def a_op(field_m: {
'GCSPath': {
'path_type': 'file',
'file_type': 'tsv'
}
}, field_o: 'Integer'):
return ContainerOp(
name='operator a',
image='gcr.io/ml-pipeline/component-b',
arguments=[
'--field-l',
field_m,
'--field-o',
field_o,
],
)
def test_basic(self):
"""Test basic usage."""
param1 = PipelineParam('param1')
param2 = PipelineParam('param2')
op1 = (ContainerOp(
name='op1',
image='image',
arguments=['%s hello %s %s' % (param1, param2, param1)],
init_containers=[
UserContainer(name='initcontainer0', image='initimage0')
],
sidecars=[Sidecar(name='sidecar0', image='image0')],
container_kwargs={
'env': [V1EnvVar(name='env1', value='value1')]
},
file_outputs={
'out1': '/tmp/b'
}).add_init_container(
UserContainer(name='initcontainer1',
image='initimage1')).add_init_container(
UserContainer(
name='initcontainer2',
image='initimage2')).add_sidecar(
Sidecar(name='sidecar1',
image='image1')).add_sidecar(
Sidecar(name='sidecar2',
image='image2')))
self.assertCountEqual([x.name for x in op1.inputs],
['param1', 'param2'])
self.assertCountEqual(list(op1.outputs.keys()), ['out1'])
self.assertCountEqual([x.op_name for x in op1.outputs.values()],
[op1.name])
self.assertEqual(op1.output.name, 'out1')
self.assertCountEqual(
[init_container.name for init_container in op1.init_containers],
['initcontainer0', 'initcontainer1', 'initcontainer2'])
self.assertCountEqual(
[init_container.image for init_container in op1.init_containers],
['initimage0', 'initimage1', 'initimage2'])
self.assertCountEqual([sidecar.name for sidecar in op1.sidecars],
['sidecar0', 'sidecar1', 'sidecar2'])
self.assertCountEqual([sidecar.image for sidecar in op1.sidecars],
['image0', 'image1', 'image2'])
self.assertCountEqual([env.name for env in op1.container.env],
['env1'])
def test_basic(self):
"""Test basic usage."""
with Pipeline('somename') as p:
param1 = PipelineParam('param1')
param2 = PipelineParam('param2')
op1 = ContainerOp(
name='op1',
image='image',
arguments=['%s hello %s %s' % (param1, param2, param1)],
file_outputs={'out1': '/tmp/b'})
self.assertCountEqual([x.name for x in op1.inputs],
['param1', 'param2'])
self.assertCountEqual(list(op1.outputs.keys()), ['out1'])
self.assertCountEqual([x.op_name for x in op1.outputs.values()],
['op1'])
self.assertEqual(op1.output.name, 'out1')
def b_op(field_x: {'customized_type': {'openapi_schema_validator': '{"type": "string", "pattern": "^gcs://.*$"}'}},
field_y: Integer(),
field_z: GCSPath()) -> {'output_model_uri': 'GcsUri'}:
return ContainerOp(
name = 'operator b',
image = 'gcr.io/ml-pipeline/component-a',
command = [
'python3',
field_x,
],
arguments = [
'--field-y', field_y,
'--field-z', field_z,
],
file_outputs = {
'output_model_uri': '/schema.txt',
}
)
def b_op(field_x: {'customized_type': {'property_a': 'value_a', 'property_b': 'value_b'}},
field_y: 'GcsUri',
field_z: GCSPath(path_type='file', file_type='tsv')) -> {'output_model_uri': 'GcsUri'}:
return ContainerOp(
name = 'operator b',
image = 'gcr.io/ml-pipeline/component-b',
command = [
'python3',
field_x,
],
arguments = [
'--field-y', field_y,
'--field-z', field_z,
],
file_outputs = {
'output_model_uri': '/schema.txt',
}
)
def b_op(field_x,
field_y: Integer(),
field_z: GCSPath(path_type='file', file_type='tsv')) -> {'output_model_uri': 'GcsUri'}:
return ContainerOp(
name = 'operator b',
image = 'gcr.io/ml-pipeline/component-a',
command = [
'python3',
field_x,
],
arguments = [
'--field-y', field_y,
'--field-z', field_z,
],
file_outputs = {
'output_model_uri': '/schema.txt',
}
)
def a_op(
field_m: {
'GCRPath': {
'openapi_schema_validator': {
"type": "string",
"pattern": "^.*gcr\\.io/.*$"
}
}
}, field_o: 'Integer'):
return ContainerOp(
name='operator a',
image='gcr.io/ml-pipeline/component-b',
arguments=[
'--field-l',
field_m,
'--field-o',
field_o,
],
)
def __init__(self, kind=None, command=None, args=None, image=None, handler=None,
metadata=None, build=None, volumes=None, volume_mounts=None,
env=None, resources=None, image_pull_policy=None,
service_account=None):
try:
from kfp.dsl import ContainerOp
except ImportError as e:
print('KubeFlow pipelines sdk is not installed, use "pip install kfp"')
raise e
super().__init__(kind, command, args, image, handler, metadata, None)
self._build = None
self.build = build
self.volumes = volumes or []
self.volume_mounts = volume_mounts or []
self.env = env or []
self.resources = resources
self.image_pull_policy = image_pull_policy
self.service_account = service_account
self._cop = ContainerOp('name', 'image')
def a_op(field_l: Integer()) -> {
'field_m': 'GCSPath',
'field_n': {
'customized_type': {
'property_a': 'value_a',
'property_b': 'value_b'
}
},
'field_o': 'Integer'
}:
return ContainerOp(name='operator a',
image='gcr.io/ml-pipeline/component-b',
arguments=[
'--field-l',
field_l,
],
file_outputs={
'field_m': '/schema.txt',
'field_n': '/feature.txt',
'field_o': '/output.txt'
})
def a_op(field_l: Integer()) -> {
'field_m': 'GCSPath',
'field_n': {
'customized_type': {
'openapi_schema_validator':
'{"type": "string", "pattern": "^gs://.*$"}'
}
},
'field_o': 'Integer'
}:
return ContainerOp(name='operator a',
image='gcr.io/ml-pipeline/component-b',
arguments=[
'--field-l',
field_l,
],
file_outputs={
'field_m': '/schema.txt',
'field_n': '/feature.txt',
'field_o': '/output.txt'
})
def training_op(script, image=None, arguments=[], file_outputs={}):
""" A template function to encapsulate similar container ops
"""
if not image and _is_ipython():
from IPython import get_ipython
image = get_ipython().user_ns.get('TRAINING_IMAGE')
if not image:
raise ValueError(f"""
`image` parameter is missing.
If you run in Jupyter Notebook you can also define a global var TRAINING_IMAGE
""")
return ContainerOp(
name=re.sub(r'[\W_]+', '-',
os.path.splitext(script.lower())[0]),
image=image,
command=['/usr/local/bin/python', script],
arguments=arguments,
file_outputs=file_outputs,
)
def test_deprecation_warnings(self):
"""Test deprecation warnings."""
op = ContainerOp(name='op1', image='image')
with self.assertWarns(PendingDeprecationWarning):
op.env_variables = [V1EnvVar(name="foo", value="bar")]
with self.assertWarns(PendingDeprecationWarning):
op.image = 'image2'
with self.assertWarns(PendingDeprecationWarning):
op.set_memory_request('10M')
with self.assertWarns(PendingDeprecationWarning):
op.set_memory_limit('10M')
with self.assertWarns(PendingDeprecationWarning):
op.set_cpu_request('100m')
with self.assertWarns(PendingDeprecationWarning):
op.set_cpu_limit('1')
with self.assertWarns(PendingDeprecationWarning):
op.set_gpu_limit('1')
with self.assertWarns(PendingDeprecationWarning):
op.add_env_variable(V1EnvVar(name="foo", value="bar"))
with self.assertWarns(PendingDeprecationWarning):
op.add_volume_mount(
V1VolumeMount(mount_path='/secret/gcp-credentials',
name='gcp-credentials'))
def test_after_op(self):
"""Test duplicate ops."""
op1 = ContainerOp(name='op1', image='image')
op2 = ContainerOp(name='op2', image='image')
op2.after(op1)
self.assertCountEqual(op2.dependent_names, [op1.name])
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 copy(self):
self._cop = None
fn = deepcopy(self)
self._cop = ContainerOp("name", "image")
fn._cop = ContainerOp("name", "image")
return fn
def container(
name: str,
arguments: str,
inputs: Optional[List[Tuple[InputArgumentPath, str]]] = None,
outputs: Optional[Dict[str, str]] = None,
) -> Tuple[ContainerOp, Dict[str, Tuple[InputArgumentPath, str]]]:
# Set the correct shell parameters
prepare_args = "set -euo pipefail\n"
# Copy the output artifacts correctly
file_outputs = {}
output_artifact_copy_args = ""
if outputs:
for k, v in outputs.items():
out = Pipeline.out_dir(v)
file_outputs[k] = out
output_artifact_copy_args += dedent("""
mkdir -p {d}
cp -r {fr} {to}
""".format(
d=os.path.dirname(out),
fr=v,
to=out,
)).lstrip()
# Create the container
ctr = ContainerOp(
image=Pipeline.IMAGE,
name=name,
command=["bash", "-c"],
output_artifact_paths=Pipeline.default_artifact_path(),
file_outputs=file_outputs,
artifact_argument_paths=[InputArgumentPath(x[0])
for x in inputs] if inputs else None,
)
ctr.container.set_image_pull_policy("Always")
# Copy input artifacts correctly
input_artifact_copy_args = ""
in_repo = False
for i, path in enumerate(ctr.input_artifact_paths.values()):
target_location = inputs[i][1]
input_artifact_copy_args += "cp -r {fr} {to}\n".format(
fr=path, to=target_location)
# Change to the repository path if available
if target_location == Pipeline.REPO:
in_repo = True
input_artifact_copy_args += "cd {}\n".format(Pipeline.REPO)
# Show the git diff to validate
if in_repo:
input_artifact_copy_args += dedent("""
echo "git diff:"
git diff --name-only
""")
# Assemble the command
ctr.arguments = prepare_args + \
input_artifact_copy_args + \
arguments + \
"\n" + \
output_artifact_copy_args
# Output Artifacts
vol = "output-artifacts"
ctr.add_volume(
k8s.V1Volume(name=vol, empty_dir=k8s.V1EmptyDirVolumeSource()))
ctr.container.add_volume_mount(
k8s.V1VolumeMount(name=vol, mount_path=Pipeline.OUT_DIR))
# GitHub Token
gh_token = "github-token"
ctr.add_volume(
k8s.V1Volume(
name=gh_token,
secret=k8s.V1SecretVolumeSource(secret_name=gh_token)))
ctr.container.add_volume_mount(
k8s.V1VolumeMount(name=gh_token,
read_only=True,
mount_path=Pipeline.GITHUB_TOKEN_MOUNT_PATH))
# Quay Login
quay = "quay"
ctr.add_volume(
k8s.V1Volume(name=quay,
secret=k8s.V1SecretVolumeSource(secret_name=quay)))
ctr.container.add_volume_mount(
k8s.V1VolumeMount(name=quay,
read_only=True,
mount_path=Pipeline.QUAY_SECRET_MOUNT_PATH))
# SSH Key
ssh_key = "ssh-key"
ctr.add_volume(
k8s.V1Volume(name=ssh_key,
secret=k8s.V1SecretVolumeSource(default_mode=0o600,
secret_name=ssh_key)))
ctr.container.add_volume_mount(
k8s.V1VolumeMount(name=ssh_key,
read_only=True,
mount_path="/root/.ssh"))
# Assemble the inputs for the next stage
consumable_inputs = {}
for k, v in file_outputs.items():
consumable_inputs[k] = (ctr.outputs[k], outputs[k])
return ctr, consumable_inputs
def _add_common_labels(op: dsl.ContainerOp) -> dsl.ContainerOp:
return op.add_pod_label('param', param)
def __init__(self, spec=None, metadata=None):
super().__init__(metadata, spec)
self._cop = ContainerOp("name", "image")
self.verbose = False
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"
def run_as_aiplatform_custom_job(
op: dsl.ContainerOp,
display_name: Optional[str] = None,
replica_count: Optional[int] = None,
machine_type: Optional[str] = None,
accelerator_type: Optional[str] = None,
accelerator_count: Optional[int] = None,
boot_disk_type: Optional[str] = None,
boot_disk_size_gb: Optional[int] = None,
timeout: Optional[str] = None,
restart_job_on_worker_restart: Optional[bool] = None,
service_account: Optional[str] = None,
network: Optional[str] = None,
output_uri_prefix: Optional[str] = None,
worker_pool_specs: Optional[List[Mapping[str, Any]]] = None,
) -> None:
"""Run a pipeline task using AI Platform (Unified) custom training job.
For detailed doc of the service, please refer to
https://cloud.google.com/ai-platform-unified/docs/training/create-custom-job
Args:
op: The task (ContainerOp) object to run as aiplatform custom job.
display_name: Optional. The name of the custom job.
replica_count: Optional. The number of replicas to be split between master
workerPoolSpec and worker workerPoolSpec. (master always has 1 replica).
machine_type: Optional. The type of the machine to run the custom job. The
default value is "n1-standard-4".
accelerator_type: Optional. The type of accelerator(s) that may be attached
to the machine as per acceleratorCount. Optional.
accelerator_count: Optional. The number of accelerators to attach to the
machine.
boot_disk_type: Optional. Type of the boot disk (default is "pd-ssd"). Valid
values: "pd-ssd" (Persistent Disk Solid State Drive) or "pd-standard"
(Persistent Disk Hard Disk Drive).
boot_disk_size_gb: Optional. Size in GB of the boot disk (default is 100GB).
timeout: Optional. The maximum job running time. The default is 7 days. A
duration in seconds with up to nine fractional digits, terminated by 's'.
Example: "3.5s"
restart_job_on_worker_restart: Optional. Restarts the entire CustomJob if a
worker gets restarted. This feature can be used by distributed training
jobs that are not resilient to workers leaving and joining a job.
service_account: Optional. Specifies the service account for workload run-as
account.
network: Optional. The full name of the Compute Engine network to which the
job should be peered. For example, projects/12345/global/networks/myVPC.
output_uri_prefix: Optional. Google Cloud Storage URI to output directory.
additional_worker_pool_specs: Optional. Additional workerPoolSpecs for
distributed training. For details, please see:
https://cloud.google.com/ai-platform-unified/docs/training/distributed-training
"""
job_spec = {}
if worker_pool_specs is not None:
worker_pool_specs = copy.deepcopy(worker_pool_specs)
def _is_output_parameter(output_key: str) -> bool:
return output_key in (
op.component_spec.output_definitions.parameters.keys())
for worker_pool_spec in worker_pool_specs:
if 'containerSpec' in worker_pool_spec:
container_spec = worker_pool_spec['containerSpec']
if 'command' in container_spec:
dsl_utils.resolve_cmd_lines(container_spec['command'],
_is_output_parameter)
if 'args' in container_spec:
dsl_utils.resolve_cmd_lines(container_spec['args'],
_is_output_parameter)
elif 'pythonPackageSpec' in worker_pool_spec:
# For custom Python training, resolve placeholders in args only.
python_spec = worker_pool_spec['pythonPackageSpec']
if 'args' in python_spec:
dsl_utils.resolve_cmd_lines(python_spec['args'],
_is_output_parameter)
else:
raise ValueError(
'Expect either "containerSpec" or "pythonPackageSpec" in each '
'workerPoolSpec. Got: {}'.format(custom_job_spec))
job_spec['workerPoolSpecs'] = worker_pool_specs
else:
worker_pool_spec = {
'machineSpec': {
'machineType': machine_type or _DEFAULT_CUSTOM_JOB_MACHINE_TYPE
},
'replicaCount': '1',
'containerSpec': {
'imageUri': op.container.image,
}
}
if op.container.command:
worker_pool_spec['containerSpec']['command'] = op.container.command
if op.container.args:
worker_pool_spec['containerSpec']['args'] = op.container.args
if accelerator_type is not None:
worker_pool_spec['machineSpec'][
'acceleratorType'] = accelerator_type
if accelerator_count is not None:
worker_pool_spec['machineSpec'][
'acceleratorCount'] = accelerator_count
if boot_disk_type is not None:
if 'diskSpec' not in worker_pool_spec:
worker_pool_spec['diskSpec'] = {}
worker_pool_spec['diskSpec']['bootDiskType'] = boot_disk_type
if boot_disk_size_gb is not None:
if 'diskSpec' not in worker_pool_spec:
worker_pool_spec['diskSpec'] = {}
worker_pool_spec['diskSpec']['bootDiskSizeGb'] = boot_disk_size_gb
job_spec['workerPoolSpecs'] = [worker_pool_spec]
if replica_count is not None and replica_count > 1:
additional_worker_pool_spec = copy.deepcopy(worker_pool_spec)
additional_worker_pool_spec['replicaCount'] = str(replica_count -
1)
job_spec['workerPoolSpecs'].append(additional_worker_pool_spec)
if timeout is not None:
if 'scheduling' not in job_spec:
job_spec['scheduling'] = {}
job_spec['scheduling']['timeout'] = timeout
if restart_job_on_worker_restart is not None:
if 'scheduling' not in job_spec:
job_spec['scheduling'] = {}
job_spec['scheduling'][
'restartJobOnWorkerRestart'] = restart_job_on_worker_restart
if service_account is not None:
job_spec['serviceAccount'] = service_account
if network is not None:
job_spec['network'] = network
if output_uri_prefix is not None:
job_spec['baseOutputDirectory'] = {
'outputUriPrefix': output_uri_prefix
}
op.custom_job_spec = {
'displayName': display_name or op.name,
'jobSpec': job_spec
}
def copy(self):
self._cop = None
fn = deepcopy(self)
self._cop = ContainerOp('name', 'image')
fn._cop = ContainerOp('name', 'image')
return fn