepHandle.parse_from_key("foo[bar]")
assert isinstance(resolved, ResolvedFromDynamicStepHandle)
assert unresolved.resolve("bar") == resolved
assert resolved.unresolved_form == unresolved
def build_subset_plan(self,
step_keys_to_execute: List[str]) -> "ExecutionPlan":
check.list_param(step_keys_to_execute,
"step_keys_to_execute",
of_type=str)
step_handles_to_execute = [
StepHandle.parse_from_key(key) for key in step_keys_to_execute
]
bad_keys = []
for handle in step_handles_to_execute:
if handle not in self.step_dict:
bad_keys.append(handle.to_key())
if bad_keys:
raise DagsterExecutionStepNotFoundError(
f"Can not build subset plan from unknown step{'s' if len(bad_keys)> 1 else ''}: {', '.join(bad_keys)}",
step_keys=bad_keys,
)
return ExecutionPlan(
self.pipeline,
self.step_dict,
step_handles_to_execute,
self.environment_config,
self.known_state,
)
def build_subset_plan(
self,
step_keys_to_execute: List[str],
pipeline_def: PipelineDefinition,
environment_config: EnvironmentConfig,
) -> "ExecutionPlan":
check.list_param(step_keys_to_execute,
"step_keys_to_execute",
of_type=str)
step_handles_to_execute = [
StepHandle.parse_from_key(key) for key in step_keys_to_execute
]
bad_keys = []
for handle in step_handles_to_execute:
if handle not in self.step_dict:
bad_keys.append(handle.to_key())
if bad_keys:
raise DagsterExecutionStepNotFoundError(
f"Can not build subset plan from unknown step{'s' if len(bad_keys)> 1 else ''}: {', '.join(bad_keys)}",
step_keys=bad_keys,
)
executable_map, resolvable_map = _compute_step_maps(
self.step_dict,
step_handles_to_execute,
self.known_state,
)
return ExecutionPlan(
self.step_dict,
executable_map,
resolvable_map,
step_handles_to_execute,
self.known_state,
_compute_artifacts_persisted(
self.step_dict,
step_handles_to_execute,
pipeline_def,
environment_config,
executable_map,
),
)
def build_subset_plan(self,
step_keys_to_execute: List[str]) -> "ExecutionPlan":
check.list_param(step_keys_to_execute,
"step_keys_to_execute",
of_type=str)
step_handles_to_execute = [
StepHandle.parse_from_key(key) for key in step_keys_to_execute
]
bad_keys = []
for handle in step_handles_to_execute:
if handle in self.step_dict:
pass # no further processing required
elif (isinstance(handle, ResolvedFromDynamicStepHandle)
and handle.unresolved_form in self.step_dict):
unresolved_step = cast(UnresolvedExecutionStep,
self.step_dict[handle.unresolved_form])
# self.step_dict updated as side effect
self.resolve(
unresolved_step.resolved_by_step_key,
{
unresolved_step.resolved_by_output_name:
[handle.mapping_key]
},
)
check.invariant(
handle in self.step_dict,
f"Handle did not resolve as expected, not found in step dict {handle}",
)
else:
bad_keys.append(handle.to_key())
if bad_keys:
raise DagsterExecutionStepNotFoundError(
f"Can not build subset plan from unknown step{'s' if len(bad_keys)> 1 else ''}: {', '.join(bad_keys)}",
step_keys=bad_keys,
)
return ExecutionPlan(
self.pipeline,
self.step_dict,
step_handles_to_execute,
self.environment_config,
)
def rebuild_from_snapshot(pipeline_name, execution_plan_snapshot):
if not execution_plan_snapshot.can_reconstruct_plan:
raise DagsterInvariantViolationError(
"Tried to reconstruct an old ExecutionPlanSnapshot that was created before snapshots "
"had enough information to fully reconstruct the ExecutionPlan"
)
step_dict = {}
for step_snap in execution_plan_snapshot.steps:
input_snaps = step_snap.inputs
output_snaps = step_snap.outputs
step_inputs = [
ExecutionPlan.rebuild_step_input(step_input_snap)
for step_input_snap in input_snaps
]
step_outputs = [
StepOutput(
step_output_snap.solid_handle,
step_output_snap.name,
step_output_snap.dagster_type_key,
step_output_snap.properties,
) for step_output_snap in output_snaps
]
if step_snap.kind == StepKind.COMPUTE:
step = ExecutionStep(
step_snap.step_handle,
pipeline_name,
step_inputs,
step_outputs,
step_snap.tags,
)
elif step_snap.kind == StepKind.UNRESOLVED_MAPPED:
step = UnresolvedMappedExecutionStep(
step_snap.step_handle,
pipeline_name,
step_inputs,
step_outputs,
step_snap.tags,
)
elif step_snap.kind == StepKind.UNRESOLVED_COLLECT:
step = UnresolvedCollectExecutionStep(
step_snap.step_handle,
pipeline_name,
step_inputs,
step_outputs,
step_snap.tags,
)
else:
raise Exception(f"Unexpected step kind {str(step_snap.kind)}")
step_dict[step.handle] = step
step_handles_to_execute = [
StepHandle.parse_from_key(key)
for key in execution_plan_snapshot.step_keys_to_execute
]
executable_map, resolvable_map = _compute_step_maps(
step_dict,
step_handles_to_execute,
execution_plan_snapshot.initial_known_state,
)
return ExecutionPlan(
step_dict,
executable_map,
resolvable_map,
step_handles_to_execute,
execution_plan_snapshot.initial_known_state,
execution_plan_snapshot.artifacts_persisted,
)
def has_step(self, key):
check.str_param(key, "key")
handle = StepHandle.parse_from_key(key)
if isinstance(handle, ResolvedFromDynamicStepHandle):
return handle.unresolved_form.to_key() in self._step_index
return key in self._step_index
def get_retry_steps_from_execution_plan(instance, execution_plan, parent_run_id):
check.inst_param(instance, "instance", DagsterInstance)
check.inst_param(execution_plan, "execution_plan", ExternalExecutionPlan)
check.opt_str_param(parent_run_id, "parent_run_id")
if not parent_run_id:
return execution_plan.step_keys_in_plan
parent_run = instance.get_run_by_id(parent_run_id)
parent_run_logs = instance.all_logs(parent_run_id)
# keep track of steps with dicts that point:
# * step_key -> set(step_handle) in the normal case
# * unresolved_step_key -> set(resolved_step_handle, ...) for dynamic outputs
all_steps_in_parent_run_logs = defaultdict(set)
failed_steps_in_parent_run_logs = defaultdict(set)
successful_steps_in_parent_run_logs = defaultdict(set)
interrupted_steps_in_parent_run_logs = defaultdict(set)
skipped_steps_in_parent_run_logs = defaultdict(set)
for record in parent_run_logs:
if record.dagster_event and record.dagster_event.step_handle:
step_handle = record.dagster_event.step_handle
_update_tracking_dict(all_steps_in_parent_run_logs, step_handle)
if record.dagster_event_type == DagsterEventType.STEP_FAILURE:
_update_tracking_dict(failed_steps_in_parent_run_logs, step_handle)
if record.dagster_event_type == DagsterEventType.STEP_SUCCESS:
_update_tracking_dict(successful_steps_in_parent_run_logs, step_handle)
if record.dagster_event_type == DagsterEventType.STEP_SKIPPED:
_update_tracking_dict(skipped_steps_in_parent_run_logs, step_handle)
for step_set in all_steps_in_parent_run_logs.values():
for step_handle in step_set:
if (
not _in_tracking_dict(step_handle, failed_steps_in_parent_run_logs)
and not _in_tracking_dict(step_handle, successful_steps_in_parent_run_logs)
and not _in_tracking_dict(step_handle, skipped_steps_in_parent_run_logs)
):
_update_tracking_dict(interrupted_steps_in_parent_run_logs, step_handle)
to_retry = defaultdict(set)
execution_deps = execution_plan.execution_deps()
for step_snap in execution_plan.topological_steps():
step_key = step_snap.key
step_handle = StepHandle.parse_from_key(step_snap.key)
if parent_run.step_keys_to_execute and step_snap.key not in parent_run.step_keys_to_execute:
continue
if step_snap.key in failed_steps_in_parent_run_logs:
to_retry[step_key].update(failed_steps_in_parent_run_logs[step_key])
# Interrupted steps can occur when graceful cleanup from a step failure fails to run,
# and a step failure event is not generated
if step_key in interrupted_steps_in_parent_run_logs:
to_retry[step_key].update(interrupted_steps_in_parent_run_logs[step_key])
# Missing steps did not execute, e.g. when a run was terminated
if step_key not in all_steps_in_parent_run_logs:
to_retry[step_key].add(step_handle)
step_dep_keys = execution_deps[step_key]
retrying_dep_keys = step_dep_keys.intersection(to_retry.keys())
# this step is downstream of a step we are about to retry
if retrying_dep_keys:
for retrying_key in retrying_dep_keys:
# If this step and its ancestor are both downstream of a dynamic output,
# add resolved instances of this step for the retrying mapping keys
if isinstance(step_handle, UnresolvedStepHandle) and all(
map(
lambda handle: isinstance(handle, ResolvedFromDynamicStepHandle),
to_retry[retrying_key],
)
):
for resolved_handle in to_retry[retrying_key]:
to_retry[step_key].add(step_handle.resolve(resolved_handle.mapping_key))
else:
to_retry[step_key].add(step_handle)
return [step_handle.to_key() for step_set in to_retry.values() for step_handle in step_set]
def get_retry_steps_from_parent_run(
instance,
parent_run_id: str = None,
parent_run: PipelineRun = None
) -> Tuple[List[str], Optional[KnownExecutionState]]:
check.inst_param(instance, "instance", DagsterInstance)
check.invariant(
bool(parent_run_id) != bool(parent_run),
"Must provide one of parent_run_id or parent_run")
check.opt_str_param(parent_run_id, "parent_run_id")
check.opt_inst_param(parent_run, "parent_run", PipelineRun)
parent_run = parent_run or instance.get_run_by_id(parent_run_id)
parent_run_id = parent_run.run_id
parent_run_logs = instance.all_logs(parent_run_id)
execution_plan_snapshot = instance.get_execution_plan_snapshot(
parent_run.execution_plan_snapshot_id)
if not execution_plan_snapshot:
raise DagsterExecutionPlanSnapshotNotFoundError(
f"Could not load execution plan snapshot for run {parent_run_id}")
execution_plan = ExternalExecutionPlan(
execution_plan_snapshot=execution_plan_snapshot)
# keep track of steps with dicts that point:
# * step_key -> set(step_handle) in the normal case
# * unresolved_step_key -> set(resolved_step_handle, ...) for dynamic outputs
all_steps_in_parent_run_logs: Dict[str, set] = defaultdict(set)
failed_steps_in_parent_run_logs: Dict[str, set] = defaultdict(set)
successful_steps_in_parent_run_logs: Dict[str, set] = defaultdict(set)
interrupted_steps_in_parent_run_logs: Dict[str, set] = defaultdict(set)
skipped_steps_in_parent_run_logs: Dict[str, set] = defaultdict(set)
for record in parent_run_logs:
if record.dagster_event and record.dagster_event.step_handle:
step_handle = record.dagster_event.step_handle
_update_tracking_dict(all_steps_in_parent_run_logs, step_handle)
if record.dagster_event_type == DagsterEventType.STEP_FAILURE:
_update_tracking_dict(failed_steps_in_parent_run_logs,
step_handle)
if record.dagster_event_type == DagsterEventType.STEP_SUCCESS:
_update_tracking_dict(successful_steps_in_parent_run_logs,
step_handle)
if record.dagster_event_type == DagsterEventType.STEP_SKIPPED:
_update_tracking_dict(skipped_steps_in_parent_run_logs,
step_handle)
for step_set in all_steps_in_parent_run_logs.values():
for step_handle in step_set:
if (not _in_tracking_dict(step_handle,
failed_steps_in_parent_run_logs)
and not _in_tracking_dict(
step_handle, successful_steps_in_parent_run_logs)
and not _in_tracking_dict(
step_handle, skipped_steps_in_parent_run_logs)):
_update_tracking_dict(interrupted_steps_in_parent_run_logs,
step_handle)
to_retry = defaultdict(set)
execution_deps = execution_plan.execution_deps()
for step_snap in execution_plan.topological_steps():
step_key = step_snap.key
step_handle = StepHandle.parse_from_key(step_snap.key)
if parent_run.step_keys_to_execute and step_snap.key not in parent_run.step_keys_to_execute:
continue
if step_snap.key in failed_steps_in_parent_run_logs:
to_retry[step_key].update(
failed_steps_in_parent_run_logs[step_key])
# Interrupted steps can occur when graceful cleanup from a step failure fails to run,
# and a step failure event is not generated
if step_key in interrupted_steps_in_parent_run_logs:
to_retry[step_key].update(
interrupted_steps_in_parent_run_logs[step_key])
# Missing steps did not execute, e.g. when a run was terminated
if step_key not in all_steps_in_parent_run_logs:
to_retry[step_key].add(step_handle)
step_dep_keys = execution_deps[step_key]
retrying_dep_keys = step_dep_keys.intersection(to_retry.keys())
# this step is downstream of a step we are about to retry
if retrying_dep_keys:
for retrying_key in retrying_dep_keys:
# If this step and its ancestor are both downstream of a dynamic output,
# add resolved instances of this step for the retrying mapping keys
if isinstance(step_handle, UnresolvedStepHandle) and all(
map(
lambda handle: isinstance(
handle, ResolvedFromDynamicStepHandle),
to_retry[retrying_key],
)):
for resolved_handle in to_retry[retrying_key]:
to_retry[step_key].add(
step_handle.resolve(resolved_handle.mapping_key))
else:
to_retry[step_key].add(step_handle)
steps_to_retry = [
step_handle.to_key() for step_set in to_retry.values()
for step_handle in step_set
]
return steps_to_retry, KnownExecutionState.for_reexecution(
parent_run_logs, steps_to_retry)