taskferry-celery

A Celery task backend for Taskferry. Enqueue Taskferry tasks onto a Celery app you already run; Taskferry supplies no worker and reimplements none of Celery.

pip install taskferry-celery[celery]

Submitting

from taskferry import Taskferry

runtime = Taskferry.from_mapping(
    {
        "backends": {"cel": {"factory": "celery", "app": "myapp.celery:app"}},
        "defaults": {"task": "cel"},
    }
)
runtime.tasks.submit("myapp.tasks:refresh", 42, queue="metadata")

Executing (worker)

Register the dispatcher once, then run Celery's own worker:

from celery import Celery
from taskferry_celery import register_dispatcher
from taskferry.functions import FunctionRegistry

app = Celery("myapp", broker="redis://localhost:6379/0")
register_dispatcher(app, registry=FunctionRegistry(allowed_modules=["myapp"]))

celery -A myapp worker

Capabilities (honest)

Retries are the engine's: the dispatcher raises Celery's own self.retry, so a failed task is genuinely re-queued with the requested countdown. Taskferry never loops in-process pretending to retry.

state maps Celery's PENDING/STARTED/SUCCESS/FAILURE/RETRY/REVOKED onto the portable states. Note Celery reports PENDING for both a queued task and an unknown id; without a result backend the two are indistinguishable, and this adapter says so rather than guessing.

Only JSON-safe arguments and a "module:function" name cross the wire — never a pickled callable.

License

Apache-2.0.