Federate Lakehouse Catalog

Federate Lakehouse Catalog via Cross-cloud Lakehouse

This skill describes how to set up a federated catalog in BigQuery to query remote catalogs like Databricks Unity Catalog or AWS Glue Data Catalog data in AWS over the public internet.

Prerequisites

• For Databricks: Databricks Workspace URL and OAuth Service Principal (Client ID and Secret) with read access.
• For AWS Glue: AWS Administrator access to create IAM roles and permissions policies.
• Active Google Cloud project with administrative access to create lakehouse resources, and secrets in the case of Databricks.

Procedure

Step 1: Information Gathering and Region Selection

Before running any commands, the agent MUST collect the following information from the user:

1. Determine which catalog the user wants to federate to (e.g., Databricks Unity or AWS Glue) and verify it is supported.
2. Determine where the remote data is located (the specific AWS region).
3. Using the Region Pairing Best Practice in the Gotchas section, help the user pick the optimal GCP region to minimize latency.
4. Collect the necessary configuration variables for the chosen flow (e.g., Databricks credentials or AWS Account ID).

Only proceed to the next steps once this information is confirmed.

Step 2: API Verification

Verify that the required Google Cloud APIs are enabled for the project:

gcloud services check biglake.googleapis.com

If the API is not enabled, explicitly ask the user for permission to enable it. Do NOT proceed without their confirmation.

Flow A: Databricks Unity Catalog

1. Create a Regional Secret for Credentials

Store the Databricks client ID and secret in Secret Manager. Ensure the secretmanager.googleapis.com API is enabled. The secret MUST be in the same region as your Lakehouse catalog.

1. Create a JSON file named credentials.json:

{
  "client_id": "<CLIENT_ID>",
  "client_secret": "<CLIENT_SECRET>"
}

1. Set the Secret Manager API endpoint override for the region:

gcloud config set api_endpoint_overrides/secretmanager https://secretmanager.<REGION>.rep.googleapis.com/

1. Create the secret:

gcloud secrets create <SECRET_NAME> \
  --location="<REGION>" \
  --project="<PROJECT_ID>" \
  --data-file=credentials.json

2. Create the Federated Catalog

Create a BigLake Iceberg catalog of type federated pointing to Databricks.

gcloud alpha biglake iceberg catalogs create <CATALOG_NAME> \
   --project="<PROJECT_ID>" \
   --primary-location="<REGION>" \
   --catalog-type="federated" \
   --federated-catalog-type="unity" \
   --secret-name="projects/<PROJECT_ID>/locations/<REGION>/secrets/<SECRET_NAME>" \
   --unity-instance-name="<UNITY_INSTANCE_NAME>" \
   --unity-catalog-name="<UNITY_CATALOG_NAME>" \
   --refresh-interval="300s"

3. Grant Catalog Access to the Secret

Grant the service account created for the catalog access to read the secret.

1. Get the service account email by describing the catalog:

gcloud alpha biglake iceberg catalogs describe <CATALOG_NAME> \
    --project="<PROJECT_ID>" \
    --location="<REGION>" \
    --format="value(biglake-service-account-id)"

1. Grant access:

gcloud secrets add-iam-policy-binding <SECRET_NAME> \
  --project="<PROJECT_ID>" \
  --location="<REGION>" \
  --member="serviceAccount:<SERVICE_ACCOUNT_EMAIL>" \
  --role="roles/secretmanager.secretAccessor"

Flow B: AWS Glue

1. Create the AWS IAM role with a placeholder trust policy

Lakehouse provisions a Google service account ID after catalog creation. Create the AWS IAM role with a placeholder trust policy first.

1. Create a file named trust_policy.json:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Principal": {
        "Federated": "accounts.google.com"
      },
      "Action": "sts:AssumeRoleWithWebIdentity",
      "Condition": {
        "StringEquals": {
          "accounts.google.com:aud": ["PLACEHOLDER_VALUE"],
          "accounts.google.com:sub": ["PLACEHOLDER_VALUE"]
        }
      }
    }
  ]
}

1. Run the AWS CLI command to create the role:

aws iam create-role \
  --role-name <AWS_ROLE_NAME> \
  --assume-role-policy-document file://trust_policy.json \
  --max-session-duration 43200

2. Attach a permissions policy

Attach a policy that allows Lakehouse to read from Glue and S3.

[!IMPORTANT] Safe IAM Scoping: The example below uses wildcard structures for illustration. You MUST consult with the user to scope the Resource ARNs to their specific catalog, database, and S3 buckets. Do NOT blindly apply wildcard permissions.

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "GlueRead",
      "Effect": "Allow",
      "Action": [
        "glue:GetCatalog",
        "glue:GetDatabase",
        "glue:GetDatabases",
        "glue:GetTable",
        "glue:GetTables"
      ],
      "Resource": "arn:aws:glue:<AWS_REGION>:<AWS_ACCOUNT_ID>:catalog"
    },
    {
      "Sid": "S3Read",
      "Effect": "Allow",
      "Action": [
        "s3:ListBucket",
        "s3:GetObject"
      ],
      "Resource": [
        "arn:aws:s3:::<SPECIFIC_BUCKET>",
        "arn:aws:s3:::<SPECIFIC_BUCKET>/*"
      ]
    }
  ]
}

Attach this permissions policy to your IAM role.

3. Create the Federated Catalog

When creating an AWS Glue federated catalog, the --glue-warehouse MUST be set to your 12-digit AWS Account ID string (not an S3 bucket URI). Best Practice: Initialize the catalog without specifying a refresh schedule to prevent premature metadata synchronization failures while AWS trust relationships are propagating.

gcloud alpha biglake iceberg catalogs create <CATALOG_NAME> \
  --project="<PROJECT_ID>" \
  --primary-location="<REGION>" \
  --catalog-type="federated" \
  --federated-catalog-type="glue" \
  --glue-warehouse="<AWS_ACCOUNT_ID>" \
  --glue-aws-region="<AWS_REGION>" \
  --glue-aws-role-arn="arn:aws:iam::<AWS_ACCOUNT_ID>:role/<AWS_ROLE_NAME>"

4. Update the trust policy

Extract the biglake-service-account-id from the created catalog, and update your AWS IAM role's trust policy to replace PLACEHOLDER_VALUE in the aud and sub conditions with this Google Service Agent ID.

5. Enable background refresh

Update the catalog to activate background refresh once the trust policy is updated.

gcloud alpha biglake iceberg catalogs update <CATALOG_NAME> \
  --project="<PROJECT_ID>" \
  --refresh-interval="300s"

Querying the Data

Once set up, you can query the tables via BigQuery.

SELECT * FROM `<PROJECT_ID>.<CATALOG_NAME>.<NAMESPACE>.<TABLE_NAME>` LIMIT 10;

Gotchas and Pitfalls

[!IMPORTANT] Regional Isolation: The Secret Manager secret and the Lakehouse catalog MUST be created in the exact same region.

[!TIP] Region Pairing Best Practice: When setting up the federated catalog, choose GCP regions with "Low Latency Dedicated" or "Partner CCI" to ensure optimal performance when federating large datasets across clouds. Examples of optimal pairings: - AWS us-east-1 (N. Virginia) pairs best with GCP us-east4 (Ashburn, VA) - AWS us-west-2 (Oregon) pairs best with GCP us-west1 (The Dalles, OR) - AWS eu-west-2 (London) pairs best with GCP europe-west2 (London) - AWS eu-central-1 (Frankfurt) pairs best with GCP europe-west3 (Frankfurt) For the exhaustive list of mappings, read the full capabilities table at: https://docs.cloud.google.com/lakehouse/docs/regions-capabilities-cross-cloud-lakehouse

[!IMPORTANT] BigQuery Query Location: When querying the federated catalog via BigQuery, you MUST ensure the query runs in the same region as the catalog (e.g., us-east4). If using the bq CLI, use the --location flag.

Step 3: Validation and Next Steps

After completing the setup, the agent MUST validate that the federation is working and propose next steps to the user.

1. Validate the Connection:

   • Attempt to list the namespaces or tables in the newly federated catalog using the bq CLI or BigQuery API. For example:

     bq ls --location="<REGION>" <PROJECT_ID>.<CATALOG_NAME>
     

   • If the command returns a list of namespaces/schemas, the federation is successful.

2. Troubleshooting:

   • If the validation fails (e.g., permission errors, empty results, timeout), the agent should consult the Cross-Cloud Lakehouse Troubleshooting documentation: https://docs.cloud.google.com/lakehouse/docs/troubleshooting.
   • For AWS Glue, verify that the trust policy correctly references the biglake-service-account-id and that the GCP and AWS regions match your configuration.
   • For Databricks, verify that the secret exists in the correct region and the service account has roles/secretmanager.secretAccessor.

3. Explore and Propose:

   • Assuming the federation is working, browse the available namespaces and a few key tables.
   • Summarize to the user what kind of data was found (e.g., "I see you have tables related to e-commerce transactions and customer profiles").
   • Propose a business or analytical question to the user that would result in a meaningful query of their data (e.g., "Would you like me to write a query to find the top 5 purchasing customers from last month?").