# Copyright 2021 The Kubeflow Authors
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Classes and methods that supports argument for ParallelFor."""
import re
from typing import Any, Dict, List, Optional, Union
from kfp.dsl import pipeline_channel
ItemList = List[Union[int, float, str, Dict[str, Any]]]
LOOP_ITEM_NAME_BASE = 'loop-item'
LOOP_ITEM_PARAM_NAME_BASE = 'loop-item-param'
def _get_loop_item_type(type_name: str) -> Optional[str]:
"""Extracts the loop item type.
This method is used for extract the item type from a collection type.
For example:
List[str] -> str
typing.List[int] -> int
typing.Sequence[str] -> str
List -> None
str -> None
Args:
type_name: The collection type name, like `List`, Sequence`, etc.
Returns:
The collection item type or None if no match found.
"""
match = re.match('(typing\.)?(?:\w+)(?:\[(?P<item_type>.+)\])', type_name)
return match['item_type'].lstrip().rstrip() if match else None
def _get_subvar_type(type_name: str) -> Optional[str]:
"""Extracts the subvar type.
This method is used for extract the value type from a dictionary type.
For example:
Dict[str, int] -> int
typing.Mapping[str, float] -> float
Args:
type_name: The dictionary type.
Returns:
The dictionary value type or None if no match found.
"""
match = re.match(
'(typing\.)?(?:\w+)(?:\[\s*(?:\w+)\s*,\s*(?P<value_type>.+)\])',
type_name)
return match['value_type'].lstrip().rstrip() if match else None
def _get_first_element_type(item_list: ItemList) -> str:
"""Returns the type of the first element of ItemList.
Args:
item_list: List of items to loop over. If a list of dicts then, all dicts must have the same keys.
Returns:
A string representing the type of the first element (e.g., "int", "Dict[str, int]").
"""
first_element = item_list[0]
if isinstance(first_element, dict):
key_type = type(list(
first_element.keys())[0]).__name__ # Get type of first key
value_type = type(list(
first_element.values())[0]).__name__ # Get type of first value
return f'Dict[{key_type}, {value_type}]'
else:
return type(first_element).__name__
def _make_name(code: str) -> str:
"""Makes a name for a loop argument from a unique code."""
return f'{LOOP_ITEM_PARAM_NAME_BASE}-{code}'
class LoopParameterArgument(pipeline_channel.PipelineParameterChannel):
"""Represents the parameter arguments that are looped over in a ParallelFor
loop.
The class shouldn't be instantiated by the end user, rather it is
created automatically by a ParallelFor ops group.
To create a LoopParameterArgument instance, use one of its factory methods::
LoopParameterArgument.from_pipeline_channel(...)
LoopParameterArgument.from_raw_items(...)
Attributes:
items_or_pipeline_channel: The raw items or the PipelineParameterChannel object
this LoopParameterArgument is associated to.
"""
def __init__(
self,
items: Union[ItemList, pipeline_channel.PipelineParameterChannel],
name_code: Optional[str] = None,
name_override: Optional[str] = None,
**kwargs,
):
"""Initializes a LoopParameterArgument object.
Args:
items: List of items to loop over. If a list of dicts then, all
dicts must have the same keys and every key must be a legal
Python variable name.
name_code: A unique code used to identify these loop arguments.
Should match the code for the ParallelFor ops_group which created
these LoopArguments. This prevents parameter name collisions.
name_override: The override name for PipelineParameterChannel.
**kwargs: Any other keyword arguments passed down to PipelineParameterChannel.
"""
if (name_code is None) == (name_override is None):
raise ValueError(
'Expect one and only one of `name_code` and `name_override` to '
'be specified.')
if name_override is None:
super().__init__(name=_make_name(name_code), **kwargs)
else:
super().__init__(name=name_override, **kwargs)
if not isinstance(
items,
(list, tuple, pipeline_channel.PipelineParameterChannel)):
raise TypeError(
f'Expected list, tuple, or PipelineParameterChannel, got {items}.'
)
if isinstance(items, tuple):
items = list(items)
self.items_or_pipeline_channel = items
self.is_with_items_loop_argument = not isinstance(
items, pipeline_channel.PipelineParameterChannel)
self._referenced_subvars: Dict[str, LoopArgumentVariable] = {}
if isinstance(items, list) and isinstance(items[0], dict):
subvar_names = set(items[0].keys())
# then this block creates loop_arg.variable_a and loop_arg.variable_b
for subvar_name in subvar_names:
loop_arg_var = LoopArgumentVariable(
loop_argument=self,
subvar_name=subvar_name,
)
self._referenced_subvars[subvar_name] = loop_arg_var
setattr(self, subvar_name, loop_arg_var)
def __getattr__(self, name: str):
# this is being overridden so that we can access subvariables of the
# LoopArgument (i.e.: item.a) without knowing the subvariable names ahead
# of time.
return self._referenced_subvars.setdefault(
name, LoopArgumentVariable(
loop_argument=self,
subvar_name=name,
))
@classmethod
def from_pipeline_channel(
cls,
channel: pipeline_channel.PipelineParameterChannel,
) -> 'LoopParameterArgument':
"""Creates a LoopParameterArgument object from a
PipelineParameterChannel object.
Provide a flexible default channel_type ('String') if extraction
from PipelineParameterChannel is unsuccessful. This maintains
compilation progress in cases of unknown or missing type
information.
"""
return LoopParameterArgument(
items=channel,
name_override=channel.name + '-' + LOOP_ITEM_NAME_BASE,
task_name=channel.task_name,
channel_type=_get_loop_item_type(channel.channel_type) or 'String',
)
@classmethod
def from_raw_items(
cls,
raw_items: ItemList,
name_code: str,
) -> 'LoopParameterArgument':
"""Creates a LoopParameterArgument object from raw item list."""
if len(raw_items) == 0:
raise ValueError('Got an empty item list for loop argument.')
return LoopParameterArgument(
items=raw_items,
name_code=name_code,
channel_type=_get_first_element_type(raw_items),
)
class LoopArtifactArgument(pipeline_channel.PipelineArtifactChannel):
"""Represents the artifact arguments that are looped over in a ParallelFor
loop.
The class shouldn't be instantiated by the end user, rather it is
created automatically by a ParallelFor ops group.
To create a LoopArtifactArgument instance, use the factory method::
LoopArtifactArgument.from_pipeline_channel(...)
Attributes:
pipeline_channel: The PipelineArtifactChannel object this
LoopArtifactArgument is associated to.
"""
def __init__(
self,
items: pipeline_channel.PipelineArtifactChannel,
name_code: Optional[str] = None,
name_override: Optional[str] = None,
**kwargs,
):
"""Initializes a LoopArtifactArgument object.
Args:
items: The PipelineArtifactChannel object this LoopArtifactArgument is
associated to.
name_code: A unique code used to identify these loop arguments.
Should match the code for the ParallelFor ops_group which created
these LoopArtifactArguments. This prevents parameter name collisions.
name_override: The override name for PipelineArtifactChannel.
**kwargs: Any other keyword arguments passed down to PipelineArtifactChannel.
"""
if (name_code is None) == (name_override is None):
raise ValueError(
'Expect one and only one of `name_code` and `name_override` to '
'be specified.')
# We don't support nested lists so `is_artifact_list` is always False.
if name_override is None:
super().__init__(
name=_make_name(name_code), is_artifact_list=False, **kwargs)
else:
super().__init__(
name=name_override, is_artifact_list=False, **kwargs)
self.items_or_pipeline_channel = items
self.is_with_items_loop_argument = not isinstance(
items, pipeline_channel.PipelineArtifactChannel)
@classmethod
def from_pipeline_channel(
cls,
channel: pipeline_channel.PipelineArtifactChannel,
) -> 'LoopArtifactArgument':
"""Creates a LoopArtifactArgument object from a PipelineArtifactChannel
object."""
if not channel.is_artifact_list:
raise ValueError(
'Cannot iterate over a single Artifact using `dsl.ParallelFor`. Expected a list of Artifacts as argument to `items`.'
)
return LoopArtifactArgument(
items=channel,
name_override=channel.name + '-' + LOOP_ITEM_NAME_BASE,
task_name=channel.task_name,
channel_type=channel.channel_type,
)
# TODO: support artifact constants here.
class LoopArgumentVariable(pipeline_channel.PipelineParameterChannel):
"""Represents a subvariable for a loop argument.
This is used for cases where we're looping over maps, each of which contains
several variables. If the user ran:
with dsl.ParallelFor([{'a': 1, 'b': 2}, {'a': 3, 'b': 4}]) as item:
...
Then there's one LoopArgumentVariable for 'a' and another for 'b'.
Attributes:
loop_argument: The original LoopArgument object this subvariable is
attached to.
subvar_name: The subvariable name.
"""
SUBVAR_NAME_DELIMITER = '-subvar-'
LEGAL_SUBVAR_NAME_REGEX = re.compile(r'^[a-zA-Z_][0-9a-zA-Z_]*$')
def __init__(
self,
loop_argument: LoopParameterArgument,
subvar_name: str,
):
"""Initializes a LoopArgumentVariable instance.
Args:
loop_argument: The LoopParameterArgument object this subvariable is based on
a subvariable to.
subvar_name: The name of this subvariable, which is the name of the
dict key that spawned this subvariable.
Raises:
ValueError is subvar name is illegal.
"""
if not self._subvar_name_is_legal(subvar_name):
raise ValueError(
f'Tried to create subvariable named {subvar_name}, but that is '
'not a legal Python variable name.')
self.subvar_name = subvar_name
self.loop_argument = loop_argument
# Handle potential channel_type extraction errors from LoopArgument by defaulting to 'String'. This maintains compilation progress.
super().__init__(
name=self._get_name_override(
loop_arg_name=loop_argument.name,
subvar_name=subvar_name,
),
task_name=loop_argument.task_name,
channel_type=_get_subvar_type(loop_argument.channel_type) or
'String',
)
@property
def items_or_pipeline_channel(
self) -> Union[ItemList, pipeline_channel.PipelineParameterChannel]:
"""Returns the loop argument items."""
return self.loop_argument.items_or_pipeline_channel
@property
def is_with_items_loop_argument(self) -> bool:
"""Whether the loop argument is originated from raw items."""
return self.loop_argument.is_with_items_loop_argument
def _subvar_name_is_legal(self, proposed_variable_name: str) -> bool:
"""Returns True if the subvar name is legal."""
return re.match(self.LEGAL_SUBVAR_NAME_REGEX,
proposed_variable_name) is not None
def _get_name_override(self, loop_arg_name: str, subvar_name: str) -> str:
"""Gets the name.
Args:
loop_arg_name: the name of the loop argument parameter that this
LoopArgumentVariable is attached to.
subvar_name: The name of this subvariable.
Returns:
The name of this loop arg variable.
"""
return f'{loop_arg_name}{self.SUBVAR_NAME_DELIMITER}{subvar_name}'
# TODO: migrate Collected to OneOfMixin style implementation
[docs]class Collected(pipeline_channel.PipelineChannel):
"""For collecting into a list the output from a task in dsl.ParallelFor
loops.
Args:
output: The output of an upstream task within a dsl.ParallelFor loop.
Example:
::
@dsl.pipeline
def math_pipeline() -> int:
with dsl.ParallelFor([1, 2, 3]) as x:
t = double(num=x)
return add(nums=dsl.Collected(t.output)).output
"""
def __init__(
self,
output: pipeline_channel.PipelineChannel,
) -> None:
self.output = output
# we know all dsl.Collected instances are lists, so set `is_artifact_list`
# for type checking, which occurs before dsl.Collected is updated to
# it's "correct" channel during compilation
if isinstance(output, pipeline_channel.PipelineArtifactChannel):
channel_type = output.channel_type
self.is_artifact_channel = True
self.is_artifact_list = True
else:
channel_type = 'LIST'
self.is_artifact_channel = False
self.is_artifact_list = False
super().__init__(
output.name,
channel_type=channel_type,
task_name=output.task_name,
)
self._validate_no_oneof_channel(self.output)
def _validate_no_oneof_channel(
self, channel: Union[pipeline_channel.PipelineParameterChannel,
pipeline_channel.PipelineArtifactChannel]
) -> None:
if isinstance(channel, pipeline_channel.OneOfMixin):
raise ValueError(
f'dsl.{pipeline_channel.OneOf.__name__} cannot be used inside of dsl.{Collected.__name__}.'
)