Airflow Reservations Policy

Airflow Cluster Policy plugin for BigQuery reservation management.

This package integrates with Airflow's Cluster Policies to automatically inject BigQuery reservation assignments into your tasks without requiring any changes to your DAG code.

Features

• Automatic re-assignment - intercepts BigQuery operators and re-assigns them to appropriate reservation based on the configuration:
  • BigQueryInsertJobOperator - Injects into configuration.reservation field
  • BigQueryExecuteQueryOperator - Injects into api_resource_configs.reservation field (supported in provider package 2.0.0 - 10.26.0)
• Lookup-based Configuration - Uses dag_id.task_id → reservation mappings
• Python API - Provides get_reservation() for custom BigQuery API calls in PythonOperator
• Performance Optimized - Config caching with file mtime-based invalidation
• Graceful Error Handling - Won't crash Airflow on config errors

Installation

Add to your requirements.txt:

airflow-reservations=X.Y.Z

Configuration

Generating Configuration

Use Masthead recommendations to generate the reservations_config.json file containing the optimal reservation assignments for your tasks. Users are responsible for pulling this configuration into their Airflow environment.

Typical workflow:

1. Masthead analyzes your BigQuery workloads
2. Read Masthead recommendations and generate reservations_config.json with optimal assignments
3. Merge the config into your DAGs repository
4. Airflow syncs the updated config file
5. The policy applies reservations on next task parse

Configuration Fields

Reservation values:

Applying Configuration

Create a reservations_config.json file in your DAGs folder:

{
  "reservation_config": [
    {
      "tag": "standard",
      "reservation": "projects/{project}/locations/{location}/reservations/{name}",
      "tasks": [
        "finance_dag.daily_report",
        "etl_dag.load_analytics",
        "etl_dag.etl_group-transform_data"
      ]
    },
    {
      "tag": "on_demand",
      "reservation": "none",
      "tasks": [
        "adhoc_dag.quick_query"
      ]
    },
    {
      "tag": "default",
      "reservation": null,
      "tasks": [
        "marketing_dag.calculate_roas"
      ]
    }
  ]
}

By default, the config file is loaded from $AIRFLOW_HOME/dags/reservations_config.json.

Override the path using the RESERVATIONS_CONFIG_PATH environment variable:

export RESERVATIONS_CONFIG_PATH=/custom/path/to/config.json

How It Works

The policy is automatically registered via Airflow's plugin entrypoint system. When Airflow parses your DAGs, this plugin's task_policy hook is called for each task. For BigQuery tasks, it:

1. Extracts dag_id and task_id from the task
2. Looks up dag_id.task_id in the configuration file
3. If found, assigns the task to a reservation

Prerequisite: BigQuery Job Labels

Masthead recommendations depend on BigQuery job labels that identify the Airflow task. Ensure every BigQuery job has:

• airflow-dag
• airflow-task

Airflow BigQuery operators set these labels by default (Airflow v2.9+), but custom job configuration can override defaults and drop them. When customizing operator configs, merge labels and keep both keys.

For jobs submitted in PythonOperator (or any manual BigQuery client usage), set labels explicitly (see Usage in Python Operators below).

Note: This package applies reservation assignment based on Airflow task identity, but it does not enforce BigQuery labels. Keep label handling in your DAG/operator/client code.

Usage in Python Operators

For custom BigQuery API calls in PythonOperator, use the provided get_reservation function, e.g.:

from airflow_reservations import get_reservation

def my_bigquery_task(**context):
    ti = context['task_instance']

    existing_labels = {"team": "finance"}
    labels = {
      **existing_labels,
      "airflow-dag": ti.dag_id.lower(),
      "airflow-task": ti.task_id.lower().replace(".", "-"), # dots are not allowed in labels
    }

    # Look up reservation for this task
    reservation = get_reservation(ti.dag_id, ti.task_id)

    # When executing, ensure labels and the reservation are included in the job config
    job_config = bigquery.QueryJobConfig(
      labels=labels,
      reservation=reservation if reservation else None
    )

    # Execute with BigQuery client...
    client.query(sql, job_config=job_config)

Troubleshooting

Config not loading

Check that:

1. The config file exists at the expected path
2. The file contains valid JSON
3. Airflow has read permissions

Enable debug logging:

import logging
logging.getLogger("airflow_reservations").setLevel(logging.DEBUG)

Reservations not being applied

Verify:

1. The task type is BigQueryInsertJobOperator or BigQueryExecuteQueryOperator
2. BigQuery jobs include airflow-dag and airflow-task labels (see Prerequisite: BigQuery Job Labels above)
3. The dag_id.task_id key exactly matches the config
4. For TaskGroups, the full path is included (e.g., dag.group.task), and in the config JSON the second dot is replaced with a dash (e.g., dag.group-task).

Supported Versions

This package is tested and compatible with:

• Airflow 2.6+ (including 2.10.x) - ✅ Fully supported and tested
• Airflow 3.x (including 3.1.x) - ✅ Fully supported and tested (see Airflow 3 notes below)
• Python 3.8+ (tested with 3.11 and 3.12)

Airflow 3 Notes

In Airflow 3, the Task SDK parses DAGs in a separate process/container from the Scheduler and Worker. To ensure the policy works correctly:

1. Installation: The airflow-reservations package must be installed in the environment where the Task SDK executes (typically your custom Airflow image).
2. Config Accessibility: The reservations_config.json file must be accessible to the Task SDK process. If you are using remote DAG storage, ensure the config file is bundled with your DAGs or placed in a shared volume.
3. Environment Variables: If you use RESERVATIONS_CONFIG_PATH, it must be set in the environment of the worker/execution container as well.