ibm-orchestration-pipelines 1.0.4

Python utilities for IBM Orchestration Pipelines

IBM Orchestration Pipelines Python Client

This package provides various utilities for working with IBM Orchestration Pipelines. Its primary usage is to enable users to store artifact results of a notebook run.

Usage

Construction

OrchestrationPipelines client is constructed from IAM/CPD APIKEY, which can be provided in a few ways:

• explicitly:

  from ibm_orchestration_pipelines import OrchestrationPipelines
  # IAM APIKEY
  client = OrchestrationPipelines(apikey)
  # or CPD APIKEY
  client = OrchestrationPipelines(apikey, url=url, username=username)
  # or
  client = OrchestrationPipelines.from_apikey(apikey, url=url, username=username)
  # or
  client = OrchestrationPipelines.from_token(token, url=url)
  

• implicitly:

  APIKEY=...
  export APIKEY
  USER_NAME=...
  export USER_NAME
  

  or

  USER_ACCESS_TOKEN=...
  export USER_ACCESS_TOKEN
  

  from ibm_orchestration_pipelines import OrchestrationPipelines
  
  # use APIKEY
  client = OrchestrationPipelines.from_apikey()
  
  # use USER_ACCESS_TOKEN
  client = OrchestrationPipelines.from_token()
  
  # try APIKEY, if absent then USER_ACCESS_TOKEN:
  client = OrchestrationPipelines()
  # or
  client = OrchestrationPipelines.new_instance()
  

All of the above may also define service_name and url.

The exact procedure of deciding which authentication method to use:

1. If from_apikey or from_token is used, the method is forced.
2. If constructor is used but either apikey or bearer_token argument was provided, that method will be forced (if both are present, an overloading error will be raised). Note that providing a nameless argument is equivalent to providing apikey.
3. If constructor or new_instance is used, APIKEY env-var is used.
4. If constructor or new_instance is used, but APIKEY env-var is not present, USER_ACCESS_TOKEN env-var is used.
5. If none of the above matches, an error is returned.

Usage in Python notebooks

Notebooks run in IBM Orchestration Pipelines get inputs and expose outputs as a node:

{
  "id": ...,
  "type": "execution_node",
  "op": "run_container",
  "app_data": {
    "pipeline_data": {
      "name": ...,
      "config": {
        "link": {
          "component_id_ref": "run-notebook"
        }
      },
      "inputs": [
        ...,
        {
          "name": "model_name",
          "group": "env_variables",
          "type": "String",
          "value_from": ...
        }
      ],
      "outputs": [
        {
          "name": "trained_model",
          "group": "output_variables",
          "type": {
            "CPDPath": {
              "path_type": "resource",
              "resource_type": "asset",
              "asset_type": "wml_model"
            }
          }
        }
      ]
    }
  },
  ...
}

Inside of the notebook, inputs are available as environmental variables:

model_name = os.environ['model_name']

Outputs are exposed using sdk method, store_results:

client = WSPipelines.from_apikey(...)
client.store_results({
  "trained_model": ... // cpd path to the trained model
})

Extracting credentials

On public cloud, this client provides a method for easy retrieval of WML instance credentials and scope storage credentials:

client.get_wml_credentials() # the scope passed in notebook
# or
client.get_wml_credentials("cpd:///projects/123456789")

client.get_storage_credentials() # the scope passed in notebook
# or
client.get_storage_credentials("cpd:///projects/123456789")

Note how the result will vary depending on the authentication method used to create the client.

CPD-Path manipulation

CPD-Path parsing is manipulation is also supported:

from ibm_orchestration_pipelines import CpdScope, OrchestrationPipelines

client = OrchestrationPipelines.from_apikey()

scope = CpdScope.from_string("cpd:///projects/123456789")

assert scope.scope_type() == "projects"
assert scope.scope_id() == "123456789"

client.get_wml_credentials(scope)

Different kinds of CPD-Paths will have different properties, providing the same interface across scopes, resource and file paths:

from ibm_orchestration_pipelines import CpdPath

scope_file_path = CpdPath.from_string("cpd:///projects/123456789/files/abc/def")
assert scope_file_path.scope_type() == "projects"
assert scope_file_path.scope_id() == "123456789"
assert scope_file_path.file_path() == "/abc/def"

connection_path = CpdPath.from_string("cpd:///projects/123456789/connections/3141592")
assert connection_path.scope_type() == "projects"
assert connection_path.scope_id() == "123456789"
assert connection_path.resource_type() == "connections"
assert connection_path.resource_id() == "3141592"

connection_file_path = CpdPath.from_string("cpd:///projects/123456789/connections/3141592/files/~/abc/def")
assert connection_file_path.scope_type() == "projects"
assert connection_file_path.scope_id() == "123456789"
assert connection_file_path.resource_type() == "connections"
assert connection_file_path.resource_id() == "3141592"
assert connection_file_path.bucket_name() == "~"
assert connection_file_path.file_path() == "/abc/def"

...additionally, for non-scope paths the scope can be extracted, if present:

from ibm_orchestration_pipelines import CpdPath

scope_path = CpdPath.from_string("cpd:///projects/123456789")
connection_path = CpdPath.from_string("cpd:///projects/123456789/connections/3141592")
assert connection_path.scope() == scope_path

Custom components for use in the pipeline

A custom pipeline component executes a script you write. You can use custom components to share reusable scripts between pipelines.

You create custom components as project assets. You can then use the components in pipelines you create in that project. You can create as many custom components for pipelines as needed. Currently, to create a custom component you must create one programmatically, using a Python function.

Creating a component as a project asset

To create a custom component, use the Python client to authenticate with IBM Orchestration Pipelines, code the component, then publish the component to the specified project. After it is available in the project, you can assign it to a node in a pipeline and run it as part of a pipeline flow.

This example demonstrates the process of publishing a component that adds two numbers together.

Publish a function as a component with the latest Python client. Run the following code in a Jupyter notebook in a project of your Cloud Pak for Data.

# Install libraries
! pip install ibm-orchestration-pipelines==1.0.4

# Authentication
from ibm_orchestration_pipelines import OrchestrationPipelines

apikey = ''
service_url = 'your_host_url'
project_id = 'your_project_id'
username = ''

client = OrchestrationPipelines.from_apikey(apikey, url=service_url, username=username)


# Define the function of the component

# If you define the input parameters, users are required to 
# input them in the UI

def add_two_numbers(a: int, b: int) -> int:
  print('Adding numbers: {} + {}.'.format(a, b))
  return a + b + 10


# Other possible functions might be sending a Slack message,
# or listing directories in a storage volume, and so on.

# Publish the component    
client.publish_component(
  name='Add numbers',  # Appears in UI as component name 
  func=add_two_numbers,
  description='Custom component adding numbers',  # Appears in UI as component description 
  project_id=project_id,
  overwrite=True,  # Overwrites an existing component with the same name 
)

Manage pipeline components

• list components from a project:

client.get_components(project_id=project_id)

• get a component by ID:

client.get_component(project_id=project_id, component_id=component_id)

• get a component by name:

client.get_component(project_id=project_id, name=component_name)	

• publish a new component:

client.publish_component(component name)

• delete a component by ID:

client.delete_component(project_id=project_id, component_id=component_id)