openlineage-dagster 0.16.1

OpenLineage integration with Dagster

Note
New integration maintainers are needed! Please open an issue to get started.

OpenLineage Dagster Integration

A library that integrates Dagster with OpenLineage for automatic metadata collection. It provides an OpenLineage sensor, a Dagster sensor that tails Dagster event logs for tracking metadata. On each sensor evaluation, the function processes a batch of event logs, converts Dagster events into OpenLineage events, and emits them to an OpenLineage backend.

Features

Metadata

• Dagster job & op lifecycle

Requirements

• Python 3.7
• Dagster 0.13.8+

Installation

$ python -m pip install openlineage-dagster

Usage

OpenLineage Sensor & Event Log Storage Requirements

Single OpenLineage sensor per Dagster instance
As the OpenLineage sensor processes all event logs for a given Dagster instance, define and enable only one sensor per instance. Running multiple sensors will result in duplicate OpenLineage job runs being emitted for Dagster steps with different OpenLineage run IDs. These are dynamically generated during sensor runs.

Non-sharded Event Log Storage
For the sensor to handle all event logs across runs, use non-sharded event log storage. If an event log storage sharded by run (i.e., the default SqliteEventLogStorage) is used, the cursor that tracks the last processed event storage ID may not update properly.

OpenLineage Sensor Setup

Get a OpenLineage sensor definition from the openlineage_sensor() factory function and add it to your Dagster repository.

from dagster import repository
from openlineage.dagster.sensor import openlineage_sensor


@repository
def my_repository():
    openlineage_sensor_def = openlineage_sensor()
    return other_defs + [openlineage_sensor_def]

With parallel sensor runs not supported at the time of writing, some tuning may be necessary to avoid affecting other sensors' performance.

See Dagster's documentation on Evaluation Interval for more detail on minimum_interval_seconds, which defaults to 30 seconds. record_filter_limit is the maximum number of event logs to process on each sensor evaluation, and it defaults to 30 records per evaluation. Default values can be overridden:

@repository
def my_repository():
    openlineage_sensor_def = openlineage_sensor(
        minimum_interval_seconds=60,
        record_filter_limit=60,
    )
    return other_defs + [openlineage_sensor_def]

The OpenLineage sensor handles event logs in ascending order of storage ID, and starts with the first log by default. Optionally, after_storage_id can be specified to customize the starting point. This is only applicable when the cursor is undefined or has been deleted.

@repository
def my_repository():
    openlineage_sensor_def = openlineage_sensor(
        after_storage_id=100
    )
    return other_defs + [openlineage_sensor_def]

OpenLineage Adapter & Client Configuration

The sensor uses an OpenLineage adapter and client to convert and push data to an OpenLineage backend, These depend on environment variables.

If using User Repository Deployments, add the below variables to the repository where the sensor is defined. Otherwise, add the variables to the Dagster Daemon.

• OPENLINEAGE_URL - point to the service which will consume OpenLineage events
• OPENLINEAGE_API_KEY - set if the consumer of OpenLineage events requires a Bearer authentication key
• OPENLINEAGE_NAMESPACE - set if you are using something other than the default as the default namespace when a Dagster repository is undefined

OpenLineage Namespace & Dagster Repository

For Dagster jobs organized in repositories, Dagster keeps track of the repository name for each pipeline run. When the repository name is present, it is always used as the OpenLineage namespace name. OPENLINEAGE_NAMESPACE option is a way to fall back and provide some other static default value.

Development

To install all dependencies for local development:

$ python -m pip install -e .[dev]  # or python -m pip install -e .\[dev\] in zsh 

To run the test suite:

$ pytest