Source code for flytekit.core.promise

from __future__ import annotations

import collections
import typing
from enum import Enum
from typing import Any, Dict, List, Optional, Tuple, Union, cast

from typing_extensions import Protocol

from flytekit.common import constants as _common_constants
from flytekit.common.exceptions import user as _user_exceptions
from flytekit.core import context_manager as _flyte_context
from flytekit.core import interface as flyte_interface
from flytekit.core import type_engine
from flytekit.core.context_manager import BranchEvalMode, ExecutionState, FlyteContext, FlyteContextManager
from flytekit.core.interface import Interface
from flytekit.core.node import Node
from flytekit.core.type_engine import DictTransformer, ListTransformer, TypeEngine
from flytekit.models import interface as _interface_models
from flytekit.models import literals as _literal_models
from flytekit.models import literals as _literals_models
from flytekit.models import types as _type_models
from flytekit.models import types as type_models
from flytekit.models.core import workflow as _workflow_model
from flytekit.models.literals import Primitive


def translate_inputs_to_literals(
    ctx: FlyteContext,
    incoming_values: Dict[str, Any],
    flyte_interface_types: Dict[str, _interface_models.Variable],
    native_types: Dict[str, type],
) -> Dict[str, _literal_models.Literal]:
    """
    The point of this function is to extract out Literals from a collection of either Python native values (which would
    be converted into Flyte literals) or Promises (the literals in which would just get extracted).

    When calling a task inside a workflow, a user might do something like this.

        def my_wf(in1: int) -> int:
            a = task_1(in1=in1)
            b = task_2(in1=5, in2=a)
            return b

    If this is the case, when task_2 is called in local workflow execution, we'll need to translate the Python native
    literal 5 to a Flyte literal.

    More interesting is this:

        def my_wf(in1: int, in2: int) -> int:
            a = task_1(in1=in1)
            b = task_2(in1=5, in2=[a, in2])
            return b

    Here, in task_2, during execution we'd have a list of Promises. We have to make sure to give task2 a Flyte
    LiteralCollection (Flyte's name for list), not a Python list of Flyte literals.

    This helper function is used both when sorting out inputs to a task, as well as outputs of a function.

    :param ctx: Context needed in case a non-primitive literal needs to be translated to a Flyte literal (like a file)
    :param incoming_values: This is a map of your task's input or wf's output kwargs basically
    :param flyte_interface_types: One side of an :py:class:`flytekit.models.interface.TypedInterface` basically.
    :param native_types: Map to native Python type.
    """

    def extract_value(
        ctx: FlyteContext, input_val: Any, val_type: type, flyte_literal_type: _type_models.LiteralType
    ) -> _literal_models.Literal:

        if isinstance(input_val, list):
            if flyte_literal_type.collection_type is None:
                raise TypeError(f"Not a collection type {flyte_literal_type} but got a list {input_val}")
            try:
                sub_type = ListTransformer.get_sub_type(val_type)
            except ValueError:
                if len(input_val) == 0:
                    raise
                sub_type = type(input_val[0])
            literal_list = [extract_value(ctx, v, sub_type, flyte_literal_type.collection_type) for v in input_val]
            return _literal_models.Literal(collection=_literal_models.LiteralCollection(literals=literal_list))
        elif isinstance(input_val, dict):
            if (
                flyte_literal_type.map_value_type is None
                and flyte_literal_type.simple != _type_models.SimpleType.STRUCT
            ):
                raise TypeError(f"Not a map type {flyte_literal_type} but got a map {input_val}")
            k_type, sub_type = DictTransformer.get_dict_types(val_type)  # type: ignore
            if flyte_literal_type.simple == _type_models.SimpleType.STRUCT:
                return TypeEngine.to_literal(ctx, input_val, type(input_val), flyte_literal_type)
            else:
                literal_map = {
                    k: extract_value(ctx, v, sub_type, flyte_literal_type.map_value_type) for k, v in input_val.items()
                }
                return _literal_models.Literal(map=_literal_models.LiteralMap(literals=literal_map))
        elif isinstance(input_val, Promise):
            # In the example above, this handles the "in2=a" type of argument
            return input_val.val
        elif isinstance(input_val, VoidPromise):
            raise AssertionError(
                f"Outputs of a non-output producing task {input_val.task_name} cannot be passed to another task."
            )
        elif isinstance(input_val, tuple):
            raise AssertionError(
                "Tuples are not a supported type for individual values in Flyte - got a tuple -"
                f" {input_val}. If using named tuple in an inner task, please, de-reference the"
                "actual attribute that you want to use. For example, in NamedTuple('OP', x=int) then"
                "return v.x, instead of v, even if this has a single element"
            )
        else:
            # This handles native values, the 5 example
            return TypeEngine.to_literal(ctx, input_val, val_type, flyte_literal_type)

    if incoming_values is None:
        raise ValueError("Incoming values cannot be None, must be a dict")

    result = {}  # So as to not overwrite the input_kwargs
    for k, v in incoming_values.items():
        if k not in flyte_interface_types:
            raise ValueError(f"Received unexpected keyword argument {k}")
        var = flyte_interface_types[k]
        t = native_types[k]
        result[k] = extract_value(ctx, v, t, var.type)

    return result


def get_primitive_val(prim: Primitive) -> Any:
    if prim.integer:
        return prim.integer
    if prim.datetime:
        return prim.datetime
    if prim.boolean:
        return prim.boolean
    if prim.duration:
        return prim.duration
    if prim.string_value:
        return prim.string_value
    return prim.float_value


class ConjunctionOps(Enum):
    AND = "and"
    OR = "or"


class ComparisonOps(Enum):
    EQ = "=="
    NE = "!="
    GT = ">"
    GE = ">="
    LT = "<"
    LE = "<="


_comparators = {
    ComparisonOps.EQ: lambda x, y: x == y,
    ComparisonOps.NE: lambda x, y: x != y,
    ComparisonOps.GT: lambda x, y: x > y,
    ComparisonOps.GE: lambda x, y: x >= y,
    ComparisonOps.LT: lambda x, y: x < y,
    ComparisonOps.LE: lambda x, y: x <= y,
}


class ComparisonExpression(object):
    """
    ComparisonExpression refers to an expression of the form (lhs operator rhs), where lhs and rhs are operands
    and operator can be any comparison expression like <, >, <=, >=, ==, !=
    """

    def __init__(self, lhs: Union["Promise", Any], op: ComparisonOps, rhs: Union["Promise", Any]):
        self._op = op
        self._lhs = None
        self._rhs = None
        if isinstance(lhs, Promise):
            self._lhs = lhs
            if lhs.is_ready:
                if lhs.val.scalar is None or lhs.val.scalar.primitive is None:
                    raise ValueError("Only primitive values can be used in comparison")
        if isinstance(rhs, Promise):
            self._rhs = rhs
            if rhs.is_ready:
                if rhs.val.scalar is None or rhs.val.scalar.primitive is None:
                    raise ValueError("Only primitive values can be used in comparison")
        if self._lhs is None:
            self._lhs = type_engine.TypeEngine.to_literal(FlyteContextManager.current_context(), lhs, type(lhs), None)
        if self._rhs is None:
            self._rhs = type_engine.TypeEngine.to_literal(FlyteContextManager.current_context(), rhs, type(rhs), None)

    @property
    def rhs(self) -> Union["Promise", _literal_models.Literal]:
        return self._rhs

    @property
    def lhs(self) -> Union["Promise", _literal_models.Literal]:
        return self._lhs

    @property
    def op(self) -> ComparisonOps:
        return self._op

    def eval(self) -> bool:
        if isinstance(self.lhs, Promise):
            lhs = self.lhs.eval()
        else:
            lhs = get_primitive_val(self.lhs.scalar.primitive)

        if isinstance(self.rhs, Promise):
            rhs = self.rhs.eval()
        else:
            rhs = get_primitive_val(self.rhs.scalar.primitive)

        return _comparators[self.op](lhs, rhs)

    def __and__(self, other):
        return ConjunctionExpression(lhs=self, op=ConjunctionOps.AND, rhs=other)

    def __or__(self, other):
        return ConjunctionExpression(lhs=self, op=ConjunctionOps.OR, rhs=other)

    def __bool__(self):
        raise ValueError(
            "Cannot perform truth value testing,"
            " This is a limitation in python. For Logical `and\\or` use `&\\|` (bitwise) instead."
            f" Expr {self}"
        )

    def __repr__(self):
        return f"Comp({self._lhs} {self._op.value} {self._rhs})"

    def __str__(self):
        return self.__repr__()


class ConjunctionExpression(object):
    """
    A Conjunction Expression is an expression of the form either (A and B) or (A or B).
    where A, B are two expressions (comparison or conjunctions) and (and, or) are logical truth operators.

    A conjunctionExpression evaluates to True or False depending on the logical operator and the truth values of
    each of the expressions A & B
    """

    def __init__(
        self,
        lhs: Union[ComparisonExpression, "ConjunctionExpression"],
        op: ConjunctionOps,
        rhs: Union[ComparisonExpression, "ConjunctionExpression"],
    ):
        self._lhs = lhs
        self._rhs = rhs
        self._op = op

    @property
    def rhs(self) -> Union[ComparisonExpression, "ConjunctionExpression"]:
        return self._rhs

    @property
    def lhs(self) -> Union[ComparisonExpression, "ConjunctionExpression"]:
        return self._lhs

    @property
    def op(self) -> ConjunctionOps:
        return self._op

    def eval(self) -> bool:
        l_eval = self.lhs.eval()
        if self.op == ConjunctionOps.AND and l_eval is False:
            return False

        if self.op == ConjunctionOps.OR and l_eval is True:
            return True

        r_eval = self.rhs.eval()
        if self.op == ConjunctionOps.AND:
            return l_eval and r_eval

        return l_eval or r_eval

    def __and__(self, other: Union[ComparisonExpression, "ConjunctionExpression"]):
        return ConjunctionExpression(lhs=self, op=ConjunctionOps.AND, rhs=other)

    def __or__(self, other: Union[ComparisonExpression, "ConjunctionExpression"]):
        return ConjunctionExpression(lhs=self, op=ConjunctionOps.OR, rhs=other)

    def __bool__(self):
        raise ValueError(
            "Cannot perform truth value testing,"
            " This is a limitation in python. For Logical `and\\or` use `&\\|` (bitwise) instead. Refer to: PEP-335"
        )

    def __repr__(self):
        return f"( {self._lhs} {self._op} {self._rhs} )"

    def __str__(self):
        return self.__repr__()


# TODO: The NodeOutput object, which this Promise wraps, has an sdk_type. Since we're no longer using sdk types,
#  we should consider adding a literal type to this object as well for downstream checking when Bindings are created.
[docs]class Promise(object): """ This object is a wrapper and exists for three main reasons. Let's assume we're dealing with a task like :: @task def t1() -> (int, str): ... #. Handling the duality between compilation and local execution - when the task function is run in a local execution mode inside a workflow function, a Python integer and string are produced. When the task is being compiled as part of the workflow, the task call creates a Node instead, and the task returns two Promise objects that point to that Node. #. One needs to be able to call :: x = t1().with_overrides(...) If the task returns an integer or a ``(int, str)`` tuple like ``t1`` above, calling ``with_overrides`` on the result would throw an error. This Promise object adds that. #. Assorted handling for conditionals. """ # TODO: Currently, NodeOutput we're creating is the slimmer core package Node class, but since only the # id is used, it's okay for now. Let's clean all this up though.
[docs] def __init__(self, var: str, val: Union[NodeOutput, _literal_models.Literal]): self._var = var self._promise_ready = True self._val = val if val and isinstance(val, NodeOutput): self._ref = val self._promise_ready = False self._val = None
def __hash__(self): return hash(id(self)) def with_var(self, new_var: str) -> Promise: if self.is_ready: return Promise(var=new_var, val=self.val) return Promise(var=new_var, val=self.ref) @property def is_ready(self) -> bool: """ Returns if the Promise is READY (is not a reference and the val is actually ready) Usage: p = Promise(...) ... if p.is_ready(): print(p.val) else: print(p.ref) """ return self._promise_ready @property def val(self) -> _literal_models.Literal: """ If the promise is ready then this holds the actual evaluate value in Flyte's type system """ return self._val @property def ref(self) -> NodeOutput: """ If the promise is NOT READY / Incomplete, then it maps to the origin node that owns the promise """ return self._ref @property def var(self) -> str: """ Name of the variable bound with this promise """ return self._var def eval(self) -> Any: if not self._promise_ready or self._val is None: raise ValueError("Cannot Eval with incomplete promises") if self.val.scalar is None or self.val.scalar.primitive is None: raise ValueError("Eval can be invoked for primitive types only") return get_primitive_val(self.val.scalar.primitive) def is_(self, v: bool) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.EQ, v) def is_false(self) -> ComparisonExpression: return self.is_(False) def is_true(self): return self.is_(True) def __eq__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.EQ, other) def __ne__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.NE, other) def __gt__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.GT, other) def __ge__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.GE, other) def __lt__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.LT, other) def __le__(self, other) -> ComparisonExpression: return ComparisonExpression(self, ComparisonOps.LE, other) def __bool__(self): raise ValueError( "Flytekit does not support Unary expressions or performing truth value testing," " This is a limitation in python. For Logical `and\\or` use `&\\|` (bitwise) instead" ) def __and__(self, other): raise ValueError("Cannot perform Logical AND of Promise with other") def __or__(self, other): raise ValueError("Cannot perform Logical OR of Promise with other") def with_overrides(self, *args, **kwargs): if not self.is_ready: # TODO, this should be forwarded, but right now this results in failure and we want to test this behavior self.ref.node.with_overrides(*args, **kwargs) return self def __repr__(self): if self._promise_ready: return f"Resolved({self._var}={self._val})" return f"Promise(node:{self.ref.node_id}.{self._var})" def __str__(self): return str(self.__repr__())
def create_native_named_tuple( ctx: FlyteContext, promises: Optional[Union[Promise, typing.List[Promise]]], entity_interface: Interface ) -> Optional[Tuple]: """ Creates and returns a Named tuple with all variables that match the expected named outputs. this makes it possible to run things locally and expect a more native behavior, i.e. address elements of a named tuple by name. """ if entity_interface is None: raise ValueError("Interface of the entity is required to generate named outputs") if promises is None: return None if isinstance(promises, Promise): k, v = [(k, v) for k, v in entity_interface.outputs.items()][0] # get output native type try: return TypeEngine.to_python_value(ctx, promises.val, v) except Exception as e: raise AssertionError(f"Failed to convert value of output {k}, expected type {v}.") from e if len(promises) == 0: return None named_tuple_name = "DefaultNamedTupleOutput" if entity_interface.output_tuple_name: named_tuple_name = entity_interface.output_tuple_name outputs = {} for p in promises: if not isinstance(p, Promise): raise AssertionError( "Workflow outputs can only be promises that are returned by tasks. Found a value of" f"type {type(p)}. Workflows cannot return local variables or constants." ) t = entity_interface.outputs[p.var] try: outputs[p.var] = TypeEngine.to_python_value(ctx, p.val, t) except Exception as e: raise AssertionError(f"Failed to convert value of output {p.var}, expected type {t}.") from e # Should this class be part of the Interface? t = collections.namedtuple(named_tuple_name, list(outputs.keys())) return t(**outputs) # To create a class that is a named tuple, we might have to create namedtuplemeta and manipulate the tuple def create_task_output( promises: Optional[Union[List[Promise], Promise]], entity_interface: Optional[Interface] = None ) -> Optional[Union[Tuple[Promise], Promise]]: # TODO: Add VoidPromise here to simplify things at call site. Consider returning for [] below as well instead of # raising an exception. if promises is None: return None if isinstance(promises, Promise): return promises if len(promises) == 0: raise Exception( "This function should not be called with an empty list. It should have been handled with a" "VoidPromise at this function's call-site." ) if len(promises) == 1: if not entity_interface: return promises[0] # See transform_signature_to_interface for more information, we're using the existence of a name as a proxy # for the user having specified a one-element typing.NamedTuple, which means we should _not_ extract it. We # should still return a tuple but it should be one of ours. if not entity_interface.output_tuple_name: return promises[0] # More than one promise, let us wrap it into a tuple # Start with just the var names in the promises variables = [p.var for p in promises] # These should be OrderedDicts so it should be safe to iterate over the keys. if entity_interface: variables = [k for k in entity_interface.outputs.keys()] named_tuple_name = "DefaultNamedTupleOutput" if entity_interface and entity_interface.output_tuple_name: named_tuple_name = entity_interface.output_tuple_name # Should this class be part of the Interface? class Output(collections.namedtuple(named_tuple_name, variables)): def with_overrides(self, *args, **kwargs): val = self.__getattribute__(self._fields[0]) val.with_overrides(*args, **kwargs) return self @property def ref(self): for p in promises: if p.ref: return p.ref return None def runs_before(self, other: Any): """ This function is just here to allow local workflow execution to run. See the corresponding function in flytekit.core.node.Node for more information. Local workflow execution in the manual ``create_node`` paradigm is already determined by the order in which the nodes were created. """ # TODO: If possible, add a check and raise an Exception if create_node was not called in the correct order. return self def __rshift__(self, other: Any): # See comment for runs_before return self return Output(*promises) # type: ignore def binding_data_from_python_std( ctx: _flyte_context.FlyteContext, expected_literal_type: _type_models.LiteralType, t_value: typing.Any, t_value_type: type, ) -> _literals_models.BindingData: # This handles the case where the given value is the output of another task if isinstance(t_value, Promise): if not t_value.is_ready: return _literals_models.BindingData(promise=t_value.ref) elif isinstance(t_value, VoidPromise): raise AssertionError( f"Cannot pass output from task {t_value.task_name} that produces no outputs to a downstream task" ) elif isinstance(t_value, list): if expected_literal_type.collection_type is None: raise AssertionError(f"this should be a list and it is not: {type(t_value)} vs {expected_literal_type}") sub_type = ListTransformer.get_sub_type(t_value_type) collection = _literals_models.BindingDataCollection( bindings=[ binding_data_from_python_std(ctx, expected_literal_type.collection_type, t, sub_type) for t in t_value ] ) return _literals_models.BindingData(collection=collection) elif isinstance(t_value, dict): if ( expected_literal_type.map_value_type is None and expected_literal_type.simple != _type_models.SimpleType.STRUCT ): raise AssertionError( f"this should be a Dictionary type and it is not: {type(t_value)} vs {expected_literal_type}" ) k_type, v_type = DictTransformer.get_dict_types(t_value_type) if expected_literal_type.simple == _type_models.SimpleType.STRUCT: lit = TypeEngine.to_literal(ctx, t_value, type(t_value), expected_literal_type) return _literals_models.BindingData(scalar=lit.scalar) else: m = _literals_models.BindingDataMap( bindings={ k: binding_data_from_python_std(ctx, expected_literal_type.map_value_type, v, v_type) for k, v in t_value.items() } ) return _literals_models.BindingData(map=m) elif isinstance(t_value, tuple): raise AssertionError( "Tuples are not a supported type for individual values in Flyte - got a tuple -" f" {t_value}. If using named tuple in an inner task, please, de-reference the" "actual attribute that you want to use. For example, in NamedTuple('OP', x=int) then" "return v.x, instead of v, even if this has a single element" ) # This is the scalar case - e.g. my_task(in1=5) scalar = TypeEngine.to_literal(ctx, t_value, t_value_type, expected_literal_type).scalar return _literals_models.BindingData(scalar=scalar) def binding_from_python_std( ctx: _flyte_context.FlyteContext, var_name: str, expected_literal_type: _type_models.LiteralType, t_value: typing.Any, t_value_type: type, ) -> _literals_models.Binding: binding_data = binding_data_from_python_std(ctx, expected_literal_type, t_value, t_value_type) return _literals_models.Binding(var=var_name, binding=binding_data) def to_binding(p: Promise) -> _literals_models.Binding: return _literals_models.Binding(var=p.var, binding=_literals_models.BindingData(promise=p.ref)) class VoidPromise(object): """ This object is returned for tasks that do not return any outputs (declared interface is empty) VoidPromise cannot be interacted with and does not allow comparisons or any operations """ def __init__(self, task_name: str): self._task_name = task_name def runs_before(self, *args, **kwargs): """ This is a placeholder and should do nothing. It is only here to enable local execution of workflows where a task returns nothing. """ def __rshift__(self, *args, **kwargs): ... # See runs_before @property def task_name(self): return self._task_name def __eq__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __and__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __or__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __le__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __ge__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __gt__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __lt__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __add__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __cmp__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __bool__(self): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __mod__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __xor__(self, other): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __str__(self): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") def __repr__(self): raise AssertionError(f"Task {self._task_name} returns nothing, NoneType return cannot be used") class NodeOutput(type_models.OutputReference): def __init__(self, node: Node, var: str): """ :param node: :param var: The name of the variable this NodeOutput references """ self._node = node super(NodeOutput, self).__init__(self._node.id, var) @property def node_id(self): """ Override the underlying node_id property to refer to SdkNode. :rtype: Text """ return self.node.id @property def node(self) -> Node: """Return Node object.""" return self._node def __repr__(self) -> str: s = f"Node({self.node if self.node.id is not None else None}:{self.var})" return s class SupportsNodeCreation(Protocol): @property def name(self) -> str: ... @property def python_interface(self) -> flyte_interface.Interface: ... def construct_node_metadata(self) -> _workflow_model.NodeMetadata: ... def extract_obj_name(name: str) -> str: """ Generates a shortened name, without the module information. Useful for node-names etc. Only extracts the final object information often separated by `.` in the python fully qualified notation """ if name is None: return "" if "." in name: return name.split(".")[-1] return name def create_and_link_node( ctx: FlyteContext, entity: SupportsNodeCreation, **kwargs, ): """ This method is used to generate a node with bindings. This is not used in the execution path. """ if ctx.compilation_state is None: raise _user_exceptions.FlyteAssertion("Cannot create node when not compiling...") used_inputs = set() bindings = [] interface = entity.python_interface typed_interface = flyte_interface.transform_interface_to_typed_interface(interface) # Mypy needs some extra help to believe that `typed_interface` will not be `None` assert typed_interface is not None for k in sorted(interface.inputs): var = typed_interface.inputs[k] if k not in kwargs: raise _user_exceptions.FlyteAssertion("Input was not specified for: {} of type {}".format(k, var.type)) v = kwargs[k] # This check ensures that tuples are not passed into a function, as tuples are not supported by Flyte # Usually a Tuple will indicate that multiple outputs from a previous task were accidentally passed # into the function. if isinstance(v, tuple): raise AssertionError( f"Variable({k}) for function({entity.name}) cannot receive a multi-valued tuple {v}." f" Check if the predecessor function returning more than one value?" ) try: bindings.append( binding_from_python_std( ctx, var_name=k, expected_literal_type=var.type, t_value=v, t_value_type=interface.inputs[k] ) ) used_inputs.add(k) except Exception as e: raise AssertionError(f"Failed to Bind variable {k} for function {entity.name}.") from e extra_inputs = used_inputs ^ set(kwargs.keys()) if len(extra_inputs) > 0: raise _user_exceptions.FlyteAssertion( "Too many inputs were specified for the interface. Extra inputs were: {}".format(extra_inputs) ) # Detect upstream nodes # These will be our core Nodes until we can amend the Promise to use NodeOutputs that reference our Nodes upstream_nodes = list( set( [ input_val.ref.node for input_val in kwargs.values() if isinstance(input_val, Promise) and input_val.ref.node_id != _common_constants.GLOBAL_INPUT_NODE_ID ] ) ) flytekit_node = Node( # TODO: Better naming, probably a derivative of the function name. id=f"{ctx.compilation_state.prefix}n{len(ctx.compilation_state.nodes)}", metadata=entity.construct_node_metadata(), bindings=sorted(bindings, key=lambda b: b.var), upstream_nodes=upstream_nodes, flyte_entity=entity, ) ctx.compilation_state.add_node(flytekit_node) if len(typed_interface.outputs) == 0: return VoidPromise(entity.name) # Create a node output object for each output, they should all point to this node of course. node_outputs = [] for output_name, output_var_model in typed_interface.outputs.items(): # TODO: If node id gets updated later, we have to make sure to update the NodeOutput model's ID, which # is currently just a static str node_outputs.append(Promise(output_name, NodeOutput(node=flytekit_node, var=output_name))) # Don't print this, it'll crash cuz sdk_node._upstream_node_ids might be None, but idl code will break return create_task_output(node_outputs, interface) class LocallyExecutable(Protocol): def local_execute(self, ctx: FlyteContext, **kwargs) -> Union[Tuple[Promise], Promise, VoidPromise]: ... def flyte_entity_call_handler(entity: Union[SupportsNodeCreation], *args, **kwargs): """ This function is the call handler for tasks, workflows, and launch plans (which redirects to the underlying workflow). The logic is the same for all three, but we did not want to create base class, hence this separate method. When one of these entities is () aka __called__, there are three things we may do: #. Compilation Mode - this happens when the function is called as part of a workflow (potentially dynamic task?). Instead of running the user function, produce promise objects and create a node. #. Workflow Execution Mode - when a workflow is being run locally. Even though workflows are functions and everything should be able to be passed through naturally, we'll want to wrap output values of the function into objects, so that potential .with_cpu or other ancillary functions can be attached to do nothing. Subsequent tasks will have to know how to unwrap these. If by chance a non-Flyte task uses a task output as an input, things probably will fail pretty obviously. #. Start a local execution - This means that we're not already in a local workflow execution, which means that we should expect inputs to be native Python values and that we should return Python native values. """ # Sanity checks # Only keyword args allowed if len(args) > 0: raise _user_exceptions.FlyteAssertion( f"When calling tasks, only keyword args are supported. " f"Aborting execution as detected {len(args)} positional args {args}" ) # Make sure arguments are part of interface for k, v in kwargs.items(): if k not in cast(SupportsNodeCreation, entity).python_interface.inputs: raise ValueError( f"Received unexpected keyword argument {k} in function {cast(SupportsNodeCreation, entity).name}" ) ctx = FlyteContextManager.current_context() if ctx.compilation_state is not None and ctx.compilation_state.mode == 1: return create_and_link_node(ctx, entity=entity, **kwargs) elif ctx.execution_state is not None and ctx.execution_state.mode == ExecutionState.Mode.LOCAL_WORKFLOW_EXECUTION: if ctx.execution_state.branch_eval_mode == BranchEvalMode.BRANCH_SKIPPED: if ( len(cast(SupportsNodeCreation, entity).python_interface.inputs) > 0 or len(cast(SupportsNodeCreation, entity).python_interface.outputs) > 0 ): output_names = list(cast(SupportsNodeCreation, entity).python_interface.outputs.keys()) if len(output_names) == 0: return VoidPromise(entity.name) vals = [Promise(var, None) for var in output_names] return create_task_output(vals, cast(SupportsNodeCreation, entity).python_interface) else: return None return cast(LocallyExecutable, entity).local_execute(ctx, **kwargs) else: with FlyteContextManager.with_context( ctx.with_execution_state( ctx.new_execution_state().with_params(mode=ExecutionState.Mode.LOCAL_WORKFLOW_EXECUTION) ) ) as child_ctx: result = cast(LocallyExecutable, entity).local_execute(child_ctx, **kwargs) expected_outputs = len(cast(SupportsNodeCreation, entity).python_interface.outputs) if expected_outputs == 0: if result is None or isinstance(result, VoidPromise): return None else: raise Exception(f"Received an output when workflow local execution expected None. Received: {result}") if (1 < expected_outputs == len(result)) or (result is not None and expected_outputs == 1): return create_native_named_tuple(ctx, result, cast(SupportsNodeCreation, entity).python_interface) raise ValueError( f"Expected outputs and actual outputs do not match." f"Result {result}. " f"Python interface: {cast(SupportsNodeCreation, entity).python_interface}" )