def test_use_azure_secret(self):
op1 = ContainerOp(name='op1', image='image')
op1 = op1.apply(use_azure_secret('foo'))
assert len(op1.container.env) == 4
index = 0
for expected in ['AZ_SUBSCRIPTION_ID', 'AZ_TENANT_ID', 'AZ_CLIENT_ID', 'AZ_CLIENT_SECRET']:
assert op1.container.env[index].name == expected
assert op1.container.env[index].value_from.secret_key_ref.name == 'foo'
assert op1.container.env[index].value_from.secret_key_ref.key == expected
index += 1
def test_use_set_volume_use_secret(self):
op1 = ContainerOp(name="op1", image="image")
secret_name = "my-secret"
secret_path = "/here/are/my/secret"
op1 = op1.apply(
use_secret(secret_name=secret_name,
secret_volume_mount_path=secret_path))
self.assertEqual(type(op1.container.env), type(None))
container_dict = op1.container.to_dict()
volume_mounts = container_dict["volume_mounts"][0]
self.assertEqual(type(volume_mounts), dict)
self.assertEqual(volume_mounts["mount_path"], secret_path)
def test_use_aws_secret(self):
op1 = ContainerOp(name='op1', image='image')
op1 = op1.apply(use_aws_secret('myaws-secret', 'key_id', 'access_key'))
assert len(op1.container.env) == 2
index = 0
for expected_name, expected_key in [('AWS_ACCESS_KEY_ID', 'key_id'),
('AWS_SECRET_ACCESS_KEY',
'access_key')]:
assert op1.container.env[index].name == expected_name
assert op1.container.env[
index].value_from.secret_key_ref.name == 'myaws-secret'
assert op1.container.env[
index].value_from.secret_key_ref.key == expected_key
index += 1
def test_use_aws_secret(self):
with Pipeline('somename') as p:
op1 = ContainerOp(name='op1', image='image')
op1 = op1.apply(
use_aws_secret('myaws-secret', 'key_id', 'access_key'))
assert len(op1.env_variables) == 2
index = 0
for expected in ['key_id', 'access_key']:
assert op1.env_variables[index].name == expected
assert op1.env_variables[
index].value_from.secret_key_ref.name == 'myaws-secret'
assert op1.env_variables[
index].value_from.secret_key_ref.key == expected
index += 1
def test_use_azure_secret(self):
with Pipeline('somename') as p:
op1 = ContainerOp(name='op1', image='image')
op1 = op1.apply(use_azure_secret('azcreds'))
assert len(op1.env_variables) == 4
index = 0
for expected in [
'AZ_SUBSCRIPTION_ID', 'AZ_TENANT_ID', 'AZ_CLIENT_ID',
'AZ_CLIENT_SECRET'
]:
print(op1.env_variables[index].name)
print(op1.env_variables[index].value_from.secret_key_ref.name)
print(op1.env_variables[index].value_from.secret_key_ref.key)
index += 1
def test_use_set_env_use_secret(self):
op1 = ContainerOp(name="op1", image="image")
secret_name = "my-secret"
secret_path = "/here/are/my/secret/"
env_variable = "MY_SECRET"
secret_file_path_in_volume = "secret.json"
op1 = op1.apply(
use_secret(secret_name=secret_name,
secret_volume_mount_path=secret_path,
env_variable=env_variable,
secret_file_path_in_volume=secret_file_path_in_volume))
self.assertEqual(len(op1.container.env), 1)
container_dict = op1.container.to_dict()
volume_mounts = container_dict["volume_mounts"][0]
self.assertEqual(type(volume_mounts), dict)
self.assertEqual(volume_mounts["mount_path"], secret_path)
env_dict = op1.container.env[0].to_dict()
self.assertEqual(env_dict["name"], env_variable)
self.assertEqual(env_dict["value"],
secret_path + secret_file_path_in_volume)
def update_op(op: dsl.ContainerOp,
pipeline_name: dsl.PipelineParam,
pipeline_root: dsl.PipelineParam,
launcher_image: Optional[str] = None) -> None:
"""Updates the passed in Op for running in v2-compatible mode.
Args:
op: The Op to update.
pipeline_spec: The PipelineSpec for the pipeline under which `op`
runs.
pipeline_root: The root output directory for pipeline artifacts.
launcher_image: An optional launcher image. Useful for tests.
"""
op.is_v2 = True
# Inject the launcher binary and overwrite the entrypoint.
image_name = launcher_image or _DEFAULT_LAUNCHER_IMAGE
launcher_container = dsl.UserContainer(
name="kfp-launcher",
image=image_name,
command=["launcher", "--copy", "/kfp-launcher/launch"],
mirror_volume_mounts=True)
op.add_init_container(launcher_container)
op.add_volume(k8s_client.V1Volume(name='kfp-launcher'))
op.add_volume_mount(
k8s_client.V1VolumeMount(name='kfp-launcher',
mount_path='/kfp-launcher'))
# op.command + op.args will have the following sections:
# 1. args passed to kfp-launcher
# 2. a separator "--"
# 3. parameters in format "key1=value1", "key2=value2", ...
# 4. a separator "--" as end of arguments passed to launcher
# 5. (start of op.args) arguments of the original user program command + args
#
# example:
# - command:
# - /kfp-launcher/launch
# - '--mlmd_server_address'
# - $(METADATA_GRPC_SERVICE_HOST)
# - '--mlmd_server_port'
# - $(METADATA_GRPC_SERVICE_PORT)
# - ... # more launcher params
# - '--pipeline_task_id'
# - $(KFP_POD_NAME)
# - '--pipeline_root'
# - ''
# - '--' # start of parameter values
# - first=first
# - second=second
# - '--' # start of user command and args
# args:
# - sh
# - '-ec'
# - |
# program_path=$(mktemp)
# printf "%s" "$0" > "$program_path"
# python3 -u "$program_path" "$@"
# - >
# import json
# import xxx
# ...
op.command = [
"/kfp-launcher/launch",
"--mlmd_server_address",
"$(METADATA_GRPC_SERVICE_HOST)",
"--mlmd_server_port",
"$(METADATA_GRPC_SERVICE_PORT)",
"--runtime_info_json",
"$(KFP_V2_RUNTIME_INFO)",
"--container_image",
"$(KFP_V2_IMAGE)",
"--task_name",
op.name,
"--pipeline_name",
pipeline_name,
"--run_id",
"$(KFP_RUN_ID)",
"--run_resource",
"workflows.argoproj.io/$(WORKFLOW_ID)",
"--namespace",
"$(KFP_NAMESPACE)",
"--pod_name",
"$(KFP_POD_NAME)",
"--pod_uid",
"$(KFP_POD_UID)",
"--pipeline_root",
pipeline_root,
"--enable_caching",
"$(ENABLE_CACHING)",
]
# Mount necessary environment variables.
op.apply(_default_transformers.add_kfp_pod_env)
op.container.add_env_variable(
k8s_client.V1EnvVar(name="KFP_V2_IMAGE", value=op.container.image))
config_map_ref = k8s_client.V1ConfigMapEnvSource(
name='metadata-grpc-configmap', optional=True)
op.container.add_env_from(
k8s_client.V1EnvFromSource(config_map_ref=config_map_ref))
op.arguments = list(op.container_spec.command) + list(
op.container_spec.args)
runtime_info = {
"inputParameters": collections.OrderedDict(),
"inputArtifacts": collections.OrderedDict(),
"outputParameters": collections.OrderedDict(),
"outputArtifacts": collections.OrderedDict(),
}
op.command += ["--"]
component_spec = op.component_spec
for parameter, spec in sorted(
component_spec.input_definitions.parameters.items()):
parameter_info = {
"type":
pipeline_spec_pb2.PrimitiveType.PrimitiveTypeEnum.Name(spec.type),
}
op.command += [f"{parameter}={op._parameter_arguments[parameter]}"]
runtime_info["inputParameters"][parameter] = parameter_info
op.command += ["--"]
for artifact_name, spec in sorted(
component_spec.input_definitions.artifacts.items()):
artifact_info = {
"metadataPath": op.input_artifact_paths[artifact_name],
"schemaTitle": spec.artifact_type.schema_title,
"instanceSchema": spec.artifact_type.instance_schema,
}
runtime_info["inputArtifacts"][artifact_name] = artifact_info
for parameter, spec in sorted(
component_spec.output_definitions.parameters.items()):
parameter_info = {
"type":
pipeline_spec_pb2.PrimitiveType.PrimitiveTypeEnum.Name(spec.type),
"path":
op.file_outputs[parameter],
}
runtime_info["outputParameters"][parameter] = parameter_info
for artifact_name, spec in sorted(
component_spec.output_definitions.artifacts.items()):
# TODO: Assert instance_schema.
artifact_info = {
# Type used to register output artifacts.
"schemaTitle": spec.artifact_type.schema_title,
"instanceSchema": spec.artifact_type.instance_schema,
# File used to write out the registered artifact ID.
"metadataPath": op.file_outputs[artifact_name],
}
runtime_info["outputArtifacts"][artifact_name] = artifact_info
op.container.add_env_variable(
k8s_client.V1EnvVar(name="KFP_V2_RUNTIME_INFO",
value=json.dumps(runtime_info)))
op.pod_annotations['pipelines.kubeflow.org/v2_component'] = "true"
op.pod_labels['pipelines.kubeflow.org/v2_component'] = "true"
def update_op(op: dsl.ContainerOp,
pipeline_name: dsl.PipelineParam,
pipeline_root: dsl.PipelineParam,
launcher_image: Optional[str] = None) -> None:
"""Updates the passed in Op for running in v2-compatible mode.
Args:
op: The Op to update.
pipeline_spec: The PipelineSpec for the pipeline under which `op`
runs.
pipeline_root: The root output directory for pipeline artifacts.
launcher_image: An optional launcher image. Useful for tests.
"""
# Inject the launcher binary and overwrite the entrypoint.
image_name = launcher_image or _DEFAULT_LAUNCHER_IMAGE
launcher_container = dsl.UserContainer(name="kfp-launcher",
image=image_name,
command="/bin/mount_launcher.sh",
mirror_volume_mounts=True)
op.add_init_container(launcher_container)
op.add_volume(k8s_client.V1Volume(name='kfp-launcher'))
op.add_volume_mount(
k8s_client.V1VolumeMount(name='kfp-launcher', mount_path='/kfp-launcher'))
op.command = [
"/kfp-launcher/launch",
"--mlmd_server_address",
"$(METADATA_GRPC_SERVICE_HOST)",
"--mlmd_server_port",
"$(METADATA_GRPC_SERVICE_PORT)",
"--runtime_info_json",
"$(KFP_V2_RUNTIME_INFO)",
"--container_image",
"$(KFP_V2_IMAGE)",
"--task_name",
op.name,
"--pipeline_name",
pipeline_name,
"--pipeline_run_id",
"$(WORKFLOW_ID)",
"--pipeline_task_id",
"$(KFP_POD_NAME)",
"--pipeline_root",
pipeline_root,
]
# Mount necessary environment variables.
op.apply(_default_transformers.add_kfp_pod_env)
op.container.add_env_variable(
k8s_client.V1EnvVar(name="KFP_V2_IMAGE", value=op.container.image))
config_map_ref = k8s_client.V1ConfigMapEnvSource(
name='metadata-grpc-configmap', optional=True)
op.container.add_env_from(
k8s_client.V1EnvFromSource(config_map_ref=config_map_ref))
op.arguments = list(op.container_spec.command) + list(op.container_spec.args)
runtime_info = {
"inputParameters": collections.OrderedDict(),
"inputArtifacts": collections.OrderedDict(),
"outputParameters": collections.OrderedDict(),
"outputArtifacts": collections.OrderedDict(),
}
component_spec = op.component_spec
for parameter, spec in sorted(
component_spec.input_definitions.parameters.items()):
parameter_info = {
"parameterType":
pipeline_spec_pb2.PrimitiveType.PrimitiveTypeEnum.Name(spec.type),
"parameterValue":
op._parameter_arguments[parameter],
}
runtime_info["inputParameters"][parameter] = parameter_info
for artifact_name, spec in sorted(
component_spec.input_definitions.artifacts.items()):
artifact_info = {"fileInputPath": op.input_artifact_paths[artifact_name]}
runtime_info["inputArtifacts"][artifact_name] = artifact_info
for parameter, spec in sorted(
component_spec.output_definitions.parameters.items()):
parameter_info = {
"parameterType":
pipeline_spec_pb2.PrimitiveType.PrimitiveTypeEnum.Name(spec.type),
"fileOutputPath":
op.file_outputs[parameter],
}
runtime_info["outputParameters"][parameter] = parameter_info
for artifact_name, spec in sorted(
component_spec.output_definitions.artifacts.items()):
# TODO: Assert instance_schema.
artifact_info = {
# Type used to register output artifacts.
"artifactSchema": spec.artifact_type.instance_schema,
# File used to write out the registered artifact ID.
"fileOutputPath": op.file_outputs[artifact_name],
}
runtime_info["outputArtifacts"][artifact_name] = artifact_info
op.container.add_env_variable(
k8s_client.V1EnvVar(name="KFP_V2_RUNTIME_INFO",
value=json.dumps(runtime_info)))
op.pod_annotations['pipelines.kubeflow.org/v2_component'] = "true"