Skip to content

LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows

833 B
Raw Blame
"""Entry point: run the PG-queue reaper (leader-elected recovery loop).

Unlike the consumer, the reaper runs no Celery tasks — it only does SQL recovery
(the barrier-orphan sweep) — so it needs **no** worker-app bootstrap: it never
``import``s ``worker``. (``run-worker.sh`` still exports ``WORKER_TYPE`` for every
worker, reaper included, but — unlike the consumer, which overwrites it before
``import worker`` to select which tasks to register — the reaper neither reads
nor mutates it.) This package exists purely so the process has a stable
``python -m pg_queue_reaper`` name that ``run-worker.sh`` can launch and match,
parallel to ``pg_queue_consumer``.

Launch via ``python -m pg_queue_reaper`` or ``./run-worker.sh reaper``.
"""

from queue_backend.pg_queue.reaper import main

if __name__ == "__main__":
    main()