---
title: Custom Connectors
slug: /connectors/custom-connectors
---

# 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.

{% note %}

This guide is based on a working example in the OpenMetadata Demos repository: [link](https://github.com/open-metadata/openmetadata-demo/tree/main/custom-connector).

We'd recommend to go through the example to better understand how all the pieces should look like.

{% /note %}

Watch OpenMetadata's [Webinar on Custom Connectors](https://www.youtube.com/watch?v=fDUj30Ub9VE)

{% youtube videoId="fDUj30Ub9VE" start="0:00" end="36:33" width="800px" height="450px" /%}

## 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 ([docs](https://docs.open-metadata.org/latest/sdk/python/build-connector/source#for-consumers-of-openmetadata-ingestion-to-define-custom-connectors-in-their-own-package-with-same-namespace)).

In [connector/my_awesome_connector.py](https://github.com/open-metadata/openmetadata-demo/blob/main/custom-connector/connector/my_awesome_connector.py) you have a minimal example of it.

Note how te important method is the `_iter`. This is the generator function that will be iterated over
to send all the Create Entity Requests to the `Sink`. Read more about the `Workflow` [here](https://docs.open-metadata.org/latest/sdk/python/build-connector).

### 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](/sdk/python).

We do not have docs and examples of all the supported Services. A way to get examples on how to create and fetch
other types of Entities is to directly refer to the `ometa` [integration tests](https://github.com/open-metadata/OpenMetadata/tree/main/ingestion/tests/integration/ometa).

#### Either & StackTraceError

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

```python
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:

```python
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

We'll need to package the code so that it can be shipped to the ingestion container and used there. In this demo
you can find a simple `setup.py` that builds the `connector` module.

### 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.

We will be running the demo against the OpenMetadata version `1.4.4`, therefore, our Dockerfile looks like:

```Dockerfile
# Base image from the right version
FROM openmetadata/ingestion:1.4.4

# 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

We have a `Makefile` prepared for you to run `make run`. This will get OpenMetadata up in Docker Compose using the
custom Ingestion image.

### 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.

{% image
  src="/images/v1.8/connectors/custom-connector.png"
  alt="Custom Connector" /%}