start_services.py

#!/usr/bin/env python3
"""
Robust launcher for Spaces:
- Starts the FastAPI app (e.g. main:app) on 127.0.0.1:5350 using uvicorn as a subprocess.
- Waits until the health endpoint responds.
- Then execs the Gradio app file (gradio_app.py by default) so Gradio becomes the main foreground process.
"""

import os
import time
import requests
import subprocess
import sys
from shutil import which

TRYON_MODULE = os.getenv("TRYON_MODULE", "main:app")  # your FastAPI module: main:app
TRYON_HOST = os.getenv("TRYON_HOST", "127.0.0.1")
TRYON_PORT = int(os.getenv("TRYON_PORT", "5350"))
HEALTH_PATH = os.getenv("TRYON_HEALTH_PATH", "/api/v1/health")
TRYON_HEALTH_URL = f"http://{TRYON_HOST}:{TRYON_PORT}{HEALTH_PATH}"

# log level: set TRYON_LOG_LEVEL=debug to get verbose uvicorn logs
TRYON_LOG_LEVEL = os.getenv("TRYON_LOG_LEVEL", "info").lower()
if TRYON_LOG_LEVEL not in ("critical", "error", "warning", "info", "debug", "trace"):
    TRYON_LOG_LEVEL = "info"

UVICORN_BIN = which("uvicorn") or "uvicorn"
UVICORN_CMD = [
    UVICORN_BIN,
    TRYON_MODULE,
    "--host", TRYON_HOST,
    "--port", str(TRYON_PORT),
    "--log-level", TRYON_LOG_LEVEL
]

print("Starting try-on server:", " ".join(UVICORN_CMD))
proc = subprocess.Popen(UVICORN_CMD, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)

# Wait for health to respond
timeout = int(os.getenv("TRYON_STARTUP_TIMEOUT", "30"))
start = time.time()
last_exception = None
while True:
    try:
        r = requests.get(TRYON_HEALTH_URL, timeout=2)
        if r.status_code == 200:
            print("Try-on health OK:", TRYON_HEALTH_URL)
            break
        else:
            print("Try-on health returned:", r.status_code, r.text[:200])
    except Exception as e:
        last_exception = e
    if time.time() - start > timeout:
        # Dump subprocess stdout/stderr for debugging
        print("Try-on failed to start within timeout. Dumping logs...")
        try:
            out, err = proc.communicate(timeout=2)
            print("try-on stdout:\n", out or "<no stdout>")
            print("try-on stderr:\n", err or "<no stderr>")
        except Exception:
            proc.kill()
            print("Could not communicate with try-on process; killed it.")
        raise RuntimeError(f"Try-on failed to start within {timeout}s. Last error: {last_exception}")
    time.sleep(0.5)

# Exec Gradio app so it becomes PID1 (Spaces expects the main app in foreground)
GRADIO_APP = os.getenv("GRADIO_APP_FILE", "gradio_app.py")
print("Launching Gradio app:", GRADIO_APP)
os.execv(sys.executable, [sys.executable, GRADIO_APP])