jobsworthy 0.5.0

No project description provided

Jobsworth

A set of utility functions and classes to aid in build Spark jobs on Azure databricks.

Job Configuration

Spark Job Module

Job provides a decorator which wraps the execution of a spark job. You use the decorator at the entry point for the job. At the moment it performs 1 function; calling all the registered initialisers.

from jobsworthy import spark_job

@spark_job.job()
def execute(args=None) -> monad.EitherMonad[value.JobState]:
    pass

To register initialisers (to be run just before the job function is called) do the following.

@spark_job.register()
def some_initialiser():
    ...

The initialisers must be imported before the job function is called; to ensure they are registered. To do that, either import them directly in the job module, or add them to a module __init__.py and import the module.

Model Library

Streamer

The Streamer module provides a fluent streaming abstraction on top of 2 hive repos (the from and to repos). The Streamer runs a pipeline as follows:

• Read from the source repo.
• Perform a transformation.
• Write to the target repo. This is where the stream starts and uses pyspark streaming to perform the read, transform and write.
• Wait for the stream to finish.

To setup a stream:

from jobsworthy import model

streamer = (model.STREAMER()
                 .stream_from(from_table)
                 .stream_to(to_table)
                 .with_transformer(transform_fn))

# Execute the stream
result = streamer.run()

# When successful it returns the Streamer wrapped in a Right.
# When there is a failure, it returns the error (subtype of JobError) wrapped in a Left   

assert result.is_right()

Some transformation functions require data from outside the input table. You can configure the streamer with additional transformation context by passing in kwargs on the with_transformer.

from dataclasses import dataclass
from jobsworthy import model


@dataclass
class TransformContext:
    run_id: int

def transform_fn_with_ctx(df, **kwargs):
    ...

    
streamer = (model.Streamer()
                 .stream_from(from_table)
                 .stream_to(to_table)
                 .with_transformer(transform_fn_with_ctx, run=TransformContext(run_id=1)))

When configuring the stream_to table, you can provide partition columns when writing the stream. Provide a tuple of column names.

streamer = model.STREAMER().stream_to(to_table, ('name', ))

Repository Module

The repo library offers a number of simple abstractions for managing Databrick/Spark databases and tables. It is by no means an object-mapper. Rather its a few classes with some simple functions we have found useful when working with Hive tables.

from jobsworthy import repo

SparkDB

Db is the base class representing a Hive Database. Once constructed it is provided to the hive table classes when they are constructed.

Db takes a spark session and a job config.

db = repo.Db(session=spark_test_session.create_session(), config=job_config())

When intialised it checks that the database (defined in the config) exists and creates it if it doesn't.

Hive Table

Util Module

Spark Session

Secrets

The Secrets module obtains secrets using the Databricks DBUtils secrets utility. The module acts as a wrapper for DButils. This allows for secrets to be mocked in tests without needing DBUtils. The CosmosDB repository is injected with the secrets provider to enable secured access to CosmosDB.

The provider requires access to the Spark session when running on Databricks. However this is not required in test. You also provide Secrets with a wrapper for DBUtils with also, optionally, takes a session. Both test and production wrappers are available in the util.databricks module.

from jobsworthy.util import secrets, databricks

provider = secrets.Secrets(session=di_container.session,
                           config=job_config(),
                           secrets_provider=databricks.DatabricksUtilsWrapper())

The default secret scope name is defined from the JobConfig properties; domain_name and data_product_name, separated by a .. This can be overridden by defining the scope on the Secrets constructor, or on the call to get_secret. It looks like this on the constructor.

provider = secrets.Secrets(session=di_container.session,
                           config=job_config(),
                           secrets_provider=databricks.DatabricksUtilsWrapper(),
                           default_scope_name="custom-scope-name")

Getting a secret.

provider.get_secret(secret_name="name-of-secret")  # returns an Either[secret-key]

Secrets is also able to return a ClientCredential using an Azure AD client credentials grant. The grant requires that client id and secrets are obtainable via DBUtils through key-vault with the key names defined in JobConfig in the properties client_id_key and client_secret_key

provider.client_credential_grant()   # returns an Either[ClientCredential]

Testing using secrets. DBUtils is not available as an open source project. When creating the secrets provider, you can provide a DBUtils mock class which is available. On this class you can also construct valid keys to be used for test (if required; the mock returns a dummy key response to any generic lookup).

The example below also shows how to use a non-default scope on the get_secrets function.

from jobsworthy.util import secrets, databricks

test_secrets = {"my_domain.my_data_product_name": {'my_secret': 'a-secret'},
                "alt_scope": {'my_secret': 'b-secret'}}

provider = secrets.Secrets(
    session=di_container.session,
    config=job_config(),
    secrets_provider=databricks.DatabricksUtilMockWrapper(spark_test_session.MockPySparkSession, test_secrets))

provider.get_secret(non_default_scope_name="alt_scope", secret_name="my_secret")