Pattern: Ordering-Only Dependencies

Use this when one step must run after another but does not need its result: seed a database after migrating it, read an external table after the step that wrote it. Start the input’s name with _, both the key in inputs= and the parameter.

Example

from barca import task


@task()
def migrate_db():
    run_migrations()


@task(inputs={"_migrate": migrate_db})
def seed_data(_migrate):
    insert_seed_records()
    return {"seeded": 2, "received": repr(_migrate)}

run_migrations() and insert_seed_records() stand for your own code. In the run below they print migrating and seeding.

barca run seed_data pipeline.py

What barca does

migrate_db runs first, then seed_data. The _migrate parameter receives None: the upstream result is not read from disk.

[pipeline.py:migrate_db] migrating
[pipeline.py:seed_data] seeding
[barca] 2/2 steps | done in 0.0s
Run 150a5b2362ad | ran 'seed_data' in 0.065s (2 steps, 1 phase)

Value:
{
  "received": "None",
  "seeded": 2
}

The same works between assets (@asset(inputs={"_events": events})). The edge is a real dependency: the downstream asset runs after the upstream and runs again when the upstream changes. Only removing the input removes the dependency.

An unused input without the prefix is loaded, and reported

An input whose name does not start with _ is loaded and passed whether or not the function uses it. At plan time barca reads each function body, and when a step never mentions an input (or only dels it), barca plan, barca get, barca run and --dry-run print one line on stderr and add an entry to the warnings array of their JSON output:

@task(inputs={"migrate": migrate_db})
def seed_data(migrate):      # never used
    insert_seed_records()

[barca] warning: unused.py:seed_data never uses its input `migrate`. It is still loaded in full each time the step runs, and it counts toward the step's cache key. Use it, remove it from inputs=, or rename the parameter `_migrate` if it is there for ordering only (a `_` input is not loaded and never flagged)

A _-prefixed input is never reported. There is no flag or configuration key that turns the warning off. A name that appears inside a string in the body counts as used (SQL such as duckdb.sql("select * from orders"), a pandas query("x > @limit")). The exact rule, and what is never reported (stubs, **kwargs, locals(), queries built outside the body, DuckDB relation inputs, sensor inputs), is in barca docs assets, “Unused inputs”.

Limits

• The function must still accept the parameter. Barca passes _migrate=None. Without the parameter every command that reads the file exits 2, before anything runs:

  Parse error: pipeline.py:seed_data (line 9): the inputs= key `_migrate` is not a parameter: seed_data() takes no parameters.
  Add a parameter `_migrate` to seed_data(), or remove the key. Barca passes each inputs= key to the function as a keyword argument. See `barca docs assets`.

• There is no after= keyword. Early prototypes had one. @task(after=[migrate_db]) is a usage error (exit 2) for every command that reads the file:

  Parse error: after.py:seed_data (line 9): `after` is not an argument of @task. @task accepts: name, inputs, partitions, serializer, freshness, timeout_seconds, retries, retry_backoff, description, tags, env
  Remove `after`, or replace it with an argument @task accepts. See `barca docs tasks`.

  Use inputs={"_name": upstream}.

• A task cannot be an ordering-only input to an asset, for the same reason it cannot be a data input: Asset-to-Task.

• A linter that reports unused parameters may flag _migrate. Most linters can be configured to allow parameters that start with _.