@simonyea/holysheep-cli 1.7.126 → 1.7.128

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@simonyea/holysheep-cli",
-  "version": "1.7.126",
+  "version": "1.7.128",
   "description": "Claude Code/Cursor/Cline API relay for China — ¥1=$1, WeChat/Alipay payment, no credit card, no VPN. One command setup for all AI coding tools.",
   "keywords": [
     "openai-china",

package/src/tools/claude-process-proxy.js

@@ -46,6 +46,26 @@ function getControlPlaneUrl(config) {
 }
 
 const leaseCache = new Map()
+const MAX_PROXY_RETRIES = 3
+
+function createForwardError(statusCode, body) {
+  const error = new Error(`HTTP ${statusCode}: ${String(body || '').slice(0, 200)}`)
+  error.statusCode = Number(statusCode) || 502
+  error.body = String(body || '')
+  return error
+}
+
+function shouldRefreshLeaseAfterError(err) {
+  const statusCode = Number(err?.statusCode || 0)
+  const body = String(err?.body || err?.message || '')
+  if (statusCode === 403 || statusCode === 503) return true
+  return (
+    body.includes('client_validation_error') ||
+    body.includes('Claude Code 必须使用 hs claude 指令启动') ||
+    body.includes('当前代理节点') ||
+    body.includes('No active Claude relay nodes are available')
+  )
+}
 
 function isRetryableNodeLeaseError(err) {
   const message = String(err?.message || '')
@@ -159,7 +179,7 @@ function forwardViaNodeProxy({ nodeProxyUrl, targetUrl, clientReq, clientRes, ex
         forwardRes.on('data', (c) => chunks.push(c))
         forwardRes.on('end', () => {
           const body = Buffer.concat(chunks).toString('utf8')
-          reject(new Error(`HTTP ${status}: ${body.slice(0, 200)}`))
+          reject(createForwardError(status, body))
         })
         return
       }
@@ -200,15 +220,21 @@ function createConnectTunnel(proxyUrl, target, headers) {
   })
 }
 
+// 等上游返回 first byte 的最大时间。Claude 长上下文（>200K tokens）的 prefill 时间
+// 可以达到 30-90 秒，30 秒太严会误杀正常长 prompt。默认 300 秒，可通过环境变量
+// HS_CLAUDE_RESPONSE_TIMEOUT_MS 覆盖。
+const RESPONSE_TIMEOUT_MS = Number(process.env.HS_CLAUDE_RESPONSE_TIMEOUT_MS) || 300000
+// 上游 stream 已开始后，相邻两个 chunk 之间的最大间隔（覆盖 extended thinking 间歇）。
+const STALL_TIMEOUT_MS = Number(process.env.HS_CLAUDE_STALL_TIMEOUT_MS) || 120000
+
 function pipeWithCleanup(a, b) {
   for (const sock of [a, b]) {
     if (typeof sock.setKeepAlive === 'function') sock.setKeepAlive(true, 10000)
   }
 
   // 双阶段超时：
-  // 1. 客户端发了数据，上游 30 秒没回第一个字节 → 断开（node proxy 挂了）
-  // 2. 上游在流数据突然停了 120 秒 → 断开（stream 中断）
-  //    120 秒足够覆盖 extended thinking 的间歇（thinking 期间服务端仍会发心跳）
+  // 1. 客户端发了数据，上游 RESPONSE_TIMEOUT_MS 没回第一个字节 → 断开（node proxy 挂了）
+  // 2. 上游在流数据突然停了 STALL_TIMEOUT_MS → 断开（stream 中断）
   let timer = null
   let streaming = false
   const kill = (reason) => {
@@ -220,15 +246,15 @@ function pipeWithCleanup(a, b) {
     // 客户端发了数据（请求），等上游响应
     if (!streaming) {
       if (timer) clearTimeout(timer)
-      timer = setTimeout(() => kill('upstream response timeout'), 30000)
+      timer = setTimeout(() => kill('upstream response timeout'), RESPONSE_TIMEOUT_MS)
     }
   })
   b.on('data', () => {
     // 上游回数据了，切换到 stream 模式
     streaming = true
-    // 每次收到数据重置 120 秒超时
+    // 每次收到数据重置 stall 超时
     if (timer) clearTimeout(timer)
-    timer = setTimeout(() => kill('upstream stream stalled'), 120000)
+    timer = setTimeout(() => kill('upstream stream stalled'), STALL_TIMEOUT_MS)
   })
 
   a.pipe(b)
@@ -277,29 +303,37 @@ function createProcessProxyServer({ sessionId, configPath = CONFIG_PATH }) {
       })
     }
 
-    try {
-      await doForward(getCachedLease(sessionId))
-    } catch (err) {
-      if (clientRes.headersSent) return
-      const shouldRefreshLease =
-        !leaseCache.has(sessionId) || (err && isRetryableNodeLeaseError(err))
-      if (shouldRefreshLease) {
-        try {
+    let lastError = null
+    for (let attempt = 0; attempt <= MAX_PROXY_RETRIES; attempt++) {
+      try {
+        if (attempt === 0) {
+          await doForward(getCachedLease(sessionId))
+        } else {
           const config = readConfig(configPath)
           leaseCache.delete(sessionId)
+          if (shouldRefreshLeaseAfterError(lastError)) {
+            await closeSession(configPath, sessionId)
+          }
           const freshLease = await fetchFreshLease(config, sessionId)
           await doForward(freshLease)
-        } catch (retryError) {
-          if (!clientRes.headersSent) {
-            clientRes.writeHead(502, { 'content-type': 'text/plain; charset=utf-8' })
-            clientRes.end(retryError.message || 'Proxy error')
-          }
         }
-      } else {
-        clientRes.writeHead(502, { 'content-type': 'text/plain; charset=utf-8' })
-        clientRes.end(err?.message || 'Proxy error')
+        lastError = null
+        break
+      } catch (err) {
+        lastError = err
+        if (clientRes.headersSent) return
+        if (!isRetryableNodeLeaseError(err) && attempt > 0) break
       }
     }
+    if (lastError && !clientRes.headersSent) {
+      const status = Number(lastError.statusCode || 502)
+      const body = String(lastError.body || lastError.message || 'Proxy error')
+      const isJson = body.trim().startsWith('{')
+      clientRes.writeHead(status, {
+        'content-type': isJson ? 'application/json; charset=utf-8' : 'text/plain; charset=utf-8'
+      })
+      clientRes.end(body)
+    }
   })
 
   // Hosts that must NOT be tunneled directly — they must go through CRS via
@@ -333,26 +367,37 @@ function createProcessProxyServer({ sessionId, configPath = CONFIG_PATH }) {
       pipeWithCleanup(clientSocket, upstreamSocket)
     }
 
-    try {
-      await doConnect(getCachedLease(sessionId))
-    } catch (err) {
-      const shouldRefreshLease =
-        !leaseCache.has(sessionId) || (err && isRetryableNodeLeaseError(err))
-      if (shouldRefreshLease) {
-        try {
+    let lastError = null
+    for (let attempt = 0; attempt <= MAX_PROXY_RETRIES; attempt++) {
+      try {
+        if (attempt === 0) {
+          await doConnect(getCachedLease(sessionId))
+        } else {
           const config = readConfig(configPath)
           leaseCache.delete(sessionId)
+          if (shouldRefreshLeaseAfterError(lastError)) {
+            await closeSession(configPath, sessionId)
+          }
           const freshLease = await fetchFreshLease(config, sessionId)
           await doConnect(freshLease)
-        } catch (retryError) {
-          clientSocket.write(`HTTP/1.1 502 Bad Gateway\r\ncontent-type: text/plain; charset=utf-8\r\n\r\n${retryError.message}`)
-          clientSocket.destroy()
         }
-      } else {
-        clientSocket.write(`HTTP/1.1 502 Bad Gateway\r\ncontent-type: text/plain; charset=utf-8\r\n\r\n${err?.message || 'Proxy error'}`)
-        clientSocket.destroy()
+        lastError = null
+        break
+      } catch (err) {
+        lastError = err
+        if (!isRetryableNodeLeaseError(err) && attempt > 0) break
       }
     }
+    if (lastError) {
+      const status = Number(lastError.statusCode || 502)
+      const body = String(lastError.body || lastError.message || 'Proxy error')
+      const statusText = status === 403 ? 'Forbidden' : status === 503 ? 'Service Unavailable' : 'Bad Gateway'
+      const contentType = body.trim().startsWith('{')
+        ? 'application/json; charset=utf-8'
+        : 'text/plain; charset=utf-8'
+      clientSocket.write(`HTTP/1.1 ${status} ${statusText}\r\ncontent-type: ${contentType}\r\n\r\n${body}`)
+      clientSocket.destroy()
+    }
   })
 
   return server