ctms/backend/app/services/monitoring_server_location.py

"""Resolve the monitoring deployment's public network location.

The map server marker is derived from the deployment's public IP instead of a
frontend or configuration coordinate. An explicit public IP is accepted for
restricted networks; otherwise the public frontend hostname and, finally, a
small set of public-IP discovery endpoints are used. Results are cached so the
monitoring API never performs per-request network discovery.
"""

from __future__ import annotations

import asyncio
import ipaddress
import logging
import socket
import time
from dataclasses import dataclass
from urllib.parse import urlparse

import httpx

from app.core.config import settings
from app.services.geo_location_metadata import resolve_geo_location_metadata
from app.services.ip_geolocation_fallback import resolve_external_ip_locations
from app.services.ip_location import resolve_ip_location

logger = logging.getLogger("ctms.monitoring_server_location")


@dataclass(frozen=True)
class MonitoringServerLocation:
    name: str
    longitude: float
    latitude: float
    accuracy_level: str
    resolution_source: str

    def to_public_dict(self) -> dict:
        return {
            "name": self.name,
            "longitude": self.longitude,
            "latitude": self.latitude,
            "accuracy_level": self.accuracy_level,
            "resolution_source": self.resolution_source,
        }


_cached_location: MonitoringServerLocation | None = None
_cache_initialized = False
_cache_expires_at = 0.0
_cache_lock = asyncio.Lock()


def _normalize_public_ip(value: str | None) -> str | None:
    candidate = (value or "").strip().splitlines()[0] if (value or "").strip() else ""
    try:
        address = ipaddress.ip_address(candidate)
    except ValueError:
        return None
    return str(address) if address.is_global else None


async def _resolve_frontend_public_ip() -> str | None:
    hostname = urlparse(settings.FRONTEND_PUBLIC_URL).hostname
    if not hostname:
        return None
    literal_ip = _normalize_public_ip(hostname)
    if literal_ip:
        return literal_ip
    try:
        loop = asyncio.get_running_loop()
        records = await loop.getaddrinfo(hostname, None, type=socket.SOCK_STREAM)
    except (OSError, socket.gaierror):
        return None
    candidates = []
    for _, _, _, _, socket_address in records:
        candidate = _normalize_public_ip(str(socket_address[0]))
        if candidate and candidate not in candidates:
            candidates.append(candidate)
    return next((item for item in candidates if ":" not in item), candidates[0] if candidates else None)


async def _discover_public_ip() -> tuple[str | None, str]:
    configured_ip = _normalize_public_ip(settings.MONITORING_SERVER_PUBLIC_IP)
    if configured_ip:
        return configured_ip, "configured_public_ip"

    frontend_ip = await _resolve_frontend_public_ip()
    if frontend_ip:
        return frontend_ip, "frontend_dns"

    if settings.ENV == "test":
        return None, "unavailable"

    urls = [
        item.strip()
        for item in settings.MONITORING_PUBLIC_IP_DISCOVERY_URLS.split(",")
        if item.strip()
    ]
    timeout = httpx.Timeout(settings.MONITORING_PUBLIC_IP_DISCOVERY_TIMEOUT_SECONDS)
    async with httpx.AsyncClient(timeout=timeout, follow_redirects=False) as client:
        for url in urls:
            try:
                response = await client.get(url, headers={"Accept": "text/plain"})
                response.raise_for_status()
            except httpx.HTTPError:
                logger.warning("Public IP discovery endpoint unavailable", extra={"endpoint": url})
                continue
            discovered_ip = _normalize_public_ip(response.text[:128])
            if discovered_ip:
                return discovered_ip, "public_ip_discovery"
    return None, "unavailable"


async def resolve_monitoring_server_location(*, force: bool = False) -> MonitoringServerLocation | None:
    global _cached_location, _cache_initialized, _cache_expires_at

    now = time.monotonic()
    if not force and _cache_initialized and now < _cache_expires_at:
        return _cached_location

    async with _cache_lock:
        now = time.monotonic()
        if not force and _cache_initialized and now < _cache_expires_at:
            return _cached_location

        public_ip, resolution_source = await _discover_public_ip()
        location = None
        if public_ip:
            ip_location = resolve_ip_location(public_ip)
            metadata = resolve_geo_location_metadata(ip_location)
            if metadata.longitude is None or metadata.latitude is None:
                external_location = (await resolve_external_ip_locations([public_ip])).get(public_ip)
                if external_location is not None:
                    ip_location = external_location.merge_ip_location(ip_location)
                    metadata = external_location.to_metadata()
            if metadata.longitude is not None and metadata.latitude is not None:
                country = metadata.country or ip_location.country
                name = " / ".join(
                    part for part in [country, ip_location.province, ip_location.city] if part
                ) or "公网服务器"
                location = MonitoringServerLocation(
                    name=name,
                    longitude=metadata.longitude,
                    latitude=metadata.latitude,
                    accuracy_level=metadata.accuracy_level,
                    resolution_source=resolution_source,
                )
                logger.info(
                    "Monitoring server location resolved",
                    extra={"resolution_source": resolution_source, "accuracy_level": metadata.accuracy_level},
                )
            else:
                logger.warning("Deployment public IP has no usable geo coordinates")
        else:
            logger.warning("Unable to discover the deployment public IP")

        _cached_location = location
        _cache_initialized = True
        cache_seconds = settings.MONITORING_SERVER_LOCATION_CACHE_SECONDS if location else 300
        _cache_expires_at = now + cache_seconds
        return location


def reset_monitoring_server_location_cache() -> None:
    """Reset process-local state for configuration reloads and tests."""
    global _cached_location, _cache_initialized, _cache_expires_at
    _cached_location = None
    _cache_initialized = False
    _cache_expires_at = 0.0