Asgarde

Error handling and dead letter queues for Apache Beam Python, without the boilerplate. Also available for Java and Kotlin with Asgarde Java.

📖 Documentation: https://tosun-si.github.io/asgarde/

Why Asgarde

With plain Beam, each step needs its own DoFn with a try/except block and tagged outputs, and all the failures must be flattened at the end. Asgarde keeps a fluent flow and gathers the failures of all the steps:

from asgarde import CollectionComposer

result = (CollectionComposer.of(values)
          .map('Trim', lambda value: value.strip())
          .map('Parse', int)
          .filter('Keep even numbers', lambda number: number % 2 == 0))

outputs = result.outputs    # Output of the last step
failures = result.failures  # The failures of all the steps, for your dead letter queue

Installation

Asgarde is published on PyPI. apache-beam is a minimum version, not a pin: your pipeline brings its own Beam version.

uv add asgarde      # or: pip install asgarde

Features

• One place for all the errors: each step catches its errors in a Failure (with the stack trace), the CollectionComposer gathers the failures of all the steps.
• map, flat_map and filter operators with side inputs and DoFn lifecycle actions, and Failure.from_exception for your own DoFn.
• Origin element: with with_origin_element, the failures also give the element that entered the flow, to debug and replay from the start. Evaluated only when a failure occurs.
• Never breaks your job: non picklable exceptions, non JSON dicts, partial flat_map outputs, deterministic labels for Dataflow updates.
• Input element format and encoded elements: choose the string format of the input element (with_input_element_to_string), and optionally keep the elements encoded with their coder (with_encoded_elements), to replay them exactly.
• Any Beam transform: apply(name, beam.Map(...)) or a custom DoFn, with the Beam native exception handling, its errors gathered with the other failures.
• Write the failures to BigQuery with Failure.to_dict() and FAILURE_BIGQUERY_SCHEMA (exception type, message, stack trace, timestamp...).
• Beam native error handling: the failures can be added to a Beam ErrorHandler with Failure.to_bad_record, for a single dead letter queue.
• Failure metrics: a Beam counter per step.
• Typed (py.typed).

Compatibility

Python 3.10 to 3.14, apache-beam>=2.60.0. The CI tests the minimum Beam version and the latest Beam release on every push and every week. See Compatibility.

Roadmap

See the roadmap.

Contributing

Contributions are welcome, see CONTRIBUTING.md.

License

MIT