Apache Livy connection

Submit Spark batches over the Apache Livy REST API from a task via a managed Leoflow Connection and Airflow’s LivyHook. The conn_type is livy. A connection carries the host:port plus credentials.

Declare the provider

# leoflow.yaml
dag_id: livy_batch
connectors:
  - livy

Fields the UI asks for

The host:port + user + password round-trip (reserved characters) is pinned by TestLivyConnectionURIShapeIntegration.

Example DAG (copy-paste)

Docs-only recipe (needs a real Livy server). The hook is imported inside the task — a top-level provider import fails the compile (see Installing a connector’s provider).

# dag.py
from __future__ import annotations

from airflow.sdk import DAG, task


@task
def submit() -> None:
    from airflow.providers.apache.livy.hooks.livy import LivyHook

    hook = LivyHook(livy_conn_id="livy_default")
    print("submit: posting batch via LivyHook(livy_default)")
    batch_id = hook.post_batch(file="s3://my-bucket/jobs/etl.py")
    print(f"submit: started batch {batch_id}")


with DAG("livy_batch", schedule=None, catchup=False, tags=["example"]):
    submit()

# leoflow.yaml
schema_version: "1.0"
dag_id: livy_batch
description: Submit a Spark batch via LivyHook.
owner: examples
tags:
  - example
python_version: "3.11"
connectors:
  - livy

Run it

1. Admin → Connections → +, type livy. Set Host, Port, and (if auth is enabled) Login and Password.
2. leoflow lite path/to/this/dag → trigger livy_batch.

Security notes

• Front Livy with auth (Kerberos / a gateway) in production; the credentials are encrypted at rest.
• Never print() the URI — it carries the password.

Related

• docs/connections/index.md — Installing a connector’s provider.
• ADR 0019 / 0021 — secret encryption + agent delivery.