/** * 感谢 群友 @utopter 投稿 * 说明: * - 增加了用于匹配对特定正则项目进行检测,检测项目为 iOS 和 Chrome,iOS 检测 ios.chat.openai.com 返回 403,Chrome 检测 chat.openai.com GET 返回 200 * - [node_name_regex] 用于匹配节点名称的正则表达式 * * HTTP META(https://github.com/xream/http-meta) 参数 * - [http_meta_protocol] 协议 默认: http * - [http_meta_host] 服务地址 默认: 127.0.0.1 * - [http_meta_port] 端口号 默认: 9876 * - [http_meta_start_delay] 初始启动延时(单位: 毫秒) 默认: 3000 * - [http_meta_proxy_timeout] 每个节点耗时(单位: 毫秒). 此参数是为了防止脚本异常退出未关闭核心. 设置过小将导致核心过早退出. 目前逻辑: 启动初始的延时 + 每个节点耗时. 默认: 10000 * * 其它参数 * - [timeout] 请求超时(单位: 毫秒) 默认 5000 * - [retries] 重试次数 默认 1 * - [retry_delay] 重试延时(单位: 毫秒) 默认 1000 * - [concurrency] 并发数 默认 10 * - [client] GPT 检测的客户端类型. 默认 iOS * - [method] 请求方法. 默认 head, 如果不支持, 可设为 get * - [node_name_regex] 用于匹配节点名称的正则表达式 * - [cache] 使用缓存, 默认不使用缓存 * 关于缓存时长 * 当使用相关脚本时, 若在对应的脚本中使用参数开启缓存, 可设置持久化缓存 sub-store-csr-expiration-time 的值来自定义默认缓存时长, 默认为 172800000 (48 * 3600 * 1000, 即 48 小时) * 🎈Loon 可在插件中设置 * 其他平台同理, 持久化缓存数据在 JSON 里 * 可以在脚本的前面添加一个脚本操作, 实现保留 1 小时的缓存. 这样比较灵活 * async function operator() { * scriptResourceCache._cleanup(undefined, 1 * 3600 * 1000); * } */ async function operator(proxies = [], targetPlatform, context) { const cacheEnabled = $arguments.cache const cache = scriptResourceCache const http_meta_host = $arguments.http_meta_host ?? '127.0.0.1' const http_meta_port = $arguments.http_meta_port ?? 9876 const http_meta_protocol = $arguments.http_meta_protocol ?? 'http' const http_meta_api = `${http_meta_protocol}://${http_meta_host}:${http_meta_port}` const http_meta_start_delay = parseFloat($arguments.http_meta_start_delay ?? 3000) const http_meta_proxy_timeout = parseFloat($arguments.http_meta_proxy_timeout ?? 10000) const method = $arguments.method || 'head' const url = $arguments.client === 'Chrome' ? `https://chat.openai.com` : `https://ios.chat.openai.com` // 获取正则表达式 const regex = new RegExp($arguments.node_name_regex || '.*') const $ = $substore // 过滤节点列表,仅保留匹配正则表达式的节点 const filteredProxies = proxies.filter(proxy => regex.test(proxy.name)) const internalProxies = [] filteredProxies.map((proxy, index) => { try { const node = ProxyUtils.produce([proxy], 'ClashMeta', 'internal')?.[0] if (node) { // $.info(JSON.stringify(node, null, 2)) internalProxies.push({ ...node, _proxies_index: index }) } } catch (e) { $.error(e) } }) // $.info(JSON.stringify(internalProxies, null, 2)) $.info(`核心支持节点数: ${internalProxies.length}/${filteredProxies.length}`) if (!internalProxies.length) return proxies if (cacheEnabled) { try { let allCached = true for (var i = 0; i < internalProxies.length; i++) { const proxy = internalProxies[i] const id = getCacheId({ proxy, url }) const cached = cache.get(id) if (cached) { if (cached.gpt) { proxies[proxy._proxies_index].name = `[GPT] ${proxies[proxy._proxies_index].name}` } } else { allCached = false break } } if (allCached) { $.info('所有节点都有有效缓存 完成') return proxies } } catch (e) {} } const http_meta_timeout = http_meta_start_delay + internalProxies.length * http_meta_proxy_timeout let http_meta_pid let http_meta_ports = [] // 启动 HTTP META const res = await http({ retries: 0, method: 'post', url: `${http_meta_api}/start`, headers: { 'Content-type': 'application/json', }, body: JSON.stringify({ proxies: internalProxies, timeout: http_meta_timeout, }), }) let body = res.body try { body = JSON.parse(body) } catch (e) {} const { ports, pid } = body if (!pid || !ports) { throw new Error(`======== HTTP META 启动失败 ====\n${body}`) } http_meta_pid = pid http_meta_ports = ports $.info( `\n======== HTTP META 启动 ====\n[端口] ${ports}\n[PID] ${pid}\n[超时] 若未手动关闭 ${ Math.round(http_meta_timeout / 60 / 10) / 100 } 分钟后自动关闭\n` ) $.info(`等待 ${http_meta_start_delay / 1000} 秒后开始检测`) await $.wait(http_meta_start_delay) const batches = [] const concurrency = parseInt($arguments.concurrency || 10) // 一组并发数 for (let i = 0; i < internalProxies.length; i += concurrency) { const batch = internalProxies.slice(i, i + concurrency) batches.push(batch) } for (const batch of batches) { await Promise.all(batch.map(check)) } // stop http meta try { const res = await http({ method: 'post', url: `${http_meta_api}/stop`, headers: { 'Content-type': 'application/json', }, body: JSON.stringify({ pid: [http_meta_pid], }), }) $.info(`\n======== HTTP META 关闭 ====\n${JSON.stringify(res, null, 2)}`) } catch (e) { $.error(e) } return proxies async function check(proxy) { // $.info(`[${proxy.name}] 检测`) // $.info(`检测 ${JSON.stringify(proxy, null, 2)}`) const id = cacheEnabled ? getCacheId({ proxy, url }) : undefined // $.info(`检测 ${id}`) try { const cached = cache.get(id) if (cacheEnabled && cached) { $.info(`[${proxy.name}] 使用缓存`) if (cached.gpt) { proxies[proxy._proxies_index].name = `[GPT] ${proxies[proxy._proxies_index].name}` } return } // $.info(JSON.stringify(proxy, null, 2)) const index = internalProxies.indexOf(proxy) const startedAt = Date.now() const res = await http({ proxy: `http://${http_meta_host}:${http_meta_ports[index]}`, method, headers: { 'User-Agent': 'Mozilla/5.0 (iPhone; CPU iPhone OS 17_4 like Mac OS X) AppleWebKit/605.1.15 (KHTML, like Gecko) Version/17.3.1 Mobile/15E148 Safari/604.1', }, url, }) const status = parseInt(res.status || res.statusCode || 200) let latency = '' latency = `${Date.now() - startedAt}` $.info(`[${proxy.name}] status: ${status}, latency: ${latency}`) // cf 拦截是 400 错误, 403 就是没被拦截, 走到了未鉴权的逻辑,网页版返回200为正确 // https://zset.cc/archives/34/ const isIOS = $arguments.client === 'iOS' const successStatus = isIOS ? 403 : 200 // 根据客户端类型设置成功状态码 if (status === successStatus) { proxies[proxy._proxies_index].name = `[GPT] ${proxies[proxy._proxies_index].name}` if (cacheEnabled) { $.info(`[${proxy.name}] 设置成功缓存`) cache.set(id, { gpt: true }) } } else { if (cacheEnabled) { $.info(`[${proxy.name}] 设置失败缓存`) cache.set(id, {}) } } } catch (e) { $.error(`[${proxy.name}] ${e.message ?? e}`) if (cacheEnabled) { $.info(`[${proxy.name}] 设置失败缓存`) cache.set(id, {}) } } } // 请求 async function http(opt = {}) { const METHOD = opt.method || $arguments.method || 'get' const TIMEOUT = parseFloat(opt.timeout || $arguments.timeout || 5000) const RETRIES = parseFloat(opt.retries ?? $arguments.retries ?? 1) const RETRY_DELAY = parseFloat(opt.retry_delay ?? $arguments.retry_delay ?? 1000) let count = 0 const fn = async () => { try { return await $.http[METHOD]({ ...opt, timeout: TIMEOUT }) } catch (e) { // $.error(e) if (count < RETRIES) { count++ const delay = RETRY_DELAY * count // $.info(`第 ${count} 次请求失败: ${e.message || e}, 等待 ${delay / 1000}s 后重试`) await $.wait(delay) return await fn() } else { throw e } } } return await fn() } function getCacheId({ proxy = {}, url }) { return `http-meta:gpt:${url}:${JSON.stringify( Object.fromEntries(Object.entries(proxy).filter(([key]) => !/^(name|collectionName|subName|id|_.*)$/i.test(key))) )}` } }