codex-app-server/cloud/chv/browser_tasks.py

"""P8 browser_tasks：每浏览器任务一个实例——建/绑/毁生命周期。

对齐 Muse B4（官方安全文）：每个浏览器任务由专门浏览器子代理
完成。P2 已把浏览器子代理独立成进程（chv-browser.py，常驻、
close 只关页不杀进程）；本轨改为每任务一个实例：建实例、绑
定一个浏览器任务、用完即毁，状态绝不跨任务复用。

transport 接口：真实 transport 走 P2 的 broker 协议
（open/snapshot/screenshot/close/ping，经 /run/chv/browser.sock）；
客机侧“每任务起一个全新 chv-browser.py、close 后杀进程”
是差距表 §3/§6 的跟进项（P2 的 chv-broker.py 里改，本轨不
复制 P2 的文件）。本层只管生命周期、单实例单任务与状态隔离，
transport 注入（测试用 fake）。

存储：宿主级 AppStateStore（round-12 已迁移，见
docs/chv-muse-p8-gap.md §6）：state_dir/app-state.sqlite 的
``p8-browser-tasks`` 命名空间，一行存整份登记；旧
browser-tasks.json 首次加载时自动导入并改名 ``.migrated``；
行损坏/库打不开时回空结构（fail-safe，orphans 同款口径），
下次成功保存即覆盖自愈。
"""

from __future__ import annotations

import json
import sqlite3
import logging
import os
import time
import uuid

from cloud.chv.app_state import AppStateStore, host_state_db_path
from cloud.chv.compat215 import resolve_inside

_LOGGER = logging.getLogger("cloud.cloud_hypervisor")

_REGISTRY_FILENAME = "browser-tasks.json"
_KV_NAMESPACE = "p8-browser-tasks"
_KV_KEY = "registry"


class BrowserTransport:
    """浏览器实例 transport 接口：真实实现走 P2 的 broker 协议。

    spawn 建实例、close 毁实例、ping 探活。子类须实现三者；
    基类方法一律 NotImplementedError（绝不静默假装成功）。"""

    def spawn(self, instance: dict) -> None:
        """建一个浏览器实例（instance 含 instance_id/task_id）。"""
        raise NotImplementedError

    def close(self, instance: dict) -> None:
        """毁一个浏览器实例（进程退出、状态清理）。"""
        raise NotImplementedError

    def ping(self, instance: dict) -> bool:
        """实例是否还活着。"""
        raise NotImplementedError


def _open_store(state_dir: str) -> AppStateStore:
    """打开宿主级状态库（每次调用新建连接，用完即关）。"""
    store = AppStateStore(host_state_db_path(state_dir))
    store.open()
    return store


def _registry_path(state_dir: str) -> str:
    """旧版实例登记文件路径（只用于一次性迁移导入）。"""
    return os.path.join(state_dir, _REGISTRY_FILENAME)


def _quarantine(path: str) -> None:
    """把损坏的登记文件挪到 <path>.corrupt 旁（尽力而为）。"""
    try:
        os.replace(path, path + ".corrupt")
    except OSError:
        _LOGGER.warning("chv_browser_tasks_quarantine_failed "
                        "path=%s", path)


def _load_registry(state_dir: str) -> dict:
    """读登记：KV 行缺失时走一次性旧文件导入；行损坏/库打不
    开时回空结构（fail-safe，orphans 同款口径；下次成功保存
    即覆盖自愈）。"""
    try:
        with _open_store(state_dir) as store:
            raw = store.kv_get(_KV_NAMESPACE, _KV_KEY)
    except (sqlite3.Error, OSError):
        _LOGGER.warning("chv_browser_tasks_store_unreadable "
                        "state_dir=%s", state_dir)
        return {"instances": {}}
    if raw is None:
        return _import_legacy(state_dir)
    try:
        data = json.loads(raw)
    except ValueError:
        _LOGGER.warning("chv_browser_tasks_corrupt state_dir=%s",
                        state_dir)
        return {"instances": {}}
    if not isinstance(data, dict) \
            or not isinstance(data.get("instances"), dict):
        return {"instances": {}}
    return data


def _import_legacy(state_dir: str) -> dict:
    """一次性导入旧 browser-tasks.json：合法即进 KV 并把文件
    改名 ``.migrated``；无文件回空；损坏则 quarantine 后回空
    （旧口径）。导入失败时返回读到的数据、文件保留，下次加
    载重试。"""
    path = _registry_path(state_dir)
    try:
        with open(path, "r") as fh:
            data = json.load(fh)
    except FileNotFoundError:
        return {"instances": {}}
    except (OSError, ValueError):
        _quarantine(path)
        return {"instances": {}}
    if not isinstance(data, dict) \
            or not isinstance(data.get("instances"), dict):
        _quarantine(path)
        return {"instances": {}}
    try:
        with _open_store(state_dir) as store:
            store.kv_put(_KV_NAMESPACE, _KV_KEY, json.dumps(data))
    except (sqlite3.Error, OSError):
        _LOGGER.warning("chv_browser_tasks_import_failed "
                        "state_dir=%s", state_dir)
        return data
    try:
        os.replace(path, path + ".migrated")
    except OSError:
        pass
    return data


def _save_registry(state_dir: str, data: dict) -> None:
    """落盘登记到 KV 行；写失败只记日志。"""
    try:
        with _open_store(state_dir) as store:
            store.kv_put(_KV_NAMESPACE, _KV_KEY, json.dumps(data))
    except (sqlite3.Error, OSError):
        _LOGGER.warning("chv_browser_tasks_save_failed "
                        "state_dir=%s", state_dir)


def _get_or_raise(data: dict, task_id: str) -> dict:
    """取某任务的实例；不存在即 ValueError。"""
    instance = data.get("instances", {}).get(task_id)
    if not isinstance(instance, dict):
        raise ValueError("browser instance not found")
    return instance


def create_instance(state_dir: str, task_id: str,
                    transport: BrowserTransport,
                    now: float | None = None) -> dict:
    """为一个浏览器任务建一个实例：单任务单实例（该任务已有
    任何实例记录——active、destroyed 乃至 failed——即
    ValueError；一个任务一生只有一个实例，毁后不复建，审计
    轨迹留在登记里）；先落盘 creating、再调 transport.spawn、
    成功后置 active；spawn 抛错时记 failed 并原样抛出
    （不留半个实例）。返回实例 dict。"""
    if now is None:
        now = time.time()
    if not task_id or not isinstance(task_id, str):
        raise ValueError("task_id required")
    data = _load_registry(state_dir)
    existing = data["instances"].get(task_id)
    if isinstance(existing, dict):
        raise ValueError("browser instance already exists")
    instance_id = uuid.uuid4().hex
    instance = {
        "instance_id": instance_id,
        "task_id": task_id,
        "status": "creating",
        "shots_dir": "browser-tasks/%s/shots" % instance_id,
        "created_at": now,
    }
    data["instances"][task_id] = instance
    _save_registry(state_dir, data)
    try:
        transport.spawn(instance)
    except Exception:
        instance["status"] = "failed"
        _save_registry(state_dir, data)
        raise
    instance["status"] = "active"
    _save_registry(state_dir, data)
    return dict(instance)


def get_instance(state_dir: str, task_id: str) -> dict | None:
    """取某任务的实例；不存在回 None。"""
    data = _load_registry(state_dir)
    instance = data.get("instances", {}).get(task_id)
    if not isinstance(instance, dict):
        return None
    return dict(instance)


def list_instances(state_dir: str) -> list[dict]:
    """列出全部实例（按创建时间排序）。"""
    data = _load_registry(state_dir)
    out = [dict(v) for v in data["instances"].values()
           if isinstance(v, dict)]
    out.sort(key=lambda i: (i.get("created_at") or 0,
                            i.get("task_id") or ""))
    return out


def destroy_instance(state_dir: str, task_id: str,
                     transport: BrowserTransport,
                     now: float | None = None) -> dict:
    """毁一个实例：须是 active（不存在/已毁即 ValueError）；
    调 transport.close，close 抛错也记下并照常置 destroyed
    （清理不能因一次 close 失败卡住），返回实例 dict。"""
    if now is None:
        now = time.time()
    data = _load_registry(state_dir)
    instance = _get_or_raise(data, task_id)
    if instance.get("status") != "active":
        raise ValueError("browser instance is not active")
    close_error = ""
    try:
        transport.close(instance)
    except Exception as exc:
        close_error = str(exc)[:300]
    instance["status"] = "destroyed"
    instance["destroyed_at"] = now
    instance["close_error"] = close_error
    _save_registry(state_dir, data)
    return dict(instance)


def ping_instance(state_dir: str, task_id: str,
                  transport: BrowserTransport) -> bool:
    """实例是否还活着：非 active 一律 False；active 时问
    transport。"""
    data = _load_registry(state_dir)
    instance = data.get("instances", {}).get(task_id)
    if not isinstance(instance, dict):
        return False
    if instance.get("status") != "active":
        return False
    try:
        return bool(transport.ping(instance))
    except Exception:
        return False


def instance_shots_dir(state_dir: str, task_id: str) -> str | None:
    """取某任务实例的截图目录（state_dir 内、resolve_inside
    校验，越界回 None）；实例不存在回 None。"""
    data = _load_registry(state_dir)
    instance = data.get("instances", {}).get(task_id)
    if not isinstance(instance, dict):
        return None
    rel = instance.get("shots_dir")
    if not isinstance(rel, str) or not rel:
        return None
    return resolve_inside(state_dir, rel)