Spaces:
Sleeping
Sleeping
File size: 7,447 Bytes
fa1fead | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 | # smartdocs/api/main.py
"""
WHY THIS EXISTS:
FastAPI application entry point.
Handles startup (DB connect, model warmup), shutdown (pool close),
middleware registration (user context), and route mounting.
LIFESPAN:
Uses the modern FastAPI lifespan context manager (not deprecated @app.on_event).
Startup: connects pgvector pool, warms up DenseEmbedder.
Shutdown: closes pgvector pool cleanly.
WHY WARMUP ON STARTUP:
multilingual-e5-large loads in 3-5s on first call.
Without warmup, the first user query experiences a 5s delay.
Warmup moves that cost to server startup where it's invisible to users.
HOW TO RUN:
From: smartdocs/
Dev: uv run uvicorn api.main:app --reload --port 8000 --log-level info
Prod: uv run uvicorn api.main:app --port 8000 --workers 1
NOTE: --workers 1 in production.
DenseEmbedder holds model in VRAM. Multiple workers = multiple model loads
= OOM on RTX 3050 6GB. One worker, async concurrency handles the load.
VERIFY SERVER IS RUNNING:
curl http://localhost:8000/health/ping
Expected: {"status": "alive"}
curl http://localhost:8000/health
Expected: {"status": "ok", "checks": {...}}
"""
from __future__ import annotations
import logging
from contextlib import asynccontextmanager
from fastapi import FastAPI
from fastapi.middleware.cors import CORSMiddleware
from api.middleware.user_context import UserContextMiddleware
from api.routes import health, ingest, query
from config.settings import get_settings
settings = get_settings()
# ββ Logging setup βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s | %(levelname)-8s | %(name)s | %(message)s",
datefmt="%Y-%m-%d %H:%M:%S",
)
logger = logging.getLogger(__name__)
# ββ Lifespan ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@asynccontextmanager
async def lifespan(app: FastAPI):
"""
Runs on startup and shutdown.
FastAPI recommends this over @app.on_event (deprecated in 0.95+).
Startup sequence:
1. Connect pgvector pool (fail fast if DB unreachable)
2. Warm up DenseEmbedder (loads multilingual-e5-large into VRAM)
3. Compile LangGraph (validates all nodes and edges)
β Server ready to accept requests
Shutdown sequence:
1. Close pgvector pool (drains in-flight queries)
"""
# ββ STARTUP βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
logger.info("SmartDocs API starting up...")
# 1. Connect pgvector pool
try:
from vectorstore.pgvector_client import get_pgvector_client
pgclient = get_pgvector_client()
await pgclient.connect()
logger.info("β pgvector pool connected")
except Exception as e:
logger.error(f"β pgvector connection FAILED: {e}")
logger.error("Check DATABASE_URL in .env β server will start but DB queries will fail")
# Don't raise β server starts degraded, /health will report the error
# 2. Warm up embedding model
try:
from embeddings.dense_embedder import get_embedder
import asyncio
embedder = get_embedder()
loop = asyncio.get_running_loop()
# embed_passages is synchronous β run in executor to avoid blocking event loop
await loop.run_in_executor(None, embedder.embed_passages, ["warmup"])
logger.info(f"β DenseEmbedder warmed up on {embedder.device}")
except Exception as e:
logger.warning(f"β DenseEmbedder warmup failed: {e} β will load on first query")
# 3. Compile LangGraph (validates all nodes and edges)
try:
from agents.smartdocs_graph import get_smartdocs_graph
get_smartdocs_graph()
logger.info("β LangGraph compiled")
except Exception as e:
logger.error(f"β LangGraph compile FAILED: {e}")
logger.info(
f"SmartDocs API ready | "
f"env={settings.environment} | "
f"model={settings.sarvam_model_30b}"
)
yield # Server runs here
# ββ SHUTDOWN ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
logger.info("SmartDocs API shutting down...")
try:
from vectorstore.pgvector_client import get_pgvector_client
pgclient = get_pgvector_client()
await pgclient.close()
logger.info("β pgvector pool closed")
except Exception as e:
logger.warning(f"pgvector pool close error: {e}")
logger.info("SmartDocs API shutdown complete")
# ββ App βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
app = FastAPI(
title="SmartDocs API",
description=(
"Multilingual PDF Q&A for Indian professionals. "
"22 Indian languages. No translation layer. "
"Powered by Sarvam-30B + multilingual-e5-large."
),
version="1.0.0",
lifespan=lifespan,
# Disable automatic redirect: /ingest β /ingest/
# Redirect strips headers including X-User-ID β breaks auth middleware
redirect_slashes=False,
)
# ββ Middleware ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Order matters: middleware executes in REVERSE registration order.
# UserContextMiddleware must run after CORS β register CORS last.
# User context β injects user_id into request.state for every request
app.add_middleware(UserContextMiddleware)
# CORS β allow all origins in development
# In production: replace "*" with your actual frontend domain
app.add_middleware(
CORSMiddleware,
allow_origins=["*"] if settings.environment == "development" else [
"https://your-streamlit-app.railway.app",
],
allow_credentials=True,
allow_methods=["GET", "POST", "DELETE", "OPTIONS"],
allow_headers=["*"],
)
# ββ Routes ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
app.include_router(health.router, tags=["Health"])
app.include_router(ingest.router, tags=["Ingestion"])
app.include_router(query.router, tags=["Query"])
# ββ Root ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@app.get("/")
async def root() -> dict:
"""
Root endpoint β confirms API is running.
curl http://localhost:8000/
"""
return {
"name": "SmartDocs API",
"version": "1.0.0",
"status": "running",
"docs": "/docs",
"health": "/health",
} |