deployment

No menu items for this category

This page is about running the Ingestion Framework externally!

There are mainly 2 ways of running the ingestion:

  1. Internally, by managing the workflows from OpenMetadata.
  2. Externally, by using any other tool capable of running Python code.

If you are looking for how to manage the ingestion process from OpenMetadata, you can follow this doc.

Run the ingestion from AWS MWAA

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.

  • It is the simplest approach
  • We don’t need to spin up any further infrastructure
  • We need to install the 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:

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

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).

We have different classes for different types of workflows. The logic is always the same, but you will need to change your import path. The rest of the method calls will remain the same.

For example, for the Metadata workflow we'll use:

The classes for each workflow type are:

  • Metadata: from metadata.workflow.metadata import MetadataWorkflow
  • Lineage: from metadata.workflow.metadata import MetadataWorkflow (same as metadata)
  • Usage: from metadata.workflow.usage import UsageWorkflow
  • dbt: from metadata.workflow.metadata import MetadataWorkflow
  • Profiler: from metadata.workflow.profiler import ProfilerWorkflow
  • Data Quality: from metadata.workflow.data_quality import TestSuiteWorkflow
  • Data Insights: from metadata.workflow.data_insight import DataInsightWorkflow
  • Elasticsearch Reindex: from metadata.workflow.metadata import MetadataWorkflow (same as metadata)
  • Completely isolated environment
  • Easy to update each version
  • 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.

  • 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:

We'll need to use the /ecs/openmetadata below when configuring the policies.

  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:

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.

  • 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:

And for the Log Group permissions

Note how you need to replace the region, account-id and the log group names for your Airflow Environment and ECS.

A DAG created using the ECS Operator will then look like this:

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

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.

  • Installation does not clash with existing libraries
  • Simpler than ECS
  • 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:

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.

This is modified from the AWS sample.

Next, create the plugins.zip file and upload it according to AWS docs. You will also need to disable lazy plugin loading in MWAA.

A DAG deployed using the PythonVirtualenvOperator would then look like:

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).

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. 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.

We have different classes for different types of workflows. The logic is always the same, but you will need to change your import path. The rest of the method calls will remain the same.

For example, for the Metadata workflow we'll use:

The classes for each workflow type are:

  • Metadata: from metadata.workflow.metadata import MetadataWorkflow
  • Lineage: from metadata.workflow.metadata import MetadataWorkflow (same as metadata)
  • Usage: from metadata.workflow.usage import UsageWorkflow
  • dbt: from metadata.workflow.metadata import MetadataWorkflow
  • Profiler: from metadata.workflow.profiler import ProfilerWorkflow
  • Data Quality: from metadata.workflow.data_quality import TestSuiteWorkflow
  • Data Insights: from metadata.workflow.data_insight import DataInsightWorkflow
  • Elasticsearch Reindex: from metadata.workflow.metadata import MetadataWorkflow (same as metadata)