# Copyright 2021 Zilliz. All rights reserved.
#
# 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.
from towhee.dag.graph_repr import GraphRepr
from towhee.dataframe import DataFrame
from towhee.errors import NoSchedulerError, EmptyInputError
from towhee.engine.graph_context import GraphContext
from towhee.engine.task_scheduler import TaskScheduler
[docs]class Pipeline:
"""
The runtime pipeline context, include graph context, all dataframes.
Args:
graph_repr: (`str` or `towhee.dag.GraphRepr`)
The graph representation either as a YAML-formatted string, or directly
as an instance of `GraphRepr`.
parallelism: (`int`)
The parallelism parameter dictates how many copies of the graph context
we create. This is likely a low number (1-4) for local engines, but may
be much higher for cloud instances.
"""
[docs] def __init__(self, graph_repr: GraphRepr, parallelism: int = 1) -> None:
self._parallelism = parallelism
self.on_graph_finish_handlers = []
self._scheduler = None
self._graph_count = 0
if isinstance(graph_repr, str):
self._graph_repr = GraphRepr.from_yaml(graph_repr)
else:
self._graph_repr = graph_repr
@property
def pipeline_type(self) -> int:
return self._graph_repr.graph_type
@property
def parallelism(self) -> int:
return self._parallelism
@property
def graph_repr(self) -> GraphRepr:
return self._graph_repr
@parallelism.setter
def parallelism(self, val):
if isinstance(val, int) and val > 0:
self._parallelism = val
else:
raise ValueError('Parallelism value must be a positive integer')
def register(self, scheduler: TaskScheduler):
self._scheduler = scheduler
[docs] def __repr__(self) -> str:
if self._graph_repr._ir is not None:
return self._graph_repr._ir
else:
return object.__repr__(self)
[docs] def __call__(self, inputs: DataFrame) -> DataFrame:
"""
Process an input `DataFrame`. This function instantiates an output `DataFrame`;
upon completion, individual `GraphContext` outputs are merged into this
dataframe. Inputs are weaved through the input `DataFrame` for each
`GraphContext` as follows (`parallelism = 3`):
data[0] -> ctx[0]
data[1] -> ctx[1]
data[2] -> ctx[2]
data[3] -> ctx[0]
data[4] -> ctx[1]
Args:
inputs (`towhee.dataframe.DataFrame`):
Input `DataFrame` (with potentially multiple rows) to process.
Returns:
(`towhee.dataframe.DataFrame`)
Output `DataFrame` with ordering matching the input `DataFrame`.
"""
if not self._scheduler:
raise NoSchedulerError('Pipeline not registered to a Scheduler.')
assert self._parallelism == 1
graph_ctx = GraphContext(self._graph_count, self._graph_repr)
self._graph_count += 1
self._scheduler.register(graph_ctx)
input_data = inputs.get(0, 1)
if input_data is None:
raise EmptyInputError('Input data is empty')
graph_ctx(input_data[0])
graph_ctx.join()
return graph_ctx.result()