mslxdff 0.1.87 → 0.1.89

package/src/upstream.js

@@ -2,37 +2,24 @@ import crypto from "node:crypto";
 import { mkdirSync } from "node:fs";
 import { appendFile } from "node:fs/promises";
 import { dirname, join } from "node:path";
-import os from "node:os";
+import { performance } from "node:perf_hooks";
 import { isFreeModel } from "./models.js";
-import { defaultStateFile } from "./state.js";
 import { fmtShanghaiYMDHMS } from "./time.js";
-
-let UndiciAgent = null;
-let UndiciFetch = null;
-try {
-  const mod = await import("undici");
-  UndiciAgent = mod.Agent;
-  UndiciFetch = mod.fetch;
-} catch {
-  // undici not installed — fallback to no dispatcher (still works, just no keep-alive tuning)
-}
+import { createTransport } from "./transport/index.js";
 
 function genId(prefix) {
   return `${prefix}${crypto.randomUUID().replace(/-/g, "")}`;
 }
-
 function envInt(name, fallback) {
   const v = Number(process.env[name]);
   return Number.isInteger(v) && v > 0 ? v : fallback;
 }
-
 function isPreheatDisabled() {
   const raw = process.env.MSLXDFF_PREHEAT;
   if (raw === undefined || raw === null || raw === "") return false;
   const s = String(raw).trim().toLowerCase();
   return s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled";
 }
-
 export function createUpstreamClient({
   baseUrl = process.env.UPSTREAM_BASE_URL || "https://opencode.ai",
   authToken = process.env.UPSTREAM_AUTH_TOKEN || "public",
@@ -48,17 +35,6 @@ export function createUpstreamClient({
   hooks,
   keepAlive = true,
 } = {}) {
-  // 使用 undici 的 fetch 与 Agent 配对，避免 global fetch 与 npm undici Agent 不兼容
-  if (!fetchImpl) fetchImpl = UndiciFetch || fetch;
-  // hooks: async (name, ctx) => ({ value, changed }) | null — 由 bin 用 runHook 绑定；错误已在上层隔离
-  const applyHook = async (name, ctx) => {
-    if (!hooks) return null;
-    try {
-      return await hooks(name, ctx);
-    } catch {
-      return null;
-    }
-  };
   function isAnonFirst() {
     const raw = process.env.MSLXDFF_OPENCOD_ANON_FIRST ?? process.env.MSLXDFF_ANON_FIRST ?? process.env.MSLXDFF_OPENCOD_ANON;
     if (raw === undefined || raw === null || raw === "") return authToken === "public";
@@ -70,7 +46,7 @@ export function createUpstreamClient({
   const baseHeaders = anonFirst
     ? {
         "Content-Type": "application/json",
-        "Authorization": "",
+        Authorization: "",
         "x-opencode-client": "desktop",
         "User-Agent": "opencode",
         "HTTP-Referer": "https://hermes-agent.nousresearch.com",
@@ -78,53 +54,40 @@ export function createUpstreamClient({
       }
     : {
         "Content-Type": "application/json",
-        "Authorization": `Bearer ${authToken}`,
+        Authorization: `Bearer ${authToken}`,
         "x-opencode-client": "desktop",
       };
-
   function buildHeaders(body, { anonymous = anonFirst } = {}) {
     const isStream = body?.stream !== false;
     const base = {
       ...baseHeaders,
-      "Accept": isStream ? "text/event-stream" : "*/*",
+      Accept: isStream ? "text/event-stream" : "*/*",
       "User-Agent": "opencode",
       "x-opencode-session": genId("ses_"),
       "x-opencode-request": genId("msg_"),
       "x-opencode-project": "global",
     };
     if (anonymous) {
-      // hermes 的 free 匿名通道：空 Authorization，保留 opencode UA 以兼容 big-pickle/mimo 的 UA 门禁
-      return {
-        ...base,
-        "Authorization": "",
-        "HTTP-Referer": "https://hermes-agent.nousresearch.com",
-        "X-Title": "Hermes Agent",
-      };
+      return { ...base, Authorization: "", "HTTP-Referer": "https://hermes-agent.nousresearch.com", "X-Title": "Hermes Agent" };
     }
-    // 显式 public 回退（anonFirst 时去掉 hermes 头）
     if (anonFirst) {
       const { "HTTP-Referer": _a, "X-Title": _b, ...rest } = base;
-      return { ...rest, "Authorization": `Bearer ${authToken}` };
+      return { ...rest, Authorization: `Bearer ${authToken}` };
     }
     return base;
   }
-
   function shouldTryAnonFree() {
     const raw = process.env.MSLXDFF_FREE_ANON;
     if (raw === "0" || raw === "off" || raw === "false" || raw === "no") return false;
     return true;
   }
-
   function isResponsesModel(model) {
-    const m = String(model || "").toLowerCase();
-    return m.startsWith("muse-spark");
+    return String(model || "").toLowerCase().startsWith("muse-spark");
   }
-
   function chatToResponsesBody(chatBody) {
     const msgs = Array.isArray(chatBody?.messages) ? chatBody.messages : [];
     const system = msgs.filter((m) => m.role === "system").map((m) => String(m.content || "")).join("\n");
     const nonSystem = msgs.filter((m) => m.role !== "system");
-    // 取最后一条 user 内容作为 input，其余拼到 input 前（保留上下文）
     const inputParts = nonSystem.map((m) => {
       const c = m.content;
       if (typeof c === "string") return `${m.role}: ${c}`;
@@ -132,11 +95,7 @@ export function createUpstreamClient({
       return `${m.role}: ${String(c || "")}`;
     });
     const input = inputParts.join("\n\n") || "hi";
-    const out = {
-      model: chatBody.model,
-      input,
-      stream: false, // muse-spark 的 chat 流式经 responses 的 SSE 格式不同，先以非流式 200 保证可用
-    };
+    const out = { model: chatBody.model, input, stream: false };
     if (system) out.instructions = system;
     if (chatBody.tools) out.tools = chatBody.tools;
     if (chatBody.tool_choice) out.tool_choice = chatBody.tool_choice;
@@ -144,9 +103,7 @@ export function createUpstreamClient({
     if (chatBody.max_tokens != null) out.max_output_tokens = chatBody.max_tokens;
     return out;
   }
-
   function responsesToChatJson(respJson) {
-    // responses: {id, model, output:[{type:reasoning},{type:message, content:[{type:output_text,text}]}]}
     let text = "";
     for (const item of respJson.output || []) {
       if (item.type === "message" && item.role === "assistant") {
@@ -157,7 +114,6 @@ export function createUpstreamClient({
       }
     }
     if (!text) {
-      // 兜底：取 output_text 的第一个
       for (const item of respJson.output || []) {
         if (item.type === "message") {
           const t = item.content?.[0]?.text;
@@ -165,288 +121,158 @@ export function createUpstreamClient({
         }
       }
     }
-    if (!text) text = "";
     const chatJson = {
       id: respJson.id || `resp_${Date.now()}`,
       object: "chat.completion",
       created: Math.floor((respJson.created_at || Date.now() / 1000)),
       model: respJson.model,
-      choices: [
-        {
-          index: 0,
-          finish_reason: respJson.status === "completed" ? "stop" : "length",
-          message: { role: "assistant", content: text },
-        },
-      ],
+      choices: [{ index: 0, finish_reason: respJson.status === "completed" ? "stop" : "length", message: { role: "assistant", content: text } }],
       usage: respJson.usage || { prompt_tokens: 0, completion_tokens: 0, total_tokens: 0 },
     };
     return chatJson;
   }
-
   function freeAnonLogFile() {
-    // 按用户要求：放在当前项目目录，方便一眼看到；可用 env 覆盖
     return process.env.MSLXDFF_FREE_ANON_LOG || join(process.cwd(), "free-anon-extra.txt");
   }
-
   function logFreeAnon({ model, publicStatus, anonStatus, anonAttempts, hit, totalMs }) {
     const file = freeAnonLogFile();
     const line = `${fmtShanghaiYMDHMS(new Date())} model=${model} public=429 anonTries=${anonAttempts} anonStatus=${anonStatus ?? "none"} hit=${hit ? "YES额外额度" : "NO无额外"} totalMs=${totalMs} count=${hit ? "1" : "0"}\n`;
-    try {
-      mkdirSync(dirname(file), { recursive: true });
-    } catch {}
+    try { mkdirSync(dirname(file), { recursive: true }); } catch {}
     appendFile(file, line).catch(() => {});
-    // 同时在控制台留痕（daemon 日志也能看到）
-    try {
-      if (process.env.MSLXDFF_DEBUG === "1") console.log(`[free-anon] ${line.trim()}`);
-    } catch {}
+    try { if (process.env.MSLXDFF_DEBUG === "1") console.log(`[free-anon] ${line.trim()}`); } catch {}
   }
-
-  // 连续命中计数（内存，方便 txt 里看出“连续几次都有”）
   let consecutiveHits = 0;
 
-  // Keep-Alive Agent（显式复用 TCP+TLS）
-  const keepAliveTimeout = envInt("MSLXDFF_UPSTREAM_KEEPALIVE_TIMEOUT", 30_000);
-  const keepAliveMaxTimeout = envInt("MSLXDFF_UPSTREAM_KEEPALIVE_MAX_TIMEOUT", 60_000);
-  const keepAliveConnections = envInt("MSLXDFF_UPSTREAM_KEEPALIVE_CONNECTIONS", 20);
-  let dispatcher = null;
-  let agent = null;
-  if (keepAlive && UndiciAgent) {
-    try {
-      agent = new UndiciAgent({
-        keepAliveTimeout,
-        keepAliveMaxTimeout,
-        connections: keepAliveConnections,
-        pipelining: 1,
-      });
-      dispatcher = agent;
-    } catch {
-      agent = null;
-      dispatcher = null;
-    }
-  }
+  // 深模块：网络层委托 StreamingTransport
+  const transport = createTransport({
+    keepAlive,
+    fetchImpl,
+    timeoutMs: connectTimeoutMs,
+    retry,
+    hooks,
+  });
 
   async function chat(body) {
     const isResp = isResponsesModel(body?.model);
     const url = isResp ? `${baseUrl}/zen/v1/responses` : `${baseUrl}/zen/v1/chat/completions`;
     const reqBody = isResp ? chatToResponsesBody(body) : body;
     const t0 = performance.now();
-    const attempts = [];
-    let waitMs = 0;
-    for (let attempt = 0; ; attempt++) {
-      const t = performance.now();
-      const result = await attemptOnce(url, reqBody);
-      attempts.push({
-        attempt,
-        type: result instanceof Error ? "network" : `http${result.status}`,
-        ms: Math.round(performance.now() - t),
+
+    // 首发请求（transport 已处理 network/429 等重试）
+    let res;
+    try {
+      res = await transport.request({
+        url,
+        method: "POST",
+        headers: buildHeaders(reqBody),
+        body: reqBody,
+        stream: body?.stream !== false,
+        timeoutMs: connectTimeoutMs,
+        retry,
       });
-      if (result instanceof Error) {
-        const entry = retry?.network;
-        if (entry && attempt < entry.attempts) {
-          await sleep(entry.delayMs);
-          waitMs += entry.delayMs;
+    } catch (e) {
+      e._t = e._t || { attempts: [], waitMs: 0, totalMs: Math.round(performance.now() - t0) };
+      throw e;
+    }
+
+    // 匿名兜底：仅非 anonFirst 时，public 429 + free 模型才走 hermes 空头重试
+    if (!anonFirst && res.status === 429 && isFreeModel(body?.model) && shouldTryAnonFree()) {
+      const anonRetries = envInt("MSLXDFF_FREE_ANON_RETRIES", 3);
+      const anonDelay = envInt("MSLXDFF_FREE_ANON_DELAY_MS", 1000);
+      let anonRes = null;
+      for (let i = 0; i < anonRetries; i++) {
+        await new Promise((r) => setTimeout(r, anonDelay));
+        try {
+          anonRes = await transport.request({
+            url,
+            method: "POST",
+            headers: buildHeaders(reqBody, { anonymous: true }),
+            body: reqBody,
+            stream: body?.stream !== false,
+            timeoutMs: connectTimeoutMs,
+            retry,
+          });
+        } catch (e) {
           continue;
         }
-        result._t = {
-          attempts,
-          waitMs,
-          totalMs: Math.round(performance.now() - t0),
-        };
-        throw result;
-      }
-      const entry = retry?.[result.status];
-      if (entry && attempt < entry.attempts) {
-        await sleep(entry.delayMs);
-        waitMs += entry.delayMs;
-        continue;
-      }
-      // 额外额度探测：仅当非 anon 优先时，public 429 且为 free 模型时，用空头（hermes 方式）每秒重试 3 次；anon 优先时首发即高额度，无需再 public 撞墙
-      if (!anonFirst && result.status === 429 && isFreeModel(body?.model) && shouldTryAnonFree()) {
-        const anonRetries = envInt("MSLXDFF_FREE_ANON_RETRIES", 3);
-        const anonDelay = envInt("MSLXDFF_FREE_ANON_DELAY_MS", 1000);
-        let anonResult = null;
-        let hit = false;
-        let hitAt = -1;
-        for (let i = 0; i < anonRetries; i++) {
-          await sleep(anonDelay);
-          waitMs += anonDelay;
-          const t2 = performance.now();
-          anonResult = await attemptOnce(url, reqBody, { anonymous: true });
-          attempts.push({
-            attempt: `anon-${i}`,
-            type: anonResult instanceof Error ? "network" : `http${anonResult.status}`,
-            ms: Math.round(performance.now() - t2),
-          });
-          if (anonResult instanceof Error) continue;
-          if (anonResult.status !== 429) {
-            // responses 模型的 anon 命中也需转回 chat 形状
-            let outAnon = anonResult;
-            if (isResp && anonResult.ok) {
-              try {
-                const txt = await anonResult.text();
-                const j = JSON.parse(txt);
-                if (j && Array.isArray(j.output)) {
-                  const chatJson = responsesToChatJson(j);
-                  const newHeaders = new Headers(anonResult.headers);
-                  newHeaders.set("content-type", "application/json");
-                  outAnon = new Response(JSON.stringify(chatJson), { status: anonResult.status, headers: newHeaders });
-                } else {
-                  outAnon = new Response(txt, { status: anonResult.status, headers: anonResult.headers });
-                }
-              } catch { outAnon = anonResult; }
-            }
-            outAnon._t = {
-              attempts,
-              waitMs,
-              totalMs: Math.round(performance.now() - t0),
-              anonTried: true,
-              anonAttempts: i + 1,
-            };
-            hit = true;
-            hitAt = i + 1;
-            consecutiveHits += 1;
-            logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonResult.status, anonAttempts: i + 1, hit: true, totalMs: outAnon._t.totalMs });
-            // 额外在 txt 里强调连续命中
+        if (anonRes.status !== 429) {
+          // responses 模型需转回 chat 形状
+          let outAnon = anonRes;
+          if (isResp && anonRes.ok) {
             try {
-              if (consecutiveHits >= 2) {
-                const file = freeAnonLogFile();
-                appendFile(file, `  -> 连续额外额度 ${consecutiveHits} 次\n`).catch(() => {});
+              const txt = await anonRes.text();
+              const j = JSON.parse(txt);
+              if (j && Array.isArray(j.output)) {
+                const chatJson = responsesToChatJson(j);
+                outAnon = new Response(JSON.stringify(chatJson), { status: anonRes.status, headers: new Headers(anonRes.headers) });
+                outAnon.headers.set("content-type", "application/json");
+              } else {
+                outAnon = new Response(txt, { status: anonRes.status, headers: anonRes.headers });
               }
-            } catch {}
-            if (process.env.MSLXDFF_DEBUG === "1") try { console.log(`[free-anon] ${body?.model} public 429 -> anon ${anonResult.status} after ${i + 1} try HIT`); } catch {}
-            return outAnon;
+            } catch { outAnon = anonRes; }
           }
-        }
-        // 3 次 anon 仍 429
-        if (anonResult) {
-          anonResult._t = {
-            attempts,
-            waitMs,
-            totalMs: Math.round(performance.now() - t0),
-            anonTried: true,
-            anonAttempts: anonRetries,
-          };
-          consecutiveHits = 0;
-          logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonResult.status, anonAttempts: anonRetries, hit: false, totalMs: anonResult._t.totalMs });
-          if (process.env.MSLXDFF_DEBUG === "1") try { console.log(`[free-anon] ${body?.model} public 429 -> anon 429 x${anonRetries} MISS`); } catch {}
-          return anonResult;
-        }
-      }
-      // Muse Spark 走 /responses，需把 responses 的 JSON 转回 chat.completions 形状，保持上层无感
-      if (isResp && result.ok) {
-        try {
-          const txt = await result.text();
-          const j = JSON.parse(txt);
-          // responses 成功态：{output:[...]}；chat 态：{choices:[...]}。非 responses 形状则原样回
-          if (j && Array.isArray(j.output)) {
-            const chatJson = responsesToChatJson(j);
-            const newBody = JSON.stringify(chatJson);
-            const newHeaders = new Headers(result.headers);
-            newHeaders.set("content-type", "application/json");
-            const transformed = new Response(newBody, { status: result.status, headers: newHeaders });
-            transformed._t = {
-              attempts,
-              waitMs,
-              totalMs: Math.round(performance.now() - t0),
-            };
-            // 保留原始 timing 供日志
-            if (result._t) transformed._t = { ...transformed._t, ...result._t };
-            return transformed;
+          outAnon._t = { ...(outAnon._t || {}), anonTried: true, anonAttempts: i + 1, totalMs: Math.round(performance.now() - t0) };
+          consecutiveHits += 1;
+          logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: i + 1, hit: true, totalMs: outAnon._t.totalMs });
+          if (consecutiveHits >= 2) {
+            try { appendFile(freeAnonLogFile(), `  -> 连续额外额度 ${consecutiveHits} 次\n`).catch(() => {}); } catch {}
           }
-          // 不是 responses 形状，回退：用原文本重建可读 Response
-          const fallback = new Response(txt, { status: result.status, headers: result.headers });
-          fallback._t = { attempts, waitMs, totalMs: Math.round(performance.now() - t0) };
-          return fallback;
-        } catch {
-          result._t = { attempts, waitMs, totalMs: Math.round(performance.now() - t0) };
-          return result;
+          return outAnon;
         }
       }
-      result._t = {
-        attempts,
-        waitMs,
-        totalMs: Math.round(performance.now() - t0),
-      };
-      return result;
+      if (anonRes) {
+        anonRes._t = { ...(anonRes._t || {}), anonTried: true, anonAttempts: anonRetries, totalMs: Math.round(performance.now() - t0) };
+        consecutiveHits = 0;
+        logFreeAnon({ model: body?.model, publicStatus: 429, anonStatus: anonRes.status, anonAttempts: anonRetries, hit: false, totalMs: anonRes._t.totalMs });
+        return anonRes;
+      }
     }
-  }
 
-  async function attemptOnce(url, body, { anonymous = anonFirst } = {}) {
-    const controller = new AbortController();
-    const timer = setTimeout(() =>
-      controller.abort(new Error(`upstream timed out after ${connectTimeoutMs}ms`)),
-      connectTimeoutMs
-    );
-    try {
-      let reqUrl = url;
-      let headers = buildHeaders(body, { anonymous });
-      // 插件 hook：upstream:headers — 返回 { headers } 可替换请求头
-      const hh = await applyHook("upstream:headers", { url, body, headers });
-      if (hh?.changed && hh.value?.headers) headers = hh.value.headers;
-      // 插件 hook：upstream:before-request — 返回 { url, headers } 可改目标地址/头（上游不限于 opencode）
-      const br = await applyHook("upstream:before-request", { url, method: "POST", body, headers });
-      if (br?.changed && br.value) {
-        if (typeof br.value.url === "string" && br.value.url) reqUrl = br.value.url;
-        if (br.value.headers && typeof br.value.headers === "object") headers = br.value.headers;
+    // responses 模型成功态转 chat
+    if (isResp && res.ok) {
+      try {
+        const txt = await res.text();
+        const j = JSON.parse(txt);
+        if (j && Array.isArray(j.output)) {
+          const chatJson = responsesToChatJson(j);
+          const headers = new Headers(res.headers);
+          headers.set("content-type", "application/json");
+          const transformed = new Response(JSON.stringify(chatJson), { status: res.status, headers });
+          transformed._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
+          return transformed;
+        }
+        const fallback = new Response(txt, { status: res.status, headers: res.headers });
+        fallback._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
+        return fallback;
+      } catch {
+        res._t = { ...(res._t || {}), totalMs: Math.round(performance.now() - t0) };
+        return res;
       }
-      const opts = {
-        method: "POST",
-        headers,
-        body: JSON.stringify(body),
-        signal: controller.signal,
-      };
-      if (dispatcher) opts.dispatcher = dispatcher;
-      const res = await fetchImpl(reqUrl, opts);
-      return res;
-    } catch (err) {
-      return err;
-    } finally {
-      clearTimeout(timer);
     }
+    res._t = { ...(res._t || {}), totalMs: res._t?.totalMs ?? Math.round(performance.now() - t0) };
+    return res;
   }
 
   async function preheat() {
     if (isPreheatDisabled()) return { ok: false, skipped: true, reason: "disabled" };
-    const url = `${baseUrl}/zen/v1/models`;
-    const controller = new AbortController();
-    const timer = setTimeout(() => controller.abort(new Error("preheat timed out after 3000ms")), 3000);
     const t0 = performance.now();
     try {
-      const headers = buildHeaders({ stream: false });
-      const opts = {
+      const res = await transport.request({
+        url: `${baseUrl}/zen/v1/models`,
         method: "GET",
-        headers,
-        signal: controller.signal,
-      };
-      if (dispatcher) opts.dispatcher = dispatcher;
-      const res = await fetchImpl(url, opts);
-      const ms = Math.round(performance.now() - t0);
-      // 消耗 body 以释放连接回池（即使不需要内容）
-      try { if (res.body) await res.text().catch(() => {}); } catch {}
-      return { ok: res.ok, status: res.status, ms };
-    } catch (err) {
-      return { ok: false, error: String(err?.message || err), ms: Math.round(performance.now() - t0) };
-    } finally {
-      clearTimeout(timer);
+        headers: buildHeaders({ stream: false }),
+        stream: false,
+        timeoutMs: 3000,
+      });
+      try { await res.text().catch(() => {}); } catch {}
+      return { ok: res.ok, status: res.status, ms: Math.round(performance.now() - t0) };
+    } catch (e) {
+      return { ok: false, error: String(e?.message || e), ms: Math.round(performance.now() - t0) };
     }
   }
-
-  let closed = false;
   async function close() {
-    if (closed) return;
-    closed = true;
-    if (agent && typeof agent.close === "function") {
-      try { await agent.close(); } catch {}
-    } else if (dispatcher && typeof dispatcher.close === "function" && dispatcher !== agent) {
-      try { await dispatcher.close(); } catch {}
-    }
+    await transport.close();
   }
-
-  // 兼容旧调用：headers 为动态生成，暴露 getter 快照（用于测试/展示）
   const headers = buildHeaders({});
-  return { chat, preheat, close, headers, buildHeaders, dispatcher, agent, [Symbol.asyncDispose]: close };
+  return { chat, preheat, close, headers, buildHeaders, get dispatcher() { return transport.dispatcher; }, get agent() { return transport.agent; }, [Symbol.asyncDispose]: close };
 }
-
-function sleep(ms) {
-  return new Promise((r) => setTimeout(r, ms));
-}