Custom Connectors

Each of the services support providing a Custom Connector. It should be a Python class available in the Python environment running the ingestion process (e.g., EC2 instance, Airflow host, Docker Image…). It should also match specific constraints on the methods to implement and how to send the Entities to be created. In this guide, we’ll walk through a possible implementation. The example is based on a Database Service, but the process is the same for Pipelines, Dashboard or Messaging services.

Review the OpenMetadata 2.0 custom connector examples for a package, image, and CLI configuration example compatible with OpenMetadata 2.0.

Watch OpenMetadata’s Webinar on Custom Connectors

Steps to Set Up a Custom Connector

Step 1 - Prepare your Connector

A connector is a class that extends from from metadata.ingestion.api.steps import Source. It should implement all the required methods in the Source API reference. Review the Custom Database connector example for a complete Source implementation. The important method is _iter. This generator function sends Create Entity Requests to the Sink. Read more about the Workflow in the Python SDK reference.

Step 2 - Yield the Data

The Sink is expecting Create Entity Requests. To get familiar with the Python SDK and understand how to create the different Entities, a recommended read is the Python SDK docs. We do not have docs and examples of all the supported Services. To see how to create and fetch other types of Entities, refer to the ometa integration tests.

Either & StackTraceError

When we yield the data, we are now wrapping the state of the execution being correct or not with an Either class:

from metadata.ingestion.api.models import Either, StackTraceError

This Either will have a left or right, and we will either return:

right with the correct CreateEntityRequest
left with the exception that we want to track with StackTraceError.

For example:

try:
    1 / 0
except Exception:
    yield Either(
        left=StackTraceError(
            name="My Error",
            error="Demoing one error",
            stack_trace=traceback.format_exc(),
        )
    )

for row in self.data:
    yield Either(
        right=CreateTableRequest(
            ...
        )
    )

Note that with the new structure, any errors are going to be properly logged at the end of the execution as:

+--------+---------------+-------------------+--------------------------------------------------------------------------------------------------------------------------+
| From   | Entity Name   | Message           | Stack Trace                                                                                                              |
+========+===============+===================+==========================================================================================================================+
| Source | My Error      | Demoing one error | Traceback (most recent call last):                                                                                       |
|        |               |                   |   File "/Users/pmbrull/github/openmetadata-demo/custom-connector/connector/my_csv_connector.py", line 182, in yield_data |
|        |               |                   |     1 / 0                                                                                                                |
|        |               |                   | ZeroDivisionError: division by zero                                                                                      |
+--------+---------------+-------------------+--------------------------------------------------------------------------------------------------------------------------+

Step 3 - Prepare the Package Installation

Package the code so the ingestion container can install and use it. Add a setup.py or pyproject.toml file to your connector package.

Step 4 - Prepare the Ingestion Image

If you want to use the connector from the UI, the Python environment running the ingestion process should contain the new code you just created. For example, if running via Docker, the openmetadata-ingestion image should be aware of your new package. Use the ingestion image tag that matches your deployed OpenMetadata version. For example:

# Base image from the right version
FROM openmetadata/ingestion:<version>

# Let's use the same workdir as the ingestion image
WORKDIR ingestion
USER airflow

# Install our custom connector
# For a PROD image, this could be picking up the package from your private package index
COPY connector connector
COPY setup.py .
RUN pip install --no-deps .

Step 5 - Run OpenMetadata with the Custom Ingestion Image

Build and run the custom ingestion image with your deployment method, such as Docker Compose or Kubernetes.

Step 6 - Configure the Connector

In the example we prepared a Database Connector. Thus, go to Database Services > Add New Service > Custom and set the Source Python Class Name as connector.my_awesome_connector.MyAwesomeConnector. Note how we are specifying the full module name so that the Ingestion Framework can import the Source class.

​Custom Connectors

​Steps to Set Up a Custom Connector

​Step 1 - Prepare your Connector

​Step 2 - Yield the Data

​Either & StackTraceError

​Step 3 - Prepare the Package Installation

​Step 4 - Prepare the Ingestion Image

​Step 5 - Run OpenMetadata with the Custom Ingestion Image

​Step 6 - Configure the Connector

Custom Connectors

Steps to Set Up a Custom Connector

Step 1 - Prepare your Connector

Step 2 - Yield the Data

Either & StackTraceError

Step 3 - Prepare the Package Installation

Step 4 - Prepare the Ingestion Image

Step 5 - Run OpenMetadata with the Custom Ingestion Image

Step 6 - Configure the Connector