Rabbit BigQuery Job Optimizer Airflow Plugin

This Airflow plugin automatically optimizes BigQuery job configurations using the Rabbit API. It patches the BigQueryHook to intercept job configurations and optimize them before execution.

Features

• Automatically intercepts all BigQuery job submissions
• Optimizes job configurations via the Rabbit API
• Routes BigQueryInsertJobOperator submits to an on-demand pool billing project when the optimizer sets optimizedJob.jobReference.projectId
• Graceful fallback to original configuration if optimization fails
• Comprehensive error handling and logging

Already running an older version? See Updating.

Installation

Option A: PyPI (recommended)

1. Add the plugin and client to your Airflow environment:

   • If using requirements.txt:

     rabbit-bq-job-optimizer==0.1.18
     rabbit-bq-optimizer-airflow-plugin==1.1.2
     

   • If using constraints.txt:

     rabbit-bq-job-optimizer==0.1.18
     rabbit-bq-optimizer-airflow-plugin==1.1.2
     

   • If using a custom Docker image, add to your Dockerfile:

     RUN pip install rabbit-bq-job-optimizer==0.1.18 rabbit-bq-optimizer-airflow-plugin==1.1.2
     

   The plugin registers via Airflow's plugin entry point — no file copy into plugins/ is required.

2. Restart your Airflow scheduler, workers, and webserver to load the plugin. If you use deferrable BigQueryInsertJobOperator, restart the triggerer as well.

Option B: Copy into plugins/

1. Add the Python client to your Airflow environment dependencies:

   rabbit-bq-job-optimizer==0.1.18
   

2. Copy the plugin file into your Airflow plugins directory:

   cp bq-job-optimizer-airflow-2/rabbit_bq_optimizer_plugin.py "$AIRFLOW_HOME/plugins/"
   

3. Restart your Airflow scheduler, workers, and webserver (and triggerer if using deferrable operators).

Updating

PyPI install

Update to the new package versions in your environment dependencies:

rabbit-bq-job-optimizer==0.1.18
rabbit-bq-optimizer-airflow-plugin==1.1.2

Apply the update using your platform's usual process, then restart Airflow components (including the triggerer if you use deferrable BigQueryInsertJobOperator). Your rabbit_api connection, rabbit_bq_optimizer_config variable, and DAGs are unchanged. If optimization fails, the plugin still submits the original job (fail-open).

Copy into plugins/ (optional)

If you installed via Option B and want to stay on the copy method, bump the client version in your environment dependencies and replace rabbit_bq_optimizer_plugin.py in your plugins location with the new file from this repo. Then restart Airflow components as above.

Switching from copy to PyPI

If you previously deployed rabbit_bq_optimizer_plugin.py to your plugins location and are switching to the PyPI package, remove that file — the PyPI install loads via an entry point and the copied file causes duplicate plugin registration. On startup the plugin logs an error if both are present. When the file is writable, you can set RABBIT_BQ_OPTIMIZER_REMOVE_LEGACY_PLUGIN=true to remove it automatically on load.

For pool billing routing, the Airflow service account also needs roles/bigquery.jobUser on the pool billing projects — contact Rabbit support for your project list. See Pool billing project routing.

Configuration

1. Create the Rabbit API connection

For security reasons the Rabbit API key must be stored in an Airflow connection instead of a variable. The plugin expects a connection with ID rabbit_api:

• Conn Type: Generic (or HTTP)
• Conn ID: rabbit_api
• Password: Rabbit API key (required)
• Extra (optional):

  {
    "api_base_url": "https://api.followrabbit.ai/bq-job-optimizer"
  }
  

  Only include api_base_url if you need to override the default Rabbit endpoint; otherwise it will fall back automatically.

CLI example:

airflow connections add rabbit_api \
    --conn-type generic \
    --conn-password "<your-rabbit-api-key>" \
    --conn-extra '{"api_base_url": "https://api.followrabbit.ai/bq-job-optimizer"}'

2. Set the optimization parameters

The remaining optimizer configuration stays in an Airflow variable. Create a JSON configuration with the following structure:

{
    "enabled": true,
    "default_pricing_mode": "on_demand",
    "reservation_ids": [
        "project:region.reservation-name1",
        "project:region.reservation-name2"
    ]
}

Configuration Fields

• enabled (required): Must be true to run optimization. If omitted, false, or the variable is not set, each BigQuery submit bypasses optimization (no API call). Takes effect on the next job without restarting Airflow.
• default_pricing_mode (required when enabled is true): The default pricing mode for jobs. Must be one of: "on_demand" or "slot_based"
• reservation_ids (required when default_pricing_mode is "on_demand"): List of reservation IDs in the format "project:region.reservation-name". On-demand routing needs at least one reservation to compare against; missing or empty fails validation and the job submits unoptimized. When default_pricing_mode is "slot_based", this field is optional (defaults to []).
• dag_whitelist (optional): List of DAG IDs to optimize. When omitted (or null), all DAGs are optimized. When set to a list, only BigQuery jobs from matching DAGs are optimized; all others pass through unmodified — and an empty list ([]) means nothing is whitelisted, so no DAG is optimized. Must be a JSON list; any other type is ignored and optimization is skipped as a safe fallback.
• debug (optional): Turns on detailed error logs for troubleshooting. When the optimizer cannot change a job and falls back to the original BigQuery submit, Airflow normally shows only a short warning. Set this to true to also include the full Python error traceback in the task log — helpful when working with Rabbit support. Leave it unset or false in normal use. Default: off.

Setting the Configuration

You can set the configuration in two ways:

1. Using the Airflow CLI:

airflow variables set rabbit_bq_optimizer_config '{
    "enabled": true,
    "default_pricing_mode": "on_demand",
    "reservation_ids": [
        "project:region.reservation-name1",
        "project:region.reservation-name2"
    ],
    "dag_whitelist": [
        "my_etl_dag",
        "my_analytics_dag"
    ]
}'

Omit dag_whitelist to optimize all DAGs. Setting it to an empty list ([]) whitelists nothing and disables optimization for every DAG.

2. Through the Airflow UI:
   • Go to Admin -> Variables
   • Add a new variable named rabbit_bq_optimizer_config
   • Paste the JSON configuration

End-to-End Test Plan

Local Testing

A setup script setup_local_test.sh is provided in the root directory to help you spin up a local Airflow environment for testing.

Prerequisites:

• Python 3.8-3.12 (Airflow 2.9.1 requirement; the script will automatically detect and use a compatible version)
• pip

Steps:

1. Run the setup script:

   ./setup_local_test.sh
   

   This script will:
   • Create a virtual environment (if not present)

• Install Airflow using the official constraints file plus dependencies from test_requirements.txt
• Initialize the Airflow database
• Create an admin user
• Deploy the plugin and test DAG
• Configure the rabbit_api connection (with a dummy key)
• Configure the rabbit_bq_optimizer_config variable

Note: The script installs Airflow using the official constraints file for your Python version so that binary wheels (including google-re2) are used. Additional dependencies (rabbit-bq-job-optimizer, apache-airflow-providers-google, google-cloud-bigquery, etc.) are installed with the same constraints to match the Airflow release.

2. Run the test DAG: The setup script automatically deploys test_dag.py to the DAGs directory.

   Execute the test DAG:

   source venv/bin/activate
   export AIRFLOW_HOME=$(pwd)/airflow_home
   airflow dags test rabbit_optimizer_test $(date +%Y-%m-%d)
   

3. Verify Results: Check the logs for the following:

   • Rabbit BQ Optimizer: patching BigQueryHook.insert_job (scheduler logs at startup)
   • Rabbit BQ Optimizer: patching BigQueryInsertJobOperator._submit_job (scheduler logs at startup)
   • Rabbit BQ Optimizer: optimization result=... at DEBUG on job submit (or failure if using dummy key/URL)

4. Run automated tests: After setup, you can run automated tests to validate the plugin:

   Test connection loading:

   source venv/bin/activate
   export AIRFLOW_HOME=$(pwd)/airflow_home
   cd bq-job-optimizer-airflow-2
   python test_plugin_connection.py
   

   This verifies that the plugin can load credentials from the Airflow connection.

   Test DAG execution (full integration):

   source venv/bin/activate
   export AIRFLOW_HOME=$(pwd)/airflow_home
   cd bq-job-optimizer-airflow-2
   python test_dag_execution.py
   

   This simulates a DAG execution and verifies that:

   • The plugin intercepts BigQuery jobs
   • The Rabbit optimization API is called with connection credentials
   • The optimized configuration is used

   Test DAG whitelist filtering:

   source venv/bin/activate
   export AIRFLOW_HOME=$(pwd)/airflow_home
   cd bq-job-optimizer-airflow-2
   python test_dag_whitelist.py
   

   This verifies the optional dag_whitelist feature:

   • Optimization runs for all DAGs when whitelist is absent or empty
   • Only whitelisted DAGs are optimized when the list is set
   • Safe fallback (skips optimization) when whitelist is set but Airflow context is unavailable

   Test plugin optimization (unit, no real BQ):

   source venv/bin/activate
   export AIRFLOW_HOME=$(pwd)/airflow_home
   python -m unittest bq-job-optimizer-airflow-2/test_plugin_optimization.py -v
   

5. Cleanup (optional): To remove the test environment (venv and airflow_home directories):

   ./setup_local_test.sh --clean
   

   Or use the short form:

   ./setup_local_test.sh -c
   

Manual Validation Steps

1. Connection setup

   • Create the rabbit_api connection as described above.
   • Verify the password is populated and (optionally) the base URL override is correct by running airflow connections get rabbit_api.

2. Variable setup

   • Set the rabbit_bq_optimizer_config variable using the sample JSON (without the API key).
   • Confirm default_pricing_mode is valid. For "on_demand", include at least one reservation ID.

3. DAG validation

   • Deploy the plugin file to $AIRFLOW_HOME/plugins.
   • Restart the scheduler/webserver/workers so the patch is applied.
   • Trigger a DAG that uses BigQueryInsertJobOperator (or any BigQuery operator).

4. Positive path (connection + variable present)

   • Confirm in the scheduler logs at startup that the plugin logs patching BigQueryHook.insert_job and patching BigQueryInsertJobOperator._submit_job.
   • Validate that the BigQuery job request reaches Rabbit (check Rabbit logs or enable DEBUG for Rabbit BQ Optimizer: optimization result=...).
   • For a query with historical data/reservations, verify the resulting BigQuery job uses the optimized configuration.

5. Missing connection safety net

   • Temporarily rename/delete the Airflow connection and re-run the DAG.
   • The scheduler logs should show Failed to load Rabbit API connection and the job should proceed without optimization (ensures graceful fallback).

6. Custom base URL regression

   • Set api_base_url in the connection extras to a staging endpoint (or use schema/host).
   • Trigger a DAG and confirm via Rabbit logs/network traces that the request hits the custom endpoint.

7. Compatibility check

   • Ensure the environment installs the latest rabbit-bq-job-optimizer package (e.g., pip install -U rabbit-bq-job-optimizer).
   • Run unit tests or a smoke DAG run to confirm no API contract regressions with the service.

Usage

The plugin works automatically with any BigQuery operator. No additional configuration is needed in your DAGs:

from airflow.providers.google.cloud.operators.bigquery import BigQueryInsertJobOperator

# The plugin will automatically optimize this job
insert_job = BigQueryInsertJobOperator(
    task_id="insert_query_job",
    configuration={
        "query": {
            "query": "SELECT * FROM `project.dataset.table`",
            "useLegacySql": False
        }
    }
)

Pool billing project routing

When the optimizer returns a pool billing project in optimizedJob.jobReference.projectId, jobs submitted through BigQueryInsertJobOperator run and poll on that project. The plugin briefly links the operator to the hook during _submit_job so operator.project_id and the BigQuery jobs.insert call use the same billing project.

Direct BigQueryHook.insert_job calls still receive the optimized configuration but keep the original project_id. Use BigQueryInsertJobOperator in DAGs when you need pool billing project routing with deferrable tasks.

If the optimizer response has no jobReference.projectId (for example reservation-only assignment), the operator keeps its configured source project_id.

Job labels

Optimized jobs include labels you can query in BigQuery (INFORMATION_SCHEMA.JOBS_BY_*) to confirm how each submit was handled:

Label	Values	Meaning
rabbit-source-project	GCP project id	Project the job was submitted from (your DAG / operator project_id)
rabbit-pool-project	GCP project id	Pool billing project recommended by the optimizer (present only when a pool was assigned)
rabbit-pool-routing	applied | skipped | none	applied: BigQueryInsertJobOperator submitted on rabbit-pool-project; skipped: pool recommended but submit stayed on the source project (typically a direct hook call); none: no pool billing project in the optimizer response

Error Handling

The plugin includes comprehensive error handling for:

• Missing or invalid configuration
• Missing required fields
• Empty reservation IDs list when default_pricing_mode is "on_demand"
• API errors

In all error cases, the plugin will log a warning and proceed with the original job configuration.

Troubleshooting

Broken plugin: `No module named 'rabbit_bq_job_optimizer'` — install `rabbit-bq-optimizer-airflow-plugin` (or `rabbit-bq-job-optimizer`) in the Airflow environment and restart.

If you see this error, it means the rabbit-bq-job-optimizer package is missing from your Airflow environment. Install it by adding the package (https://pypi.org/project/rabbit-bq-job-optimizer/) to your Airflow requirements or environment.

[2025-06-15, 17:25:20 UTC] {rabbit_bq_optimizer_plugin.py} WARNING - Rabbit BQ Optimizer: config error: rabbit_bq_optimizer_config not set or invalid. Using original job.

If you see this on job submit, the Airflow variable rabbit_bq_optimizer_config is not set or has invalid content. Make sure you have:

1. Created the variable named exactly rabbit_bq_optimizer_config
2. Set valid JSON content that includes all required fields:

   {
       "enabled": true,
       "default_pricing_mode": "on_demand",
       "reservation_ids": [
           "project:us-central1.reservation-name1",
           "project:us-east1.reservation-name2"
       ]
   }
   

3. Verified the JSON is properly formatted without any syntax errors

Development

Pre-commit Hooks

This repository uses pre-commit hooks to ensure code quality. To set them up:

# Install pre-commit
pip install pre-commit

# Install the hooks
pre-commit install

# (Optional) Run hooks on all files
pre-commit run --all-files

The hooks will automatically:

• Auto-fix formatting with Black (CI checks formatting)
• Auto-fix linting issues with Ruff (CI checks linting)
• Check for common issues (trailing whitespace, large files, etc.)

Note: Pre-commit hooks auto-fix issues for convenience, while CI runs in check-only mode to ensure nothing slips through.

CI/CD

GitHub Actions automatically runs on every push and pull request:

• Linting and Formatting: Checks code style with Black and Ruff
• Tests: Runs test suite on Python 3.9, 3.10, 3.11, and 3.12

Uninstalling

1. Remove the plugin file from your Airflow plugins directory
2. Restart Airflow

Support

For questions, API endpoint details, API key, or any issues, contact the Rabbit support team at success@followrabbit.ai