mslxdff 0.1.233 → 0.1.235

package/src/routes/chat/gateway.js

@@ -1,47 +1,47 @@
-import { clientIp, json, readBody, parseHops } from "../helpers.js";
-import { runHook } from "../../plugins.js";
-import { createChatPipeline } from "../../chat-pipeline/index.js";
-
-/**
- * ChatGateway 薄适配层：仅负责 HTTP I/O（readBody + request:received hook）→ 委托 ChatPipeline.execute
- * Policy/Order/AutoRace/hedge/peer/broadband 全部在 chat-pipeline 深模块内。
- */
-export function createChatGateway(deps = {}) {
-  const pipeline = createChatPipeline(deps);
-
-  async function handle({ req, res }) {
-    let body;
-    try {
-      body = await readBody(req);
-    } catch {
-      return json(res, 400, { error: "Invalid JSON body" });
-    }
-    if (deps.plugins?.length) {
-      const rc = await runHook(deps.plugins, "request:received", {
-        ip: clientIp(req),
-        hops: parseHops(req.headers["x-mslxdff-hops"]),
-        headers: { "content-type": req.headers["content-type"] },
-        body,
-      });
-      for (const e of rc.errors) deps.logs?.appendEvent?.({ ts: Date.now(), type: "plugin-hook-error", hook: "request:received", plugin: e.plugin, error: e.error });
-      const respond = rc.value?.respond;
-      if (respond && typeof respond === "object") return json(res, respond.status || 200, respond.body ?? {});
-    }
-    req.body = body;
-    try {
-      await pipeline.execute({ req, res });
-    } catch (err) {
-      const msg = err?.message || String(err);
-      if (!res.headersSent) return json(res, 502, { error: msg });
-      try { res.end(); } catch {}
-    }
-  }
-
-  return { handle };
-}
-
-// 薄适配：保持原 chatHandler 签名兼容
-export async function chatHandler(ctx) {
-  const gw = createChatGateway(ctx);
-  return gw.handle(ctx);
+import { clientIp, json, readBody, parseHops } from "../helpers.js";
+import { runHook } from "../../plugins.js";
+import { createChatPipeline } from "../../chat-pipeline/index.js";
+
+/**
+ * ChatGateway 薄适配层：仅负责 HTTP I/O（readBody + request:received hook）→ 委托 ChatPipeline.execute
+ * Policy/Order/AutoRace/hedge/peer/broadband 全部在 chat-pipeline 深模块内。
+ */
+export function createChatGateway(deps = {}) {
+  const pipeline = createChatPipeline(deps);
+
+  async function handle({ req, res }) {
+    let body;
+    try {
+      body = await readBody(req);
+    } catch {
+      return json(res, 400, { error: "Invalid JSON body" });
+    }
+    if (deps.plugins?.length) {
+      const rc = await runHook(deps.plugins, "request:received", {
+        ip: clientIp(req),
+        hops: parseHops(req.headers["x-mslxdff-hops"]),
+        headers: { "content-type": req.headers["content-type"] },
+        body,
+      });
+      for (const e of rc.errors) deps.logs?.appendEvent?.({ ts: Date.now(), type: "plugin-hook-error", hook: "request:received", plugin: e.plugin, error: e.error });
+      const respond = rc.value?.respond;
+      if (respond && typeof respond === "object") return json(res, respond.status || 200, respond.body ?? {});
+    }
+    req.body = body;
+    try {
+      await pipeline.execute({ req, res });
+    } catch (err) {
+      const msg = err?.message || String(err);
+      if (!res.headersSent) return json(res, 502, { error: msg });
+      try { res.end(); } catch {}
+    }
+  }
+
+  return { handle };
+}
+
+// 薄适配：保持原 chatHandler 签名兼容
+export async function chatHandler(ctx) {
+  const gw = createChatGateway(ctx);
+  return gw.handle(ctx);
 }

package/src/routes/chat/peer-handler.js

@@ -30,10 +30,20 @@ export async function handlePeerRelay({
   const peerErrors = [];
   // 三批共享同一 peer 阶段总预算（同 hedge：旧行为是每批各吃满一次等头超时，串行叠加到 90s+）
   const pctx = peerStageCtx({ ...handlerCtx, peerErrors });
-  const win =
-    (await race(peers.ordered(), pctx)) ||
-    (await race(peers.orderedByLastError(), pctx)) ||
-    (await race(peers.coolingByLastError(), pctx));
+  // singlePeer：指定模型 5xx 重试耗尽后只换 1 个节点，该节点仍失败就交回上层收场。
+  // 选人仍按「组长第一 → 延迟升序」，只是取链头一个；三批候选不叠加，也不进 IPv6 阶段。
+  const singlePeer = handlerCtx?.singlePeer === true;
+  let win = null;
+  if (singlePeer) {
+    pctx.peerMaxPeers = 1;
+    const pick = peers.ordered()[0] || peers.orderedByLastError?.()?.[0] || peers.coolingByLastError?.()?.[0] || null;
+    win = pick ? await race([pick], pctx) : null;
+  } else {
+    win =
+      (await race(peers.ordered(), pctx)) ||
+      (await race(peers.orderedByLastError(), pctx)) ||
+      (await race(peers.coolingByLastError(), pctx));
+  }
   if (!win) {
     evt("peer-race-lose", { reqId: handlerCtx.reqId, model });
     // 组员全失败：把每个组员的真实返回汇总给调用者（否则只剩一个无信息的 429/502）