# -*- coding: utf-8 -*-
"""207 断电重启后恢复服务。"""
from __future__ import annotations

import sys
import time
import urllib.error
import urllib.request

import paramiko

HOST = "192.168.110.207"
USER = "root"
KEY = r"C:\Users\DCKJ\.ssh\id_ed25519"


def ssh() -> paramiko.SSHClient:
    c = paramiko.SSHClient()
    c.set_missing_host_key_policy(paramiko.AutoAddPolicy())
    c.connect(
        HOST,
        username=USER,
        key_filename=KEY,
        timeout=30,
        allow_agent=False,
        look_for_keys=False,
    )
    return c


def run(c: paramiko.SSHClient, cmd: str, timeout: int = 300) -> tuple[int, str]:
    print(f"\n>>> {cmd[:220]}")
    _, o, e = c.exec_command(cmd, timeout=timeout, get_pty=True)
    out = o.read().decode(errors="replace")
    err = e.read().decode(errors="replace")
    code = o.channel.recv_exit_status()
    text = (out + ("\n" + err if err.strip() else "")).strip()
    if text:
        print(text[-6000:] if len(text) > 6000 else text)
    print(f"[exit={code}]")
    return code, text


def http_ok(url: str, timeout: float = 6.0) -> str:
    try:
        r = urllib.request.urlopen(url, timeout=timeout)
        return f"OK {r.status}"
    except urllib.error.HTTPError as e:
        return f"HTTP {e.code}"
    except Exception as e:
        return f"DOWN {type(e).__name__}"


def probe() -> None:
    checks = [
        ("Dify :80", "http://192.168.110.207/"),
        ("LiteLLM", "http://192.168.110.207:4000/health/liveliness"),
        ("Langfuse", "http://192.168.110.207:3000/api/public/health"),
        ("Xinference", "http://192.168.110.207:9997/v1/models"),
        ("Higress", "http://192.168.110.207:18080/"),
        ("Higress Console", "http://192.168.110.207:8001/"),
        ("GitLab", "http://192.168.110.207:8929/users/sign_in"),
        ("Argo CD", "http://192.168.110.207:30080/"),
        ("demo-nginx", "http://192.168.110.207:30088/"),
        ("Grafana", "http://192.168.110.207:3001/"),
        ("Prometheus", "http://192.168.110.207:9092/-/ready"),
    ]
    print("\n=== HTTP probe ===")
    for name, url in checks:
        print(f"{name:16} {http_ok(url):20} {url}")


def main() -> int:
    probe()
    c = ssh()
    print("[OK] SSH connected")

    run(c, "uptime; free -h | head -2; systemctl is-active docker k3s litellm 2>/dev/null || true")
    run(c, "docker ps -a --format 'table {{.Names}}\t{{.Status}}' | head -80")

    # 1) Docker
    run(c, "systemctl enable docker 2>/dev/null; systemctl start docker; sleep 2; systemctl is-active docker")

    # 2) LiteLLM systemd
    run(c, "systemctl enable litellm 2>/dev/null; systemctl restart litellm; sleep 2; systemctl is-active litellm || true")

    # 3) Dify compose
    run(
        c,
        "cd /dify/dify-main/docker && (docker compose up -d || docker-compose up -d) 2>&1 | tail -40",
        timeout=600,
    )

    # 4) Langfuse
    run(
        c,
        "if [ -d /Langfuse/langfuse ]; then cd /Langfuse/langfuse && (docker compose up -d || docker-compose up -d) 2>&1 | tail -30; "
        "elif [ -d /opt/langfuse ]; then cd /opt/langfuse && (docker compose up -d || docker-compose up -d) 2>&1 | tail -30; "
        "else echo 'langfuse dir missing'; fi",
        timeout=600,
    )

    # 5) Xinference
    run(
        c,
        "if docker ps -a --format '{{.Names}}' | grep -qx xinference; then docker start xinference; "
        "elif docker ps -a --format '{{.Names}}' | grep -qi xinference; then "
        "docker start $(docker ps -a --format '{{.Names}}' | grep -i xinference | head -1); "
        "else echo 'xinference container missing'; docker ps -a | grep -i xinf || true; fi",
        timeout=120,
    )

    # 6) Higress
    run(
        c,
        "if [ -x /opt/higress/bin/startup.sh ]; then /opt/higress/bin/startup.sh 2>&1 | tail -40; "
        "elif [ -d /opt/higress/compose ]; then cd /opt/higress/compose && (docker compose up -d || docker-compose up -d) 2>&1 | tail -40; "
        "else echo 'higress missing'; ls /opt/higress 2>/dev/null || true; fi",
        timeout=600,
    )

    # 7) Monitoring hw8
    run(
        c,
        "if [ -f /opt/workbuddy/patches_cursor/作业8_SLA监控大盘/docker-compose.monitoring.yml ]; then "
        "cd /opt/workbuddy/patches_cursor/作业8_SLA监控大盘 && "
        "(docker compose -f docker-compose.monitoring.yml up -d || docker-compose -f docker-compose.monitoring.yml up -d) 2>&1 | tail -20; "
        "else echo 'monitoring compose missing'; fi",
        timeout=300,
    )

    # 8) k3s + GitLab demo
    run(c, "systemctl enable k3s 2>/dev/null; systemctl start k3s; sleep 5; systemctl is-active k3s || true")
    run(
        c,
        "if docker ps -a --format '{{.Names}}' | grep -qx gitlab-argocd-demo; then docker start gitlab-argocd-demo; "
        "else echo 'gitlab-argocd-demo missing'; fi",
        timeout=120,
    )
    run(
        c,
        "export KUBECONFIG=/etc/rancher/k3s/k3s.yaml; "
        "k3s kubectl get nodes 2>&1 | head -5; "
        "k3s kubectl -n argocd get pods 2>&1 | head -20; "
        "k3s kubectl -n argocd-demo get pods 2>&1 | head -10",
        timeout=120,
    )

    # wait for backends
    print("\n[INFO] waiting 45s for containers…")
    time.sleep(45)
    run(c, "docker ps --format 'table {{.Names}}\t{{.Status}}' | head -80")
    run(
        c,
        "cd /dify/dify-main/docker && (docker compose ps || docker-compose ps) 2>&1 | head -40",
        timeout=60,
    )

    c.close()
    probe()
    print("\n[DONE] restore script finished")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())