Pattern: Asset-to-Task

Use this when a cached value feeds an action: a deploy, an upload, a notification. The asset is computed when its code or inputs change. The task runs every time you ask for it.

Example

from barca import asset, task


@asset()
def model() -> dict:
    return {"version": 3, "auc": 0.91}


@task(inputs={"m": model})
def deploy(m: dict) -> dict:
    print(f"deploying model v{m['version']}")
    return {"deployed": m["version"]}

A task is run with barca run, target first, then the file:

barca run deploy pipeline.py

What barca does

First run: two steps, model and then deploy. The task’s print goes to stderr, prefixed with the step’s id. In a terminal run prints the task’s value under the summary.

[pipeline.py:deploy] deploying model v3
[barca] 2/2 steps | done in 0.0s
Run 150dcb149a6c | ran 'deploy' in 0.064s (2 steps, 1 phase)

Value:
{
  "deployed": 3
}

Second run: one step. model is served from cache and deploy runs again.

[pipeline.py:deploy] deploying model v3
[barca] 1/2 steps | done in 0.0s
Run 150f3857378c | ran 'deploy' in 0.061s (1 step, 1 phase)

Value:
{
  "deployed": 3
}

--dry-run shows the same decision without running anything:

barca run deploy pipeline.py --dry-run

Dry run: barca run deploy (nothing executed, nothing written)

STATUS    WHY                  STEP
cached    -                    pipeline.py:model
will run  tasks always re-run  pipeline.py:deploy

1 will run, 1 cached, 0 unknown

To recompute the asset as well, name it: barca run deploy pipeline.py --refresh model (two steps again). --refresh-all recomputes every upstream asset.

Limits

get is for assets and run is for tasks. barca get deploy pipeline.py exits 2 with 'deploy' is a task — use `barca run` instead, and barca run model pipeline.py exits 2 with 'model' is an asset — use `barca get` instead.
barca get pipeline.py with no target never runs a task. It materializes the assets and sensors and prints the tasks it skipped on stderr:
```
[barca] skipped 1 task (deploy): `barca get` without a target materializes assets only. Run a task with: barca run deploy pipeline.py
```
A task cannot be an input to an asset or a sensor. A task’s result is never cached, so barca rejects the edge when it builds the graph (exit 2):
```
DAG error: task 'bad.py:migrate_db' cannot be an input to asset 'bad.py:user_counts' (tasks are never cached, so this would poison caching)
```
A task may be an input to another task; see Ordering-Only Dependencies.
A task that fails is not retried unless you ask. See Error Handling.

The same material is in the terminal manual: barca docs tasks.