def download_pipeline_files(id): # noqa: E501
"""Returns the pipeline YAML compressed into a .tgz (.tar.gz) file.
:param id:
:type id: str
:rtype: file
"""
tar, bytes_io = create_tarfile(bucket_name="mlpipeline", prefix=f"pipelines/{id}",
file_extensions=[""],
keep_open=False)
if len(tar.members) == 0:
return f"Could not find pipeline with id '{id}'", 404
return bytes_io.getvalue(), 200, {"Content-Disposition": f"attachment; filename={id}.tgz"}
def download_model_files(id, include_generated_code=None): # noqa: E501
"""
Returns the model artifacts compressed into a .tgz (.tar.gz) file.
:param id:
:type id: str
:param include_generated_code: Include generated run scripts in download
:type include_generated_code: bool
:rtype: file | binary
"""
tar, bytes_io = create_tarfile(
bucket_name="mlpipeline",
prefix=f"models/{id}/",
file_extensions=[".yaml", ".yml", ".py", ".md"],
keep_open=include_generated_code)
if len(tar.members) == 0:
return f"Could not find model with id '{id}'", 404
if include_generated_code:
generate_code_response: ApiGenerateModelCodeResponse = generate_model_code(
id)[0]
for s in generate_code_response.scripts:
file_name = f"run_{s.pipeline_stage}_{s.execution_platform}.py"
if file_name in tar.getnames():
file_name = file_name.replace(".py", "_generated.py")
file_content = s.script_code
file_size = len(file_content)
file_obj = BytesIO(file_content.encode('utf-8'))
tarinfo = tarfile.TarInfo(name=file_name)
tarinfo.size = file_size
tar.addfile(tarinfo, file_obj)
tar.close()
return bytes_io.getvalue(), 200, {
"Content-Disposition": f"attachment; filename={id}.tgz"
}
def download_dataset_files(id, include_generated_code=None): # noqa: E501
"""
Returns the dataset artifacts compressed into a .tgz (.tar.gz) file.
:param id:
:type id: str
:param include_generated_code: Include generated run script in download
:type include_generated_code: bool
:rtype: file | binary
"""
tar, bytes_io = create_tarfile(
bucket_name="mlpipeline",
prefix=f"datasets/{id}/",
file_extensions=[".yaml", ".yml", ".py", ".md"],
keep_open=include_generated_code)
if len(tar.members) == 0:
return f"Could not find dataset with id '{id}'", 404
if include_generated_code:
generate_code_response, api_status = generate_dataset_code(id)
if api_status == 200:
file_content = generate_code_response.script
file_name = f"run_dataset.py"
if file_name in tar.getnames():
file_name = file_name.replace(".py", "_generated.py")
tarinfo = tarfile.TarInfo(name=file_name)
tarinfo.size = len(file_content)
file_obj = BytesIO(file_content.encode('utf-8'))
tar.addfile(tarinfo, file_obj)
tar.close()
return bytes_io.getvalue(), 200, {
"Content-Disposition": f"attachment; filename={id}.tgz"
}
def generate_notebook_run_script(api_notebook: ApiNotebook,
parameters: dict = {},
run_name: str = None,
hide_secrets: bool = True):
if "dataset_pvc" in parameters:
template_file = "run_notebook_with_dataset.TEMPLATE.py"
else:
template_file = "run_notebook.TEMPLATE.py"
with open(join(CODE_TEMPLATE_DIR, template_file), 'r') as f:
template_raw = f.read()
notebook_file = api_notebook.url.split("/")[-1]
requirements_url = get_object_url(bucket_name="mlpipeline",
prefix=f"notebooks/{api_notebook.id}/",
file_extensions=[".txt"],
file_name_filter="requirements")
cos_dependencies_archive_url = get_object_url(bucket_name="mlpipeline",
prefix=f"notebooks/{api_notebook.id}/",
file_extensions=[".tar.gz"],
file_name_filter="elyra-dependencies-archive")
if not cos_dependencies_archive_url:
tar, bytes_io = create_tarfile(bucket_name="mlpipeline",
prefix=f"notebooks/{api_notebook.id}/",
file_extensions=[".ipynb"])
cos_dependencies_archive_url = store_file(bucket_name="mlpipeline",
prefix=f"notebooks/{api_notebook.id}/",
file_name="elyra-dependencies-archive.tar.gz",
file_content=bytes_io.getvalue())
cos_dependencies_archive = cos_dependencies_archive_url.split("/")[-1]
# TODO: move this into a ApiNotebook.image as opposed to parsing yaml here
yaml_file_content = retrieve_file_content(bucket_name="mlpipeline",
prefix=f"notebooks/{api_notebook.id}/",
file_extensions=[".yaml", ".yml"])
metadata_yaml = yaml.load(yaml_file_content, Loader=yaml.FullLoader)
image = metadata_yaml["implementation"]["github"].get("image", "tensorflow/tensorflow:latest")
# TODO: elyra-ai/kfp-notebook generates output notebook as: "-output.ipynb"
# https://github.com/elyra-ai/kfp-notebook/blob/c8f1298/etc/docker-scripts/bootstrapper.py#L188-L190
# so here we may consider renaming the generated file with a datetimestamp
# output_folder = f"notebooks/{api_notebook.id}/runs/{datetime.now().strftime('%Y%m%d-%H%M%S')}"
# output_file_name = notebook_file_name.replace(r'.ipynb', '-output.ipynb')
# output_file_path = f"{output_folder}/{output_file_name}"
# output_file_url = f"http://{minio_host}:{minio_port}/mlpipeline/{output_file_path}"
kfp_url = f"'{_pipeline_service_url}'" if "POD_NAMESPACE" not in os.environ else ""
substitutions = {
"name": api_notebook.name,
"notebook": notebook_file,
"cos_bucket": "mlpipeline",
"cos_directory": f"notebooks/{api_notebook.id}/",
"cos_dependencies_archive": cos_dependencies_archive,
"cos_endpoint": "***",
"cos_username": "******",
"cos_password": "******",
"requirements_url": requirements_url or "",
"image": image,
"pipeline_server": kfp_url,
"run_name": run_name or api_notebook.name
}
# TODO: make the `dataset_pvc` and `mount_path` parameters part of the Swagger spec?
if "dataset_pvc" in parameters:
substitutions.update({
"dataset_pvc": parameters["dataset_pvc"],
"mount_path": parameters.get("mount_path", "/tmp/data")
})
if not hide_secrets:
substitutions.update({
"cos_endpoint": f"http://{minio_host}:{minio_port}/minio",
"cos_username": minio_access_key,
"cos_password": minio_secret_key
})
run_script = Template(template_raw).substitute(substitutions)
return run_script
