node_runtime.py

"""User-space Node.js runtime bootstrap for the Respite backend.

Runs the existing Next.js backend inside the Gradio Space container without
root: downloads an official self-contained Node tarball to a local cache,
extracts it, and launches the Next.js standalone server as a child process
bound to localhost. The FastAPI layer reverse-proxies /api/* to it.

Env:
  NODE_VERSION   Node version to fetch (default 22.14.0 LTS)
  NEXT_PORT      localhost port for the Node server (default 3210)
  NEXT_DIR       directory of the standalone build (default ./respite-backend)
  NODE_PROXY_OFF  set to "1" to skip the download/proxy entirely
"""

import os
import shutil
import subprocess
import tarfile
import threading
import urllib.request

BACKEND_DIR = os.environ.get("NEXT_DIR", os.path.join(os.path.dirname(os.path.abspath(__file__)), "respite-backend"))
NODE_VERSION = os.environ.get("NODE_VERSION", "22.14.0")
NODE_PORT = int(os.environ.get("NEXT_PORT", "3210"))

NODE_TARBALL = f"https://nodejs.org/dist/v{NODE_VERSION}/node-v{NODE_VERSION}-linux-x64.tar.xz"
RUNTIME_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), ".node-runtime")
NODE_BIN = os.path.join(RUNTIME_DIR, f"node-v{NODE_VERSION}-linux-x64", "bin", "node")

_node_proc = None
_node_lock = threading.Lock()
_node_ready = threading.Event()


def _log(msg):
    print(f"[node-runtime] {msg}", flush=True)


def ensure_node() -> str:
    """Download + extract Node into the local cache once. Returns node path."""
    if os.path.exists(NODE_BIN):
        return NODE_BIN
    os.makedirs(RUNTIME_DIR, exist_ok=True)
    tgz = os.path.join(RUNTIME_DIR, "node.tar.xz")
    _log(f"downloading {NODE_TARBALL}")
    urllib.request.urlretrieve(NODE_TARBALL, tgz)
    _log("extracting")
    with tarfile.open(tgz, "r:xz") as tf:
        tf.extractall(RUNTIME_DIR)
    os.remove(tgz)
    os.chmod(NODE_BIN, 0o755)
    _log(f"node ready at {NODE_BIN}")
    return NODE_BIN


def _wait_ready(proc: subprocess.Popen, timeout: float = 90.0) -> bool:
    """Wait until the Node server answers on its localhost port."""
    import socket
    import time

    deadline = time.time() + timeout
    while time.time() < deadline:
        if proc.poll() is not None:
            _log(f"node exited early with code {proc.returncode}")
            return False
        try:
            with socket.create_connection(("127.0.0.1", NODE_PORT), timeout=1):
                return True
        except OSError:
            time.sleep(0.5)
    return False


def _ensure_backend_extracted() -> bool:
    """Extract backend-bundle/respite-backend.tar.xz if the backend dir is absent."""
    if os.path.isdir(BACKEND_DIR) and os.path.exists(os.path.join(BACKEND_DIR, "server.js")):
        return True
    bundle = os.path.join(os.path.dirname(os.path.abspath(__file__)), "backend-bundle", "respite-backend.tar.xz")
    if not os.path.exists(bundle):
        _log(f"no bundle at {bundle}")
        return False
    _log(f"extracting {bundle} -> {BACKEND_DIR}")
    os.makedirs(BACKEND_DIR, exist_ok=True)
    with tarfile.open(bundle, "r:xz") as tf:
        tf.extractall(BACKEND_DIR)
    _log("backend extracted")
    return True


def start_node_backend() -> bool:
    """Launch the Next.js standalone server. Safe to call once at startup."""
    global _node_proc
    if os.environ.get("NODE_PROXY_OFF") == "1":
        _log("disabled via NODE_PROXY_OFF")
        return False
    if not _ensure_backend_extracted():
        return False
    with _node_lock:
        if _node_proc and _node_proc.poll() is None:
            return True
        try:
            node = ensure_node()
        except Exception as e:
            _log(f"node bootstrap failed: {e}")
            return False
        server_js = os.path.join(BACKEND_DIR, "server.js")
        if not os.path.exists(server_js):
            _log(f"no server.js in {BACKEND_DIR}")
            return False
        env = os.environ.copy()
        env.update({
            "PORT": str(NODE_PORT),
            "HOSTNAME": "127.0.0.1",
            "NODE_ENV": "production",
        })
        _log(f"starting node server on 127.0.0.1:{NODE_PORT}")
        _node_proc = subprocess.Popen(
            [node, server_js],
            cwd=BACKEND_DIR,
            env=env,
            stdout=subprocess.PIPE,
            stderr=subprocess.STDOUT,
            text=True,
            bufsize=1,
        )
        threading.Thread(target=_pump_logs, daemon=True).start()
        ok = _wait_ready(_node_proc)
        if ok:
            _log("node backend is READY")
            _node_ready.set()
        else:
            _log("node backend failed to become ready")
        return ok


def _pump_logs():
    assert _node_proc and _node_proc.stdout
    for line in _node_proc.stdout:
        _log(f"[next] {line.rstrip()}")


def node_ready() -> bool:
    return _node_ready.is_set() and _node_proc is not None and _node_proc.poll() is None