OpenRAG is a comprehensive, single package Retrieval-Augmented Generation platform built on Langflow, Docling, and Opensearch.
Python
"""OpenRAG backend entry point.
Most of what used to live in this file moved into focused modules:
- app/factory.py — FastAPI app construction
- app/container.py — service DI container
- app/lifespan.py — startup + shutdown lifecycle
- app/middleware.py — request logging
- app/routes/ — route registrars (internal, public_v1, mcp)
- services/default_docs_service.py — bundled-docs onboarding/refresh
- services/startup_orchestrator.py — per-startup tasks (OpenSearch, flows, MCP URLs)
- utils/opensearch_init.py — OpenSearch index/security setup
- utils/jwt_keygen.py — RSA keypair generation
- utils/url_content_fetcher.py — URL → text-file helper
- observability/instana_boot.py — Instana APM gate (run from bootstrap)
- api/health.py — liveness + readiness probes
This module is intentionally thin: import bootstrap, run Alembic, build
the app, run uvicorn. The re-exports below preserve the
`from main import …` contract used by tests and api/settings.py.
"""
# Importing bootstrap also boots the Instana tracer when INSTANA_ENABLED is
# set — it has to patch httpx / sqlalchemy / starlette before the imports
# below pull them in.
import bootstrap # noqa: F401 — must be first; loads .env + structured logging
import asyncio
import atexit
# Re-exported so tests that monkeypatch "main.httpx.AsyncClient" keep working.
# Module attributes in Python are shared singletons, so the patch propagates
# to services/default_docs_service.py which calls httpx.AsyncClient directly.
import httpx # noqa: F401
from app.factory import create_app
from config.settings import ACCESS_LOG_ENABLED, OPENRAG_BACKEND_PORT
from services.default_docs_service import (
_get_remote_docs_signature,
_should_use_url_default_docs_ingest,
ingest_default_documents_when_ready,
refresh_default_openrag_docs,
)
from services.startup_orchestrator import startup_tasks
from utils.encryption import enforce_startup_prerequisites
from utils.jwt_keygen import generate_jwt_keys
from utils.logging_config import get_logger
from utils.opensearch_init import _ensure_opensearch_index, init_index
enforce_startup_prerequisites()
logger = get_logger(__name__)
__all__ = [
"create_app",
"startup_tasks",
"generate_jwt_keys",
"init_index",
"_ensure_opensearch_index",
"ingest_default_documents_when_ready",
"refresh_default_openrag_docs",
"_get_remote_docs_signature",
"_should_use_url_default_docs_ingest",
]
def cleanup():
"""Cleanup on application shutdown (atexit hook)."""
logger.info("Application shutting down")
if __name__ == "__main__":
import uvicorn
atexit.register(cleanup)
# Run Alembic upgrade SYNCHRONOUSLY before the app builds. This
# avoids two pitfalls:
# 1. Alembic's env.py uses asyncio.run() which collides with a
# live event loop.
# 2. Anything done inside `asyncio.run(create_app())` binds to a
# loop that is closed before uvicorn starts — including DB
# engines. Putting the schema migration here keeps the runtime
# `init_engine()` deferred to the lifespan startup, on the
# live uvicorn loop.
try:
from db.migrations_runtime import run_alembic_upgrade
run_alembic_upgrade("head")
except Exception as _e:
logger.error("Alembic upgrade failed at startup", error=str(_e))
raise
app = asyncio.run(create_app())
# Optionally spin up the standalone ingestion-callback proxy router in this
# same process (own daemon thread + port) so Langflow calls back to it
# instead of the full backend internal API surface.
from config.settings import OPENRAG_BACKEND_ROUTER_ENABLE, OPENRAG_BACKEND_ROUTER_PORT
if OPENRAG_BACKEND_ROUTER_ENABLE:
from app.router_app import start_backend_router
start_backend_router()
logger.info("Backend ingestion router enabled", port=OPENRAG_BACKEND_ROUTER_PORT)
uvicorn.run(
app,
workers=1,
host="0.0.0.0",
port=OPENRAG_BACKEND_PORT,
reload=False,
access_log=ACCESS_LOG_ENABLED,
log_config=None,
)