pipelines/sdk/python/kfp/dsl/_pipeline.py

345 lines
11 KiB
Python

# Copyright 2018-2019 Google LLC
#
# 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 typing import Callable, Optional, Union
from kubernetes.client.models import V1PodDNSConfig
from . import _container_op
from . import _resource_op
from . import _ops_group
from ._component_bridge import \
_create_container_op_from_component_and_arguments, \
_sanitize_python_function_name
from ..components import _components
from ..components._naming import _make_name_unique_by_adding_index
import sys
# This handler is called whenever the @pipeline decorator is applied.
# It can be used by command-line DSL compiler to inject code that runs for every pipeline definition.
_pipeline_decorator_handler = None
def pipeline(
name: Optional[str] = None,
description: Optional[str] = None,
pipeline_root: Optional[str] = None):
"""Decorator of pipeline functions.
Example
::
@pipeline(
name='my awesome pipeline',
description='Is it really awesome?'
pipeline_root='gs://my-bucket/my-output-path'
)
def my_pipeline(a: PipelineParam, b: PipelineParam):
...
Args:
name: The pipeline name. Default to a sanitized version of the function
name.
description: Optionally, a human-readable description of the pipeline.
pipeline_root: The root directory to generate input/output URI under this
pipeline. This is required if input/output URI placeholder is used in this
pipeline.
"""
def _pipeline(func: Callable):
if name:
func._component_human_name = name
if description:
func._component_description = description
if pipeline_root:
func.output_directory = pipeline_root
if _pipeline_decorator_handler:
return _pipeline_decorator_handler(func) or func
else:
return func
return _pipeline
class PipelineConf():
"""PipelineConf contains pipeline level settings."""
def __init__(self):
self.image_pull_secrets = []
self.timeout = 0
self.ttl_seconds_after_finished = -1
self._pod_disruption_budget_min_available = None
self.op_transformers = []
self.default_pod_node_selector = {}
self.image_pull_policy = None
self.parallelism = None
self._data_passing_method = None
self.dns_config = None
def set_image_pull_secrets(self, image_pull_secrets):
"""Configures the pipeline level imagepullsecret
Args:
image_pull_secrets: a list of Kubernetes V1LocalObjectReference
For detailed description, check Kubernetes V1LocalObjectReference definition
https://github.com/kubernetes-client/python/blob/master/kubernetes/docs/V1LocalObjectReference.md
"""
self.image_pull_secrets = image_pull_secrets
return self
def set_timeout(self, seconds: int):
"""Configures the pipeline level timeout
Args:
seconds: number of seconds for timeout
"""
self.timeout = seconds
return self
def set_parallelism(self, max_num_pods: int):
"""Configures the max number of total parallel pods that can execute at the same time in a workflow.
Args:
max_num_pods: max number of total parallel pods.
"""
if max_num_pods < 1:
raise ValueError('Pipeline max_num_pods set to < 1, allowed values are > 0')
self.parallelism = max_num_pods
return self
def set_ttl_seconds_after_finished(self, seconds: int):
"""Configures the ttl after the pipeline has finished.
Args:
seconds: number of seconds for the workflow to be garbage collected after it is finished.
"""
self.ttl_seconds_after_finished = seconds
return self
def set_pod_disruption_budget(self, min_available: Union[int, str]):
""" PodDisruptionBudget holds the number of concurrent disruptions that you allow for pipeline Pods.
Args:
min_available (Union[int, str]): An eviction is allowed if at least "minAvailable" pods selected by
"selector" will still be available after the eviction, i.e. even in the
absence of the evicted pod. So for example you can prevent all voluntary
evictions by specifying "100%". "minAvailable" can be either an absolute number or a percentage.
"""
self._pod_disruption_budget_min_available = min_available
return self
def set_default_pod_node_selector(self, label_name: str, value: str):
"""Add a constraint for nodeSelector for a pipeline. Each constraint is a key-value pair label. For the
container to be eligible to run on a node, the node must have each of the constraints appeared
as labels.
Args:
label_name: The name of the constraint label.
value: The value of the constraint label.
"""
self.default_pod_node_selector[label_name] = value
return self
def set_image_pull_policy(self, policy: str):
"""Configures the default image pull policy
Args:
policy: the pull policy, has to be one of: Always, Never, IfNotPresent.
For more info: https://github.com/kubernetes-client/python/blob/10a7f95435c0b94a6d949ba98375f8cc85a70e5a/kubernetes/docs/V1Container.md
"""
self.image_pull_policy = policy
return self
def add_op_transformer(self, transformer):
"""Configures the op_transformers which will be applied to all ops in the pipeline.
The ops can be ResourceOp, VolumeOp, or ContainerOp.
Args:
transformer: A function that takes a kfp Op as input and returns a kfp Op
"""
self.op_transformers.append(transformer)
def set_dns_config(self, dns_config: V1PodDNSConfig):
"""Set the dnsConfig to be given to each pod.
Args:
dns_config: Kubernetes V1PodDNSConfig
For detailed description, check Kubernetes V1PodDNSConfig definition
https://github.com/kubernetes-client/python/blob/master/kubernetes/docs/V1PodDNSConfig.md
Example:
::
import kfp
from kubernetes.client.models import V1PodDNSConfig, V1PodDNSConfigOption
pipeline_conf = kfp.dsl.PipelineConf()
pipeline_conf.set_dns_config(dns_config=V1PodDNSConfig(
nameservers=["1.2.3.4"],
options=[V1PodDNSConfigOption(name="ndots", value="2")]
))
"""
self.dns_config = dns_config
@property
def data_passing_method(self):
return self._data_passing_method
@data_passing_method.setter
def data_passing_method(self, value):
"""Sets the object representing the method used for intermediate data passing.
Example:
::
from kfp.dsl import PipelineConf, data_passing_methods
from kubernetes.client.models import V1Volume, V1PersistentVolumeClaim
pipeline_conf = PipelineConf()
pipeline_conf.data_passing_method = data_passing_methods.KubernetesVolume(
volume=V1Volume(
name='data',
persistent_volume_claim=V1PersistentVolumeClaim('data-volume'),
),
path_prefix='artifact_data/',
)
"""
self._data_passing_method = value
def get_pipeline_conf():
"""Configure the pipeline level setting to the current pipeline
Note: call the function inside the user defined pipeline function.
"""
return Pipeline.get_default_pipeline().conf
# TODO: Pipeline is in fact an opsgroup, refactor the code.
class Pipeline():
"""A pipeline contains a list of operators.
This class is not supposed to be used by pipeline authors since pipeline authors can use
pipeline functions (decorated with @pipeline) to reference their pipelines. This class
is useful for implementing a compiler. For example, the compiler can use the following
to get the pipeline object and its ops:
Example:
::
with Pipeline() as p:
pipeline_func(*args_list)
traverse(p.ops)
"""
# _default_pipeline is set when it (usually a compiler) runs "with Pipeline()"
_default_pipeline = None
@staticmethod
def get_default_pipeline():
"""Get default pipeline. """
return Pipeline._default_pipeline
@staticmethod
def add_pipeline(name, description, func):
"""Add a pipeline function with the specified name and description."""
# Applying the @pipeline decorator to the pipeline function
func = pipeline(name=name, description=description)(func)
def __init__(self, name: str):
"""Create a new instance of Pipeline.
Args:
name: the name of the pipeline. Once deployed, the name will show up in Pipeline System UI.
"""
self.name = name
self.ops = {}
# Add the root group.
self.groups = [_ops_group.OpsGroup('pipeline', name=name)]
self.group_id = 0
self.conf = PipelineConf()
self._metadata = None
def __enter__(self):
if Pipeline._default_pipeline:
raise Exception('Nested pipelines are not allowed.')
Pipeline._default_pipeline = self
self._old_container_task_constructor = _components._container_task_constructor
_components._container_task_constructor = _create_container_op_from_component_and_arguments
def register_op_and_generate_id(op):
return self.add_op(op, op.is_exit_handler)
self._old__register_op_handler = _container_op._register_op_handler
_container_op._register_op_handler = register_op_and_generate_id
return self
def __exit__(self, *args):
Pipeline._default_pipeline = None
_container_op._register_op_handler = self._old__register_op_handler
_components._container_task_constructor = self._old_container_task_constructor
def add_op(self, op: _container_op.BaseOp, define_only: bool):
"""Add a new operator.
Args:
op: An operator of ContainerOp, ResourceOp or their inherited types.
Returns
op_name: a unique op name.
"""
# Sanitizing the op name.
# Technically this could be delayed to the compilation stage, but string serialization of PipelineParams make unsanitized names problematic.
op_name = _sanitize_python_function_name(op.human_name).replace('_', '-')
#If there is an existing op with this name then generate a new name.
op_name = _make_name_unique_by_adding_index(op_name, list(self.ops.keys()), ' ')
if op_name == '':
op_name = _make_name_unique_by_adding_index('task', list(self.ops.keys()), ' ')
self.ops[op_name] = op
if not define_only:
self.groups[-1].ops.append(op)
return op_name
def push_ops_group(self, group: _ops_group.OpsGroup):
"""Push an OpsGroup into the stack.
Args:
group: An OpsGroup. Typically it is one of ExitHandler, Branch, and Loop.
"""
self.groups[-1].groups.append(group)
self.groups.append(group)
def pop_ops_group(self):
"""Remove the current OpsGroup from the stack."""
del self.groups[-1]
def remove_op_from_groups(self, op):
for group in self.groups:
group.remove_op_recursive(op)
def get_next_group_id(self):
"""Get next id for a new group. """
self.group_id += 1
return self.group_id
def _set_metadata(self, metadata):
"""_set_metadata passes the containerop the metadata information
Args:
metadata (ComponentMeta): component metadata
"""
self._metadata = metadata