A zero-dependency parallel script runner for Python.

Define units of work (Pi), execute them in parallel with a Runner, or feed them continuously through a Line.

pip install pi-line

from piline import Pi, Runner

pis = [
    Pi(name="train", script="train.py", args=["--epochs", "10"]),
    Pi(name="eval", script="eval.py"),
]

runner = Runner(max_workers=4)
results = runner.run(pis)

for r in results:
    print(f"{r.pi_name}: {'PASS' if r.succeeded else 'FAIL'}")

from piline import Pi, Runner, Line

runner = Runner(max_workers=4)

with Line(runner, max_results=5000) as line:
    line.put(Pi(name="job1", script="job.py"))
    line.put(Pi(name="job2", script="job.py"))

    # look up a result by id
    result = line.get(pi_id)

    # flush results to external storage
    batch = line.drain_results()
# line.stop() called automatically

Use {artefact_dir} and {task_dir} placeholders in args — they get resolved to the real paths before execution:

Pi(
    name="train",
    script="train.py",
    args=["--output", "{artefact_dir}/model.pt"],
)

• Pi — A unit of work. Wraps a script with args, env vars, and an optional timeout. Gets a unique ID on creation. .py files run with Python; anything else runs directly.
• Result — Outcome of running a Pi. Exit code, timing, paths to stdout/stderr logs and artefacts.
• Runner — Executes a batch of Pi's in parallel using ProcessPoolExecutor.
• Line — Thread-safe queue that feeds Pi's to a Runner in batches. Results stored in a capped dict keyed by pi_id. Supports context manager, per-Pi and per-batch callbacks, and drain_results() for periodic flushing.

.piline/runs/<pi_name>/<pi_id>/
    stdout.log
    stderr.log
    artefact/

Scripts receive these env vars automatically:

• PILINE_PI_ID — The Pi's unique ID
• PILINE_PI_NAME — The Pi's name
• PILINE_TASK_DIR — Path to the task directory
• PILINE_ARTEFACT_DIR — Path to the artefact subdirectory

Line(
    runner,
    on_pi_complete=lambda r: print(f"Done: {r.pi_name} ({'PASS' if r.succeeded else 'FAIL'})"),
    on_batch_complete=lambda results: print(f"Batch of {len(results)} finished"),
)

See docs/index.md.