xas-test-library/deploy/dedicated.py

"""First deployment / explicit port cutover. Only manages this application's resources."""
import argparse
import json
import os
from pathlib import Path
import shutil
import socket
import subprocess
import sys

ROOT = Path("/adamwen-work-space/xas-test-library")
NAME = "adamwen-script-console"
IMAGE = "adamwen/script-console:20261020-1"
UNIT = "adamwen-script-worker.service"


def validate_runtime(root):
    for path in (root / "runtime", *(root / "runtime" / name for name in ("console-data", "library", "runs", "native-ipc"))):
        if not path.resolve().is_relative_to(root.resolve()):
            raise ValueError("Runtime must remain inside this project")
        if any(part.is_symlink() for part in (path, *path.parents)):
            raise ValueError("Runtime symlink rejected")


def command(*args, timeout=30):
    return subprocess.check_output(args, text=True, timeout=timeout).strip()


def web_arguments(port):
    validate_runtime(ROOT)
    runtime = ROOT / "runtime"
    args = ["docker", "run", "-d", "--name", NAME, "--label", "xas.application=script-console",
            "--runtime", "runc", "--network", "host", "--user", "0:0", "--cap-drop", "ALL",
            "--security-opt", "no-new-privileges", "--read-only", "--memory", "512m", "--cpus", "1",
            "--pids-limit", "96", "--tmpfs", "/tmp:rw,nosuid,nodev,noexec,size=67108864",
            "--log-driver", "local", "--log-opt", "max-size=10m", "--log-opt", "max-file=2",
            "--env", "XAS_CONSOLE_ROOT=" + str(runtime)]
    for name in ("console-data", "library", "runs", "native-ipc"):
        path = runtime / name
        args += ["--mount", f"type=bind,src={path},dst={path}" + (",readonly" if name in {"runs", "native-ipc"} else "")]
    sources = json.loads((ROOT / "catalog/external-sources.json").read_text())["sources"]
    for source in sources:
        path = Path(source["path"])
        if path.is_dir():
            args += ["--mount", f"type=bind,src={path},dst={path},readonly"]
            if path.name == "scripts":
                for folder in ("logs", "results"):
                    logs = path.parent / folder
                    if logs.is_dir():
                        args += ["--mount", f"type=bind,src={logs},dst={logs},readonly"]
    return args + [IMAGE, "--port", str(port)]


def prepare():
    if os.geteuid() != 0 or ROOT.resolve() != ROOT or not (ROOT / "console/worker.py").is_file():
        raise RuntimeError("Run through the documented sudo flow in the confirmed project")
    validate_runtime(ROOT)
    for name in ("console-data", "library", "runs", "native-ipc"):
        path = ROOT / "runtime" / name
        if path.is_symlink():
            raise RuntimeError("Runtime symlink rejected")
        path.mkdir(mode=0o750, exist_ok=True)
    runner = command("docker", "image", "inspect", "--format", "{{.Id}}", "adamwen/test-console:20261013-2")
    config = {"project": str(ROOT), "runner_image": runner,
              "smoke_upstream": "http://100.64.0.55:18100",
              "smoke_credential": "/dengyuyun-joy-work-space/workpc-claude-code/20261010-dsv41-trial/configs/vllm.env"}
    path = ROOT / "deploy/console-worker.json"
    if path.exists() and json.loads(path.read_text()) != config:
        raise RuntimeError("Existing worker configuration differs; review before replacing")
    path.write_text(json.dumps(config, indent=2))
    path.chmod(0o600)
    # Existing roots are 0700/root. Container root has no capabilities and read-only mounts.
    # Runtime code used by non-root jobs needs read access, without changing original scripts.
    for p in (ROOT / "console").rglob("*"):
        p.chmod(0o755 if p.is_dir() else 0o644)
    (ROOT / "console").chmod(0o755)


def start(port):
    existing = command("docker", "ps", "-aq", "--filter", "name=^/" + NAME + "$")
    if existing:
        raise RuntimeError("Console container already exists; inspect before replacement")
    with socket.socket() as probe:
        probe.bind(("10.88.118.188", port))
    print(command(*web_arguments(port)))
    service = Path("/etc/systemd/system") / UNIT
    content = (ROOT / "deploy/console-worker.service").read_bytes()
    if service.exists() and service.read_bytes() != content:
        raise RuntimeError("Existing systemd service differs")
    service.write_bytes(content)
    command("systemctl", "daemon-reload")
    # Start only; no automatic boot scheduling is enabled by this deployment.
    command("systemctl", "start", UNIT)
    print(command("systemctl", "is-active", UNIT))


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--port", type=int, choices=[14440, 14442], default=14442)
    options = parser.parse_args()
    prepare()
    start(options.port)