def _PrepareModelPath(
self, model_uri: Text,
serving_spec: infra_validator_pb2.ServingSpec) -> Text:
model_path = path_utils.serving_model_path(model_uri)
serving_binary = serving_spec.WhichOneof('serving_binary')
if serving_binary == 'tensorflow_serving':
# TensorFlow Serving requires model to be stored in its own directory
# structure flavor. If current model_path does not conform to the flavor,
# we need to make a copy to the temporary path.
try:
# Check whether current model_path conforms to the tensorflow serving
# model path flavor. (Parsed without exception)
tf_serving_flavor.parse_model_path(
model_path, expected_model_name=serving_spec.model_name)
except ValueError:
# Copy the model to comply with the tensorflow serving model path
# flavor.
temp_model_path = tf_serving_flavor.make_model_path(
model_base_path=self._get_tmp_dir(),
model_name=serving_spec.model_name,
version=int(time.time()))
io_utils.copy_dir(src=model_path, dst=temp_model_path)
return temp_model_path
return model_path
def testDoWithCustomSplits(self):
# Update input dict.
io_utils.copy_dir(
os.path.join(self._testdata_dir, 'iris/data/train'),
os.path.join(self._output_data_dir, 'data/training'))
io_utils.copy_dir(
os.path.join(self._testdata_dir, 'iris/data/eval'),
os.path.join(self._output_data_dir, 'data/evaluating'))
examples = standard_artifacts.Examples()
examples.uri = os.path.join(self._output_data_dir, 'data')
examples.split_names = artifact_utils.encode_split_names(
['training', 'evaluating'])
self._input_dict['examples'] = [examples]
# Update exec properties skeleton with custom splits.
self._exec_properties['train_args'] = json_format.MessageToJson(
trainer_pb2.TrainArgs(splits=['training'], num_steps=1000),
preserving_proto_field_name=True)
self._exec_properties['eval_args'] = json_format.MessageToJson(
trainer_pb2.EvalArgs(splits=['evaluating'], num_steps=500),
preserving_proto_field_name=True)
self._exec_properties['module_file'] = os.path.join(self._testdata_dir,
'module_file',
'tuner_module.py')
tuner = executor.Executor(self._context)
tuner.Do(
input_dict=self._input_dict,
output_dict=self._output_dict,
exec_properties=self._exec_properties)
self._verify_output()
def setUp(self):
super(CliAirflowEndToEndTest, self).setUp()
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup airflow_home in a temp directory
self._airflow_home = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', tempfile.mkdtemp()),
self._testMethodName)
self._old_airflow_home = os.environ.get('AIRFLOW_HOME')
os.environ['AIRFLOW_HOME'] = self._airflow_home
self._old_home = os.environ.get('HOME')
os.environ['HOME'] = self._airflow_home
tf.logging.info('Using %s as AIRFLOW_HOME and HOME in this e2e test',
self._airflow_home)
# Do not load examples to make this a bit faster.
os.environ['AIRFLOW__CORE__LOAD_EXAMPLES'] = 'False'
# Following environment variables make scheduler process dags faster.
os.environ['AIRFLOW__SCHEDULER__JOB_HEARTBEAT_SEC'] = '1'
os.environ['AIRFLOW__SCHEDULER__SCHEDULER_HEARTBEAT_SEC'] = '1'
os.environ['AIRFLOW__SCHEDULER__RUN_DURATION'] = '-1'
os.environ['AIRFLOW__SCHEDULER__MIN_FILE_PROCESS_INTERVAL'] = '1'
os.environ['AIRFLOW__SCHEDULER__PRINT_STATS_INTERVAL'] = '30'
# Using more than one thread results in a warning for sqlite backend.
# See https://github.com/tensorflow/tfx/issues/141
os.environ['AIRFLOW__SCHEDULER__MAX_THREADS'] = '1'
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(
os.path.abspath(__file__))))), 'examples',
'chicago_taxi_pipeline', '')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = tf.gfile.ListDirectory(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._airflow_home, 'taxi', 'data',
'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert tf.gfile.IsDirectory(target_data_dir)
content = tf.gfile.ListDirectory(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._airflow_home, 'taxi', 'taxi_utils.py'))
# Initialize database.
_ = subprocess.check_output(['airflow', 'initdb'])
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
def _PrepareModelPath(
self, model: types.Artifact,
serving_spec: infra_validator_pb2.ServingSpec) -> str:
model_path = path_utils.serving_model_path(
model.uri, path_utils.is_old_model_artifact(model))
serving_binary = serving_spec.WhichOneof('serving_binary')
if serving_binary == _TENSORFLOW_SERVING:
# TensorFlow Serving requires model to be stored in its own directory
# structure flavor. If current model_path does not conform to the flavor,
# we need to make a copy to the temporary path.
try:
# Check whether current model_path conforms to the tensorflow serving
# model path flavor. (Parsed without exception)
tf_serving_flavor.parse_model_path(
model_path, expected_model_name=serving_spec.model_name)
except ValueError:
# Copy the model to comply with the tensorflow serving model path
# flavor.
temp_model_path = tf_serving_flavor.make_model_path(
model_base_path=self._get_tmp_dir(),
model_name=serving_spec.model_name,
version=int(time.time()))
io_utils.copy_dir(src=model_path, dst=temp_model_path)
self._AddCleanup(io_utils.delete_dir,
self._context.get_tmp_path())
return temp_model_path
return model_path
def testDoWithCustomSplits(self):
# Update input dict.
io_utils.copy_dir(
os.path.join(self._testdata_dir, 'penguin/data/train'),
os.path.join(self._output_data_dir, 'data/training'))
io_utils.copy_dir(
os.path.join(self._testdata_dir, 'penguin/data/eval'),
os.path.join(self._output_data_dir, 'data/evaluating'))
examples = standard_artifacts.Examples()
examples.uri = os.path.join(self._output_data_dir, 'data')
examples.split_names = artifact_utils.encode_split_names(
['training', 'evaluating'])
self._input_dict[standard_component_specs.EXAMPLES_KEY] = [examples]
# Update exec properties skeleton with custom splits.
self._exec_properties[standard_component_specs.
TRAIN_ARGS_KEY] = proto_utils.proto_to_json(
trainer_pb2.TrainArgs(splits=['training'],
num_steps=1000))
self._exec_properties[standard_component_specs.
EVAL_ARGS_KEY] = proto_utils.proto_to_json(
trainer_pb2.EvalArgs(splits=['evaluating'],
num_steps=500))
self._exec_properties[
standard_component_specs.MODULE_FILE_KEY] = os.path.join(
self._testdata_dir, 'module_file', 'tuner_module.py')
tuner = executor.Executor(self._context)
tuner.Do(input_dict=self._input_dict,
output_dict=self._output_dict,
exec_properties=self._exec_properties)
self._verify_output()
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Copies over recorded data to pipeline output uri.
Args:
input_dict: Input dict from input key to a list of Artifacts.
output_dict: Output dict from output key to a list of Artifacts.
exec_properties: A dict of execution properties.
Returns:
None
Raises:
FileNotFoundError: If the recorded test data dir doesn't exist any more.
"""
for output_key, artifact_list in output_dict.items():
for idx, artifact in enumerate(artifact_list):
dest = artifact.uri
src = os.path.join(self._test_data_dir, self._component_id,
output_key, str(idx))
if not os.path.exists(src):
raise FileNotFoundError("{} does not exist".format(src))
io_utils.copy_dir(src, dest)
logging.info("Finished copying from %s to %s", src, dest)
def testDoWithCustomSplits(self):
# Update input dict.
io_utils.copy_dir(
os.path.join(self._source_data_dir,
'transform/transformed_examples/data/train'),
os.path.join(self._output_data_dir, 'data/training'))
io_utils.copy_dir(
os.path.join(self._source_data_dir,
'transform/transformed_examples/data/eval'),
os.path.join(self._output_data_dir, 'data/evaluating'))
examples = standard_artifacts.Examples()
examples.uri = os.path.join(self._output_data_dir, 'data')
examples.split_names = artifact_utils.encode_split_names(
['training', 'evaluating'])
self._input_dict[constants.EXAMPLES_KEY] = [examples]
# Update exec properties skeleton with custom splits.
self._exec_properties['train_args'] = json_format.MessageToJson(
trainer_pb2.TrainArgs(splits=['training'], num_steps=1000),
preserving_proto_field_name=True)
self._exec_properties['eval_args'] = json_format.MessageToJson(
trainer_pb2.EvalArgs(splits=['evaluating'], num_steps=500),
preserving_proto_field_name=True)
self._exec_properties['module_file'] = self._module_file
self._do(self._trainer_executor)
self._verify_model_exports()
self._verify_model_run_exports()
def testDoWithCustomSplits(self):
# Update input dict.
io_utils.copy_dir(
os.path.join(self._source_data_dir,
'transform/transformed_examples/data/train'),
os.path.join(self._output_data_dir, 'data/training'))
io_utils.copy_dir(
os.path.join(self._source_data_dir,
'transform/transformed_examples/data/eval'),
os.path.join(self._output_data_dir, 'data/evaluating'))
examples = standard_artifacts.Examples()
examples.uri = os.path.join(self._output_data_dir, 'data')
examples.split_names = artifact_utils.encode_split_names(
['training', 'evaluating'])
self._input_dict[standard_component_specs.EXAMPLES_KEY] = [examples]
# Update exec properties skeleton with custom splits.
self._exec_properties[standard_component_specs.
TRAIN_ARGS_KEY] = proto_utils.proto_to_json(
trainer_pb2.TrainArgs(splits=['training'],
num_steps=1000))
self._exec_properties[standard_component_specs.
EVAL_ARGS_KEY] = proto_utils.proto_to_json(
trainer_pb2.EvalArgs(splits=['evaluating'],
num_steps=500))
self._exec_properties[
standard_component_specs.MODULE_FILE_KEY] = self._module_file
self._do(self._trainer_executor)
self._verify_model_exports()
self._verify_model_run_exports()
def _CreateWarmupModel(self, blessing: types.Artifact, model_path: str,
warmup_requests: List[iv_types.Request]):
output_model_path = path_utils.stamped_model_path(blessing.uri)
io_utils.copy_dir(src=model_path, dst=output_model_path)
io_utils.write_tfrecord_file(
path_utils.warmup_file_path(output_model_path),
*[_convert_to_prediction_log(r) for r in warmup_requests])
blessing.set_int_custom_property(_MODEL_FLAG_KEY, 1)
def _create_tflite_compatible_saved_model(src: Text, dst: Text):
io_utils.copy_dir(src, dst)
assets_path = os.path.join(dst, tf.saved_model.ASSETS_DIRECTORY)
if fileio.exists(assets_path):
fileio.rmtree(assets_path)
assets_extra_path = os.path.join(dst, EXTRA_ASSETS_DIRECTORY)
if fileio.exists(assets_extra_path):
fileio.rmtree(assets_extra_path)
def record_pipeline(output_dir: Text,
metadata_db_uri: Optional[Text] = None,
host: Optional[Text] = None,
port: Optional[int] = None,
pipeline_name: Optional[Text] = None,
run_id: Optional[Text] = None) -> None:
"""Record pipeline run with run_id to output_dir.
For the beam pipeline, metadata_db_uri is required. For KFP pipeline,
host and port should be specified. If run_id is not specified, then
pipeline_name ought to be specified in order to fetch the latest execution
for the specified pipeline.
Args:
output_dir: Directory path where the pipeline outputs should be recorded.
metadata_db_uri: Uri to metadata db.
host: Hostname of the metadata grpc server
port: Port number of the metadata grpc server.
pipeline_name: Pipeline name, which is required if run_id isn't specified.
run_id: Pipeline execution run_id.
Raises:
ValueError: In cases of invalid arguments:
- metadata_db_uri is None or host and/or port is None.
- run_id is None and pipeline_name is None.
FileNotFoundError: if the source artifact uri does not already exist.
"""
if host is not None and port is not None:
metadata_config = metadata_store_pb2.MetadataStoreClientConfig(
host=host, port=port)
elif metadata_db_uri is not None:
metadata_config = metadata.sqlite_metadata_connection_config(
metadata_db_uri)
else:
raise ValueError('For KFP, host and port are required. '
'For beam pipeline, metadata_db_uri is required.')
with metadata.Metadata(metadata_config) as metadata_connection:
if run_id is None:
if pipeline_name is None:
raise ValueError('If the run_id is not specified,'
' pipeline_name should be specified')
# fetch executions of the most recently updated execution context.
executions = _get_latest_executions(metadata_connection,
pipeline_name)
else:
execution_dict = _get_execution_dict(metadata_connection)
if run_id in execution_dict:
executions = execution_dict[run_id]
else:
raise ValueError(
'run_id {} is not recorded in the MLMD metadata'.format(
run_id))
for src_uri, dest_uri in _get_paths(metadata_connection, executions,
output_dir):
io_utils.copy_dir(src_uri, dest_uri)
logging.info('Pipeline Recorded at %s', output_dir)
def testCopyDir(self):
old_path = os.path.join(self._base_dir, 'old', 'path')
new_path = os.path.join(self._base_dir, 'new', 'path')
io_utils.write_string_file(old_path, 'testing')
io_utils.copy_dir(os.path.dirname(old_path), os.path.dirname(new_path))
self.assertTrue(file_io.file_exists(new_path))
f = file_io.FileIO(new_path, mode='r')
self.assertEqual('testing', f.read())
self.assertEqual(7, f.tell())
def testCopyDir(self):
old_path = os.path.join(self._base_dir, 'old', 'path')
new_path = os.path.join(self._base_dir, 'new', 'path')
io_utils.write_string_file(old_path, 'testing')
io_utils.copy_dir(os.path.dirname(old_path), os.path.dirname(new_path))
self.assertTrue(file_io.file_exists(new_path))
f = file_io.FileIO(new_path, mode='r')
self.assertEqual('testing', f.read())
self.assertEqual(7, f.tell())
def setUp(self):
super(CliAirflowEndToEndTest, self).setUp()
# List of packages installed.
self._pip_list = str(subprocess.check_output(['pip', 'freeze', '--local']))
# Check if Apache Airflow is installed before running E2E tests.
if labels.AIRFLOW_PACKAGE_NAME not in self._pip_list:
sys.exit('Apache Airflow not installed.')
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup airflow_home in a temp directory
self._airflow_home = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', tempfile.mkdtemp()),
self._testMethodName, 'airflow')
self._old_airflow_home = os.environ.get('AIRFLOW_HOME')
os.environ['AIRFLOW_HOME'] = self._airflow_home
self._old_home = os.environ.get('HOME')
os.environ['HOME'] = self._airflow_home
absl.logging.info('Using %s as AIRFLOW_HOME and HOME in this e2e test',
self._airflow_home)
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
# Do not load examples to make this a bit faster.
os.environ['AIRFLOW__CORE__LOAD_EXAMPLES'] = 'False'
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__))))),
'examples', 'chicago_taxi_pipeline', '')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = tf.io.gfile.listdir(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._airflow_home, 'taxi', 'data', 'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert tf.io.gfile.isdir(target_data_dir)
content = tf.io.gfile.listdir(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._airflow_home, 'taxi', 'taxi_utils.py'))
self._airflow_initdb()
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Push model to target directory if blessed.
Args:
input_dict: Input dict from input key to a list of artifacts, including:
- model_export: exported model from trainer.
output_dict: Output dict from key to a list of artifacts, including:
- model_push: A list of 'ModelPushPath' artifact of size one. It will
include the model in this push execution if the model was pushed.
exec_properties: A dict of execution properties, including:
- push_destination: JSON string of pusher_pb2.PushDestination instance,
providing instruction of destination to push model.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
model_push = artifact_utils.get_single_instance(
output_dict[PUSHED_MODEL_KEY])
model_push_uri = model_push.uri
model_export = artifact_utils.get_single_instance(
input_dict[MODEL_KEY])
model_export_uri = model_export.uri
logging.info('Model pushing.')
# Copy the model to pushing uri.
model_path = path_utils.serving_model_path(model_export_uri)
model_version = str(int(time.time()))
# model_version = path_utils.get_serving_model_version(model_export_uri)
logging.info('Model version is %s', model_version)
io_utils.copy_dir(model_path,
os.path.join(model_push_uri, model_version))
logging.info('Model written to %s.', model_push_uri)
push_destination = pusher_pb2.PushDestination()
json_format.Parse(exec_properties['push_destination'],
push_destination)
serving_path = os.path.join(push_destination.filesystem.base_directory,
model_version)
if tf.io.gfile.exists(serving_path):
logging.info(
'Destination directory %s already exists, skipping current push.',
serving_path)
else:
# tf.serving won't load partial model, it will retry until fully copied.
io_utils.copy_dir(model_path, serving_path)
logging.info('Model written to serving path %s.', serving_path)
model_push.set_int_custom_property('pushed', 1)
model_push.set_string_custom_property('pushed_model', model_export_uri)
model_push.set_int_custom_property('pushed_model_id', model_export.id)
logging.info('Model pushed to %s.', serving_path)
def setUp(self):
super().setUp()
# List of packages installed.
self._pip_list = pip_utils.get_package_names()
# Check if Apache Airflow is installed before running E2E tests.
if labels.AIRFLOW_PACKAGE_NAME not in self._pip_list:
sys.exit('Apache Airflow not installed.')
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup airflow_home in a temp directory
self._airflow_home = os.path.join(self.tmp_dir, 'airflow')
self.enter_context(
test_case_utils.override_env_var('AIRFLOW_HOME', self._airflow_home))
self.enter_context(
test_case_utils.override_env_var('HOME', self._airflow_home))
absl.logging.info('Using %s as AIRFLOW_HOME and HOME in this e2e test',
self._airflow_home)
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
self._pipeline_name = 'chicago_taxi_simple'
self._pipeline_path = os.path.join(self._testdata_dir,
'test_pipeline_airflow_1.py')
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(os.path.abspath(__file__))))),
'examples', 'chicago_taxi_pipeline')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = fileio.listdir(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._airflow_home, 'taxi', 'data', 'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert fileio.isdir(target_data_dir)
content = fileio.listdir(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._airflow_home, 'taxi', 'taxi_utils.py'))
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
def setUp(self):
super(CliAirflowEndToEndTest, self).setUp()
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup airflow_home in a temp directory
self._airflow_home = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', tempfile.mkdtemp()),
self._testMethodName)
self._old_airflow_home = os.environ.get('AIRFLOW_HOME')
os.environ['AIRFLOW_HOME'] = self._airflow_home
self._old_home = os.environ.get('HOME')
os.environ['HOME'] = self._airflow_home
tf.logging.info('Using %s as AIRFLOW_HOME and HOME in this e2e test',
self._airflow_home)
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(
os.path.abspath(__file__))))), 'examples',
'chicago_taxi_pipeline', '')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = tf.gfile.ListDirectory(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._airflow_home, 'taxi', 'data',
'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert tf.gfile.IsDirectory(target_data_dir)
content = tf.gfile.ListDirectory(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._airflow_home, 'taxi', 'taxi_utils.py'))
# Initialize database.
_ = subprocess.check_output(['airflow', 'initdb'])
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
# Start scheduler.
self._scheduler = subprocess.Popen(['airflow', 'scheduler'])
def setUpClass(cls):
super(ExecutorTest, cls).setUpClass()
source_example_dir = os.path.join(cls._SOURCE_DATA_DIR,
'csv_example_gen')
io_utils.copy_dir(source_example_dir, cls._ARTIFACT1_URI)
io_utils.copy_dir(source_example_dir, cls._ARTIFACT2_URI)
# Duplicate the number of train and eval records such that
# second artifact has twice as many as first.
artifact2_pattern = os.path.join(cls._ARTIFACT2_URI, '*', '*')
artifact2_files = tf.io.gfile.glob(artifact2_pattern)
for filepath in artifact2_files:
directory, filename = os.path.split(filepath)
io_utils.copy_file(filepath,
os.path.join(directory, 'dup_' + filename))
def testCopyDirWithTrailingSlashes(self):
old_path1 = os.path.join(self._base_dir, 'old1', '')
old_path_file1 = os.path.join(old_path1, 'child', 'file')
new_path1 = os.path.join(self._base_dir, 'new1')
new_path_file1 = os.path.join(new_path1, 'child', 'file')
io_utils.write_string_file(old_path_file1, 'testing')
io_utils.copy_dir(old_path1, new_path1)
self.assertTrue(file_io.file_exists(new_path_file1))
old_path2 = os.path.join(self._base_dir, 'old2')
old_path_file2 = os.path.join(old_path2, 'child', 'file')
new_path2 = os.path.join(self._base_dir, 'new2', '')
new_path_file2 = os.path.join(new_path2, 'child', 'file')
io_utils.write_string_file(old_path_file2, 'testing')
io_utils.copy_dir(old_path2, new_path2)
self.assertTrue(file_io.file_exists(new_path_file2))
def testCopyDir(self):
old_path = os.path.join(self._base_dir, 'old')
old_path_file1 = os.path.join(old_path, 'file1')
old_path_file2 = os.path.join(old_path, 'dir', 'dir2', 'file2')
new_path = os.path.join(self._base_dir, 'new')
new_path_file1 = os.path.join(new_path, 'file1')
new_path_file2 = os.path.join(new_path, 'dir', 'dir2', 'file2')
io_utils.write_string_file(old_path_file1, 'testing')
io_utils.write_string_file(old_path_file2, 'testing2')
io_utils.copy_dir(old_path, new_path)
self.assertTrue(file_io.file_exists(new_path_file1))
f = file_io.FileIO(new_path_file1, mode='r')
self.assertEqual('testing', f.readline())
self.assertTrue(file_io.file_exists(new_path_file2))
f = file_io.FileIO(new_path_file2, mode='r')
self.assertEqual('testing2', f.readline())
def setUp(self):
super(CliBeamEndToEndTest, self).setUp()
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup beam_home in a temp directory
self._home = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', tempfile.mkdtemp()),
self._testMethodName)
self._old_home = os.environ.get('HOME')
os.environ['HOME'] = self._home
self._old_beam_home = os.environ.get('BEAM_HOME')
os.environ['BEAM_HOME'] = os.path.join(self._home, 'beam', '')
self._beam_home = os.environ['BEAM_HOME']
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(
os.path.abspath(__file__))))), 'examples',
'chicago_taxi_pipeline', '')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = tf.io.gfile.listdir(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._home, 'taxi', 'data', 'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert tf.io.gfile.isdir(target_data_dir)
content = tf.io.gfile.listdir(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._home, 'taxi', 'taxi_utils.py'))
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
def run_fn(fn_args: executor.TrainerFnArgs):
"""Train the model based on given args.
Args:
fn_args: Holds args used to train the model as name/value pairs.
"""
schema = io_utils.parse_pbtxt_file(fn_args.schema_file,
schema_pb2.Schema())
training_spec = _trainer_fn(fn_args, schema)
# Train the model
absl.logging.info('Training model.')
tf.estimator.train_and_evaluate(training_spec['estimator'],
training_spec['train_spec'],
training_spec['eval_spec'])
absl.logging.info('Training complete. Model written to %s',
fn_args.serving_model_dir)
# Export an eval savedmodel for TFMA
# NOTE: When trained in distributed training cluster, eval_savedmodel must be
# exported only by the chief worker (check TF_CONFIG).
absl.logging.info('Exporting eval_savedmodel for TFMA.')
eval_export_dir = path_utils.eval_model_dir(fn_args.model_run_dir)
tfma.export.export_eval_savedmodel(
estimator=training_spec['estimator'],
export_dir_base=eval_export_dir,
eval_input_receiver_fn=training_spec['eval_input_receiver_fn'])
absl.logging.info('Exported eval_savedmodel to %s.',
fn_args.eval_model_dir)
# TODO(b/160795287): Deprecate estimator based executor.
# Copy serving and eval model from model_run to model artifact directory.
serving_source = path_utils.serving_model_path(fn_args.model_run_dir)
io_utils.copy_dir(serving_source, fn_args.serving_model_dir)
absl.logging.info('Serving model copied to: %s.',
fn_args.serving_model_dir)
eval_source = path_utils.eval_model_path(fn_args.model_run_dir)
io_utils.copy_dir(eval_source, fn_args.eval_model_dir)
absl.logging.info('Eval model copied to: %s.', fn_args.eval_model_dir)
def testCopyDir(self):
self.createFiles({
'old': {
'file1.txt': 'testing',
'dir1': {
'dir2': {
'file2.txt': 'testing2'
}
}
}
})
io_utils.copy_dir(self.relpath('old'), self.relpath('new'))
self.assertDirectoryEqual(self.relpath('new'), {
'file1.txt': 'testing',
'dir1': {
'dir2': {
'file2.txt': 'testing2'
}
}
})
def setUp(self):
super().setUp()
# Change the encoding for Click since Python 3 is configured to use ASCII as
# encoding for the environment.
if codecs.lookup(locale.getpreferredencoding()).name == 'ascii':
os.environ['LANG'] = 'en_US.utf-8'
# Setup beam_home in a temp directory
self._home = self.tmp_dir
self._beam_home = os.path.join(self._home, 'beam')
self.enter_context(
test_case_utils.override_env_var('BEAM_HOME', self._beam_home))
self.enter_context(test_case_utils.override_env_var(
'HOME', self._home))
# Testdata path.
self._testdata_dir = os.path.join(
os.path.dirname(os.path.dirname(__file__)), 'testdata')
# Copy data.
chicago_taxi_pipeline_dir = os.path.join(
os.path.dirname(
os.path.dirname(
os.path.dirname(os.path.dirname(
os.path.abspath(__file__))))), 'examples',
'chicago_taxi_pipeline', '')
data_dir = os.path.join(chicago_taxi_pipeline_dir, 'data', 'simple')
content = fileio.listdir(data_dir)
assert content, 'content in {} is empty'.format(data_dir)
target_data_dir = os.path.join(self._home, 'taxi', 'data', 'simple')
io_utils.copy_dir(data_dir, target_data_dir)
assert fileio.isdir(target_data_dir)
content = fileio.listdir(target_data_dir)
assert content, 'content in {} is {}'.format(target_data_dir, content)
io_utils.copy_file(
os.path.join(chicago_taxi_pipeline_dir, 'taxi_utils.py'),
os.path.join(self._home, 'taxi', 'taxi_utils.py'))
# Initialize CLI runner.
self.runner = click_testing.CliRunner()
def setUp(self):
super(PenguinPipelineLocalEndToEndTest, self).setUp()
self._test_dir = os.path.join(
os.environ.get('TEST_UNDECLARED_OUTPUTS_DIR', self.get_temp_dir()),
self._testMethodName)
self._pipeline_name = 'penguin_test'
self._data_root = os.path.join(os.path.dirname(__file__), 'data')
# Create a data root for rolling window test
# - data
# - day1
# - penguins_processed.csv
# - day2
# - penguins_processed.csv
# - day3
# - penguins_processed.csv
self._data_root_span = os.path.join(self._test_dir, 'data')
io_utils.copy_dir(self._data_root,
os.path.join(self._data_root_span, 'day1'))
io_utils.copy_dir(self._data_root,
os.path.join(self._data_root_span, 'day2'))
io_utils.copy_dir(self._data_root,
os.path.join(self._data_root_span, 'day3'))
self._module_file = os.path.join(os.path.dirname(__file__),
'penguin_utils.py')
self._serving_model_dir = os.path.join(self._test_dir, 'serving_model')
self._pipeline_root = os.path.join(self._test_dir, 'tfx', 'pipelines',
self._pipeline_name)
self._metadata_path = os.path.join(self._test_dir, 'tfx', 'metadata',
self._pipeline_name, 'metadata.db')
def setUp(self):
super().setUp()
self._test_dir = self.tmp_dir
self.enter_context(test_case_utils.change_working_dir(self.tmp_dir))
self._test_output_dir = 'gs://{}/test_output'.format(self._BUCKET_NAME)
test_id = test_utils.random_id()
self._testdata_root = 'gs://{}/test_data/{}'.format(
self._BUCKET_NAME, test_id)
io_utils.copy_dir(self._TEST_DATA_ROOT, self._testdata_root)
self._data_root = os.path.join(self._testdata_root, 'external', 'csv')
self._transform_module = os.path.join(self._MODULE_ROOT,
'transform_module.py')
self._trainer_module = os.path.join(self._MODULE_ROOT,
'trainer_module.py')
self._serving_model_dir = os.path.join(self._testdata_root, 'output')
self.addCleanup(self._delete_test_dir, test_id)
def setUp(self):
super().setUp()
penguin_examples_dir = os.path.join(self._REPO_BASE, 'tfx', 'examples',
'penguin')
# The location of the penguin test data and schema. The input files are
# copied to a test-local location for each invocation, and cleaned up at the
# end of test.
penguin_test_data_root = os.path.join(penguin_examples_dir, 'data')
penguin_test_schema_file = os.path.join(penguin_examples_dir, 'schema',
'user_provided', 'schema.pbtxt')
# The location of the user module for penguin. Will be packaged and copied
# to under the pipeline root before pipeline execution.
self._penguin_dependency_file = os.path.join(
penguin_examples_dir, 'penguin_utils_cloud_tuner.py')
self._penguin_data_root = os.path.join(self._testdata_root, 'data')
io_utils.copy_dir(penguin_test_data_root, self._penguin_data_root)
self._penguin_schema_file = os.path.join(self._testdata_root,
'schema.pbtxt')
io_utils.copy_file(penguin_test_schema_file, self._penguin_schema_file)
def testCopyDirWithTrailingSlashes(self):
self.createFiles({
'old': {
'dir': {
'file.txt': 'testing'
}
}
})
with self.subTest('Copy old/ to new1'):
io_utils.copy_dir(self.relpath('old', ''), self.relpath('new1'))
self.assertDirectoryEqual(self.relpath('new1'), {
'dir': {
'file.txt': 'testing'
}
})
with self.subTest('Copy old to new2/'):
io_utils.copy_dir(self.relpath('old'), self.relpath('new2', ''))
self.assertDirectoryEqual(self.relpath('new2'), {
'dir': {
'file.txt': 'testing'
}
})
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]) -> None:
"""Push model to target directory if blessed.
Args:
input_dict: Input dict from input key to a list of artifacts, including:
- model: exported model from trainer.
- model_blessing: model blessing path from model_validator. A push
action delivers the model exports produced by Trainer to the
destination defined in component config.
output_dict: Output dict from key to a list of artifacts, including:
- pushed_model: A list of 'ModelPushPath' artifact of size one. It will
include the model in this push execution if the model was pushed.
exec_properties: A dict of execution properties, including:
- push_destination: JSON string of pusher_pb2.PushDestination instance,
providing instruction of destination to push model.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
model_push = artifact_utils.get_single_instance(
output_dict[standard_component_specs.PUSHED_MODEL_KEY])
if not self.CheckBlessing(input_dict):
self._MarkNotPushed(model_push)
return
model_export = artifact_utils.get_single_instance(
input_dict[standard_component_specs.MODEL_KEY])
model_path = path_utils.serving_model_path(
model_export.uri, path_utils.is_old_model_artifact(model_export))
# Push model to the destination, which can be listened by a model server.
#
# If model is already successfully copied to outside before, stop copying.
# This is because model validator might blessed same model twice (check
# mv driver) with different blessing output, we still want Pusher to
# handle the mv output again to keep metadata tracking, but no need to
# copy to outside path again..
# TODO(jyzhao): support rpc push and verification.
push_destination = pusher_pb2.PushDestination()
proto_utils.json_to_proto(
exec_properties[standard_component_specs.PUSH_DESTINATION_KEY],
push_destination)
destination_kind = push_destination.WhichOneof('destination')
if destination_kind == 'filesystem':
fs_config = push_destination.filesystem
if fs_config.versioning == _Versioning.AUTO:
fs_config.versioning = _Versioning.UNIX_TIMESTAMP
if fs_config.versioning == _Versioning.UNIX_TIMESTAMP:
model_version = str(int(time.time()))
else:
raise NotImplementedError('Invalid Versioning {}'.format(
fs_config.versioning))
logging.info('Model version: %s', model_version)
serving_path = os.path.join(fs_config.base_directory,
model_version)
if fileio.exists(serving_path):
logging.info(
'Destination directory %s already exists, skipping current push.',
serving_path)
else:
# tf.serving won't load partial model, it will retry until fully copied.
io_utils.copy_dir(model_path, serving_path)
logging.info('Model written to serving path %s.', serving_path)
else:
raise NotImplementedError(
'Invalid push destination {}'.format(destination_kind))
# Copy the model to pushing uri for archiving.
io_utils.copy_dir(model_path, model_push.uri)
self._MarkPushed(model_push,
pushed_destination=serving_path,
pushed_version=model_version)
logging.info('Model pushed to %s.', model_push.uri)
def Do(self, input_dict: Dict[Text, List[types.Artifact]],
output_dict: Dict[Text, List[types.Artifact]],
exec_properties: Dict[Text, Any]):
"""Overrides the tfx_pusher_executor.
Args:
input_dict: Input dict from input key to a list of artifacts, including:
- model_export: exported model from trainer.
- model_blessing: model blessing path from model_validator.
output_dict: Output dict from key to a list of artifacts, including:
- model_push: A list of 'ModelPushPath' artifact of size one. It will
include the model in this push execution if the model was pushed.
exec_properties: Mostly a passthrough input dict for
tfx.components.Pusher.executor. custom_config.bigquery_serving_args is
consumed by this class. For the full set of parameters supported by
Big Query ML, refer to https://cloud.google.com/bigquery-ml/
Returns:
None
Raises:
ValueError:
If bigquery_serving_args is not in exec_properties.custom_config.
If pipeline_root is not 'gs://...'
RuntimeError: if the Big Query job failed.
"""
self._log_startup(input_dict, output_dict, exec_properties)
model_push = artifact_utils.get_single_instance(
output_dict[tfx_pusher_executor.PUSHED_MODEL_KEY])
if not self.CheckBlessing(input_dict):
self._MarkNotPushed(model_push)
return
model_export = artifact_utils.get_single_instance(
input_dict[tfx_pusher_executor.MODEL_KEY])
model_export_uri = model_export.uri
custom_config = json_utils.loads(
exec_properties.get(_CUSTOM_CONFIG_KEY, 'null'))
if custom_config is not None and not isinstance(custom_config, Dict):
raise ValueError('custom_config in execution properties needs to be a '
'dict.')
bigquery_serving_args = custom_config.get(SERVING_ARGS_KEY)
# if configuration is missing error out
if bigquery_serving_args is None:
raise ValueError('Big Query ML configuration was not provided')
bq_model_uri = '.'.join([
bigquery_serving_args[_PROJECT_ID_KEY],
bigquery_serving_args[_BQ_DATASET_ID_KEY],
bigquery_serving_args[_MODEL_NAME_KEY],
])
# Deploy the model.
io_utils.copy_dir(
src=path_utils.serving_model_path(model_export_uri), dst=model_push.uri)
model_path = model_push.uri
if not model_path.startswith(_GCS_PREFIX):
raise ValueError('pipeline_root must be gs:// for BigQuery ML Pusher.')
logging.info('Deploying the model to BigQuery ML for serving: %s from %s',
bigquery_serving_args, model_path)
query = _BQML_CREATE_OR_REPLACE_MODEL_QUERY_TEMPLATE.format(
model_uri=bq_model_uri, model_path=model_path)
# TODO(zhitaoli): Refactor the executor_class_path creation into a common
# utility function.
executor_class_path = '%s.%s' % (self.__class__.__module__,
self.__class__.__name__)
with telemetry_utils.scoped_labels(
{telemetry_utils.LABEL_TFX_EXECUTOR: executor_class_path}):
default_query_job_config = bigquery.job.QueryJobConfig(
labels=telemetry_utils.get_labels_dict())
client = bigquery.Client(default_query_job_config=default_query_job_config)
try:
query_job = client.query(query)
query_job.result() # Waits for the query to finish
except Exception as e:
raise RuntimeError('BigQuery ML Push failed: {}'.format(e))
logging.info('Successfully deployed model %s serving from %s', bq_model_uri,
model_path)
# Setting the push_destination to bigquery uri
self._MarkPushed(model_push, pushed_destination=bq_model_uri)
def Do(self, input_dict, output_dict, exec_properties):
"""Push model to target if blessed.
Args:
input_dict: Input dict from input key to a list of artifacts, including:
- model_export: exported model from trainer.
- model_blessing: model blessing path from model_validator.
output_dict: Output dict from key to a list of artifacts, including:
- model_push: A list of 'ModelPushPath' artifact of size one. It will
include the model in this push execution if the model was pushed.
exec_properties: A dict of execution properties, including:
- push_destination: JSON string of pusher_pb2.PushDestination instance,
providing instruction of destination to push model.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
model_export = types.get_single_instance(input_dict['model_export'])
model_export_uri = model_export.uri
model_blessing_uri = types.get_single_uri(input_dict['model_blessing'])
model_push = types.get_single_instance(output_dict['model_push'])
model_push_uri = model_push.uri
# TODO(jyzhao): should this be in driver or executor.
if not tf.gfile.Exists(os.path.join(model_blessing_uri, 'BLESSED')):
model_push.set_int_custom_property('pushed', 0)
tf.logging.info('Model on %s was not blessed', )
return
tf.logging.info('Model pushing.')
# Copy the model we are pushing into
model_path = path_utils.serving_model_path(model_export_uri)
# Note: we do not have a logical model version right now. This
# model_version is a timestamp mapped to trainer's exporter.
model_version = os.path.basename(model_path)
tf.logging.info('Model version is %s', model_version)
io_utils.copy_dir(model_path,
os.path.join(model_push_uri, model_version))
tf.logging.info('Model written to %s.', model_push_uri)
# Copied to a fixed outside path, which can be listened by model server.
#
# If model is already successfully copied to outside before, stop copying.
# This is because model validator might blessed same model twice (check
# mv driver) with different blessing output, we still want Pusher to
# handle the mv output again to keep metadata tracking, but no need to
# copy to outside path again..
# TODO(jyzhao): support rpc push and verification.
push_destination = pusher_pb2.PushDestination()
json_format.Parse(exec_properties['push_destination'],
push_destination)
serving_path = os.path.join(push_destination.filesystem.base_directory,
model_version)
if tf.gfile.Exists(serving_path):
tf.logging.info(
'Destination directory %s already exists, skipping current push.',
serving_path)
else:
# tf.serving won't load partial model, it will retry until fully copied.
io_utils.copy_dir(model_path, serving_path)
tf.logging.info('Model written to serving path %s.', serving_path)
model_push.set_int_custom_property('pushed', 1)
model_push.set_string_custom_property('pushed_model', model_export_uri)
model_push.set_int_custom_property('pushed_model_id', model_export.id)
tf.logging.info('Model pushed to %s.', serving_path)
if exec_properties.get('custom_config'):
cmle_serving_args = exec_properties.get(
'custom_config', {}).get('cmle_serving_args')
if cmle_serving_args is not None:
return cmle_runner.deploy_model_for_serving(
serving_path, model_version, cmle_serving_args,
exec_properties['log_root'])
def Do(self, input_dict,
output_dict,
exec_properties):
"""Push model to target if blessed.
Args:
input_dict: Input dict from input key to a list of artifacts, including:
- model_export: exported model from trainer.
- model_blessing: model blessing path from model_validator.
output_dict: Output dict from key to a list of artifacts, including:
- model_push: A list of 'ModelPushPath' artifact of size one. It will
include the model in this push execution if the model was pushed.
exec_properties: A dict of execution properties, including:
- push_destination: JSON string of pusher_pb2.PushDestination instance,
providing instruction of destination to push model.
Returns:
None
"""
self._log_startup(input_dict, output_dict, exec_properties)
model_export = types.get_single_instance(input_dict['model_export'])
model_export_uri = model_export.uri
model_blessing_uri = types.get_single_uri(input_dict['model_blessing'])
model_push = types.get_single_instance(output_dict['model_push'])
model_push_uri = model_push.uri
# TODO(jyzhao): should this be in driver or executor.
if not tf.gfile.Exists(os.path.join(model_blessing_uri, 'BLESSED')):
model_push.set_int_custom_property('pushed', 0)
tf.logging.info('Model on %s was not blessed',)
return
tf.logging.info('Model pushing.')
# Copy the model we are pushing into
model_path = path_utils.serving_model_path(model_export_uri)
# Note: we do not have a logical model version right now. This
# model_version is a timestamp mapped to trainer's exporter.
model_version = os.path.basename(model_path)
tf.logging.info('Model version is %s', model_version)
io_utils.copy_dir(model_path, os.path.join(model_push_uri, model_version))
tf.logging.info('Model written to %s.', model_push_uri)
# Copied to a fixed outside path, which can be listened by model server.
#
# If model is already successfully copied to outside before, stop copying.
# This is because model validator might blessed same model twice (check
# mv driver) with different blessing output, we still want Pusher to
# handle the mv output again to keep metadata tracking, but no need to
# copy to outside path again..
# TODO(jyzhao): support rpc push and verification.
push_destination = pusher_pb2.PushDestination()
json_format.Parse(exec_properties['push_destination'], push_destination)
serving_path = os.path.join(push_destination.filesystem.base_directory,
model_version)
if tf.gfile.Exists(serving_path):
tf.logging.info(
'Destination directory %s already exists, skipping current push.',
serving_path)
else:
# tf.serving won't load partial model, it will retry until fully copied.
io_utils.copy_dir(model_path, serving_path)
tf.logging.info('Model written to serving path %s.', serving_path)
model_push.set_int_custom_property('pushed', 1)
model_push.set_string_custom_property('pushed_model', model_export_uri)
model_push.set_int_custom_property('pushed_model_id', model_export.id)
tf.logging.info('Model pushed to %s.', serving_path)
if exec_properties.get('custom_config'):
cmle_serving_args = exec_properties.get('custom_config',
{}).get('cmle_serving_args')
if cmle_serving_args is not None:
return cmle_runner.deploy_model_for_serving(serving_path, model_version,
cmle_serving_args,
exec_properties['log_root'])
