When running ingestion workflows from MWAA we have three approaches:
1. Install the openmetadata-ingestion package as a requirement in the Airflow environment. We will then run the process using a `PythonOperator`
2. Configure an ECS cluster and run the ingestion as an `ECSOperator`.
3. Install a plugin and run the ingestion with the `PythonVirtualenvOperator`.
We will now discuss pros and cons of each aspect and how to configure them.
## Ingestion Workflows as a Python Operator
### PROs
- It is the simplest approach
- We don’t need to spin up any further infrastructure
### CONs
- We need to install the [openmetadata-ingestion](https://pypi.org/project/openmetadata-ingestion/) package in the MWAA environment
- The installation can clash with existing libraries
- Upgrading the OM version will require to repeat the installation process
To install the package, we need to update the `requirements.txt` file from the MWAA environment to add the following line:
```
openmetadata-ingestion[<plugin>]==x.y.z
```
Where `x.y.z` is the version of the OpenMetadata ingestion package. Note that the version needs to match the server version. If we are using the server at 1.3.1, then the ingestion package needs to also be 1.3.1.
The plugin parameter is a list of the sources that we want to ingest. An example would look like this `openmetadata-ingestion[mysql,snowflake,s3]==1.3.1`.
A DAG deployed using a Python Operator would then look like follows
```python
import json
from datetime import timedelta
from airflow import DAG
try:
from airflow.operators.python import PythonOperator
except ModuleNotFoundError:
from airflow.operators.python_operator import PythonOperator
from airflow.utils.dates import days_ago
from metadata.workflow.metadata import MetadataWorkflow
- We need to set up an ECS cluster and the required policies in MWAA to connect to ECS and handle Log Groups.
We will now describe the steps, following the official AWS documentation.
### 1. Create an ECS Cluster & Task Definition
- The cluster needs a task to run in `FARGATE` mode.
- The required image is `docker.getcollate.io/openmetadata/ingestion-base:x.y.z`
- The same logic as above applies. The `x.y.z` version needs to match the server version. For example, `docker.getcollate.io/openmetadata/ingestion-base:1.3.1`
We have tested this process with a Task Memory of 512MB and Task CPU (unit) of 256. This can be tuned depending on the amount of metadata that needs to be ingested.
When creating the Task Definition, take notes on the **log groups** assigned, as we will need them to prepare the MWAA Executor Role policies.
For example, if in the JSON from the Task Definition we see:
```json
"logConfiguration": {
"logDriver": "awslogs",
"options": {
"awslogs-create-group": "true",
"awslogs-group": "/ecs/openmetadata",
"awslogs-region": "us-east-2",
"awslogs-stream-prefix": "ecs"
},
"secretOptions": []
}
```
We'll need to use the `/ecs/openmetadata` below when configuring the policies.
### 2. Task Definition ARN & Networking
1. From the AWS Console, copy your task definition ARN. It will look something like this `arn:aws:ecs:<region>:<account>:task-definition/<name>:<revision>`.
2. Get the network details on where the task should execute. We will be using a JSON like:
```json
{
"awsvpcConfiguration": {
"subnets": [
"subnet-xxxyyyzzz",
"subnet-xxxyyyzzz"
],
"securityGroups": [
"sg-xxxyyyzzz"
],
"assignPublicIp": "ENABLED"
}
}
```
{% note %}
If you want to extract MWAA metadata, add the **VPC**, **subnets** and **security groups** used when setting up MWAA. We need to
be in the same network environment as MWAA to reach the underlying database.
{% /note %}
### 3. Update MWAA Executor Role policies
- Identify your MWAA executor role. This can be obtained from the details view of your MWAA environment.
- Add the following two policies to the role, the first with ECS permissions:
Note that depending on the kind of workflow you will be deploying, the YAML configuration will need to updated following
the official OpenMetadata docs, and the value of the `pipelineType` configuration will need to hold one of the following values:
-`metadata`
-`usage`
-`lineage`
-`profiler`
-`TestSuite`
Which are based on the `PipelineType` [JSON Schema definitions](https://github.com/open-metadata/OpenMetadata/blob/main/openmetadata-spec/src/main/resources/json/schema/entity/services/ingestionPipelines/ingestionPipeline.json#L14)
Moreover, one of the imports will depend on the MWAA Airflow version you are using:
- If using Airflow <2.5:`from airflow.providers.amazon.aws.operators.ecs import ECSOperator`
- If using Airflow > 2.5: `from airflow.providers.amazon.aws.operators.ecs import EcsRunTaskOperator`
Make sure to update the `ecs_operator_task` task call accordingly.
## Ingestion Workflows as a Python Virtualenv Operator
### PROs
- Installation does not clash with existing libraries
- Simpler than ECS
### CONs
- We need to install an additional plugin in MWAA
- DAGs take longer to run due to needing to set up the virtualenv from scratch for each run.
We need to update the `requirements.txt` file from the MWAA environment to add the following line:
```
virtualenv
```
Then, we need to set up a custom plugin in MWAA. Create a file named virtual_python_plugin.py. Note that you may need to update the python version (eg, python3.7 -> python3.10) depending on what your MWAA environment is running.
```python
"""
Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
Permission is hereby granted, free of charge, to any person obtaining a copy of
this software and associated documentation files (the "Software"), to deal in
the Software without restriction, including without limitation the rights to
This is modified from the [AWS sample](https://docs.aws.amazon.com/mwaa/latest/userguide/samples-virtualenv.html).
Next, create the plugins.zip file and upload it according to [AWS docs](https://docs.aws.amazon.com/mwaa/latest/userguide/configuring-dag-import-plugins.html). You will also need to [disable lazy plugin loading in MWAA](https://docs.aws.amazon.com/mwaa/latest/userguide/samples-virtualenv.html#samples-virtualenv-airflow-config).
A DAG deployed using the PythonVirtualenvOperator would then look like:
```python
from datetime import timedelta
from airflow import DAG
from airflow.operators.python import PythonVirtualenvOperator
from airflow.utils.dates import days_ago
default_args = {
"retries": 3,
"retry_delay": timedelta(seconds=10),
"execution_timeout": timedelta(minutes=60),
}
def metadata_ingestion_workflow():
from metadata.workflow.metadata import MetadataWorkflow
description="An example DAG which runs a OpenMetadata ingestion workflow",
start_date=days_ago(1),
is_paused_upon_creation=False,
catchup=False,
) as dag:
ingest_task = PythonVirtualenvOperator(
task_id="ingest_redshift",
python_callable=metadata_ingestion_workflow,
requirements=['openmetadata-ingestion==1.0.5.0',
'apache-airflow==2.4.3', # note, v2.4.3 is the first version that does not conflict with OpenMetadata's 'tabulate' requirements
'apache-airflow-providers-amazon==6.0.0', # Amazon Airflow provider is necessary for MWAA
'watchtower',],
system_site_packages=False,
dag=dag,
)
```
Where you can update the YAML configuration and workflow classes accordingly. accordingly. Further examples on how to
run the ingestion can be found on the documentation (e.g., [Snowflake](/connectors/database/snowflake)).
You will also need to determine the OpenMetadata ingestion extras and Airflow providers you need. Note that the Openmetadata version needs to match the server version. If we are using the server at 0.12.2, then the ingestion package needs to also be 0.12.2. An example of the extras would look like this `openmetadata-ingestion[mysql,snowflake,s3]==0.12.2.2`.
For Airflow providers, you will want to pull the provider versions from [the matching constraints file](https://raw.githubusercontent.com/apache/airflow/constraints-2.4.3/constraints-3.7.txt). Since this example installs Airflow Providers v2.4.3 on Python 3.7, we use that constraints file.
Also note that the ingestion workflow function must be entirely self-contained as it will run by itself in the virtualenv. Any imports it needs, including the configuration, must exist within the function itself.