dsh-router-core 0.3.18 → 0.3.20

package/lib/index.js

@@ -1235,29 +1235,43 @@ function tapStreamUsage(source, startedAt) {
 			outputChars += frameText(obj).length;
 		}
 	};
+	/** 源 reader：cancel 时要靠它把取消传下去（见下方 cancel 分支）。 */
+	let reader;
 	return {
-		stream: new ReadableStream({ start: async (ctrl) => {
-			const reader = source.getReader();
-			try {
-				for (;;) {
-					const { done, value } = await reader.read();
-					if (done) break;
-					if (firstAt === 0) firstAt = Date.now();
-					buf += dec.decode(value, { stream: true });
-					let sep;
-					while ((sep = buf.indexOf("\n\n")) >= 0) {
-						const frame = buf.slice(0, sep + 2);
-						buf = buf.slice(sep + 2);
-						parseFrame(frame);
+		stream: new ReadableStream({
+			start: async (ctrl) => {
+				reader = source.getReader();
+				try {
+					for (;;) {
+						const { done, value } = await reader.read();
+						if (done) break;
+						if (firstAt === 0) firstAt = Date.now();
+						buf += dec.decode(value, { stream: true });
+						let sep;
+						while ((sep = buf.indexOf("\n\n")) >= 0) {
+							const frame = buf.slice(0, sep + 2);
+							buf = buf.slice(sep + 2);
+							parseFrame(frame);
+						}
+						ctrl.enqueue(value);
 					}
-					ctrl.enqueue(value);
+					if (buf !== "") parseFrame(buf);
+				} catch {} finally {
+					try {
+						reader.releaseLock();
+					} catch {}
 				}
-				if (buf !== "") parseFrame(buf);
-			} finally {
-				reader.releaseLock();
-			}
-			ctrl.close();
-		} }),
+				try {
+					ctrl.close();
+				} catch {}
+			},
+			/**
+			* 取消要往下传（同 normalizeSSEStream 的理由）：不传 → 客户端断开时
+			* 上游 fetch body 永不 cancel，连接与 in-flight 请求一直挂着。
+			* 必须用 reader.cancel()：流已被 getReader() 锁住，stream.cancel() 会抛。
+			*/
+			cancel: (reason) => reader?.cancel(reason)
+		}),
 		done: async () => ({
 			usage: merged,
 			outputChars,
@@ -1399,59 +1413,65 @@ async function aggregateSSE(stream) {
 	const calls = /* @__PURE__ */ new Map();
 	const reader = stream.getReader();
 	let buf = "";
-	for (;;) {
-		const { done, value } = await reader.read();
-		if (done) break;
-		buf += dec.decode(value, { stream: true });
-		let sep;
-		while ((sep = buf.indexOf("\n\n")) >= 0) {
-			const frame = buf.slice(0, sep);
-			buf = buf.slice(sep + 2);
-			for (const line of frame.split("\n")) {
-				if (!line.startsWith("data:")) continue;
-				const payload = line.slice(5).trim();
-				if (payload === "" || payload === "[DONE]") continue;
-				let obj;
-				try {
-					obj = JSON.parse(payload);
-				} catch {
-					continue;
-				}
-				if (typeof obj.id === "string" && obj.id !== "") id = obj.id;
-				if (typeof obj.model === "string" && obj.model !== "") model = obj.model;
-				if (typeof obj.created === "number") created = obj.created;
-				if (obj.usage !== void 0) usage = obj.usage;
-				const ch = obj.choices?.[0];
-				if (ch === void 0) continue;
-				if (typeof ch.finish_reason === "string") finish = ch.finish_reason;
-				const delta = ch.delta;
-				if (delta === void 0) {
-					const msg = ch.message;
-					if (msg !== void 0) {
-						if (typeof msg.content === "string") content += msg.content;
-						if (typeof msg.reasoning_content === "string") reasoning += msg.reasoning_content;
+	try {
+		for (;;) {
+			const { done, value } = await reader.read();
+			if (done) break;
+			buf += dec.decode(value, { stream: true });
+			let sep;
+			while ((sep = buf.indexOf("\n\n")) >= 0) {
+				const frame = buf.slice(0, sep);
+				buf = buf.slice(sep + 2);
+				for (const line of frame.split("\n")) {
+					if (!line.startsWith("data:")) continue;
+					const payload = line.slice(5).trim();
+					if (payload === "" || payload === "[DONE]") continue;
+					let obj;
+					try {
+						obj = JSON.parse(payload);
+					} catch {
+						continue;
+					}
+					if (typeof obj.id === "string" && obj.id !== "") id = obj.id;
+					if (typeof obj.model === "string" && obj.model !== "") model = obj.model;
+					if (typeof obj.created === "number") created = obj.created;
+					if (obj.usage !== void 0) usage = obj.usage;
+					const ch = obj.choices?.[0];
+					if (ch === void 0) continue;
+					if (typeof ch.finish_reason === "string") finish = ch.finish_reason;
+					const delta = ch.delta;
+					if (delta === void 0) {
+						const msg = ch.message;
+						if (msg !== void 0) {
+							if (typeof msg.content === "string") content += msg.content;
+							if (typeof msg.reasoning_content === "string") reasoning += msg.reasoning_content;
+						}
+						continue;
+					}
+					if (typeof delta.content === "string") content += delta.content;
+					if (typeof delta.reasoning_content === "string") reasoning += delta.reasoning_content;
+					const tcs = delta.tool_calls;
+					if (!Array.isArray(tcs)) continue;
+					for (const tc of tcs) {
+						const idx = typeof tc.index === "number" ? tc.index : 0;
+						const fn = tc.function;
+						const prev = calls.get(idx) ?? {
+							name: "",
+							args: ""
+						};
+						if (typeof tc.id === "string" && tc.id !== "") prev.id = tc.id;
+						if (typeof tc.type === "string") prev.type = tc.type;
+						if (typeof fn?.name === "string" && fn.name !== "") prev.name = fn.name;
+						if (typeof fn?.arguments === "string") prev.args += fn.arguments;
+						calls.set(idx, prev);
 					}
-					continue;
-				}
-				if (typeof delta.content === "string") content += delta.content;
-				if (typeof delta.reasoning_content === "string") reasoning += delta.reasoning_content;
-				const tcs = delta.tool_calls;
-				if (!Array.isArray(tcs)) continue;
-				for (const tc of tcs) {
-					const idx = typeof tc.index === "number" ? tc.index : 0;
-					const fn = tc.function;
-					const prev = calls.get(idx) ?? {
-						name: "",
-						args: ""
-					};
-					if (typeof tc.id === "string" && tc.id !== "") prev.id = tc.id;
-					if (typeof tc.type === "string") prev.type = tc.type;
-					if (typeof fn?.name === "string" && fn.name !== "") prev.name = fn.name;
-					if (typeof fn?.arguments === "string") prev.args += fn.arguments;
-					calls.set(idx, prev);
 				}
 			}
 		}
+	} finally {
+		try {
+			reader.releaseLock();
+		} catch {}
 	}
 	const toolCalls = [...calls.entries()].sort((a, b) => a[0] - b[0]).filter(([, c]) => c.name !== "").map(([, c]) => ({
 		id: c.id ?? `call_${c.name}`,
@@ -1505,31 +1525,47 @@ function normalizeSSEStream(stream) {
 	const dec = new TextDecoder();
 	let buf = "";
 	let sawDone = false;
+	/** 上游 reader：cancel 时要靠它把取消传下去（见下方 cancel 分支）。 */
+	let reader;
 	return {
-		stream: new ReadableStream({ start: async (ctrl) => {
-			const reader = stream.getReader();
-			try {
-				for (;;) {
-					const { done, value } = await reader.read();
-					if (done) break;
-					buf += dec.decode(value, { stream: true });
-					let sep;
-					while ((sep = buf.indexOf("\n\n")) >= 0) {
-						const frame = buf.slice(0, sep + 2);
-						buf = buf.slice(sep + 2);
-						if (isDoneFrame(frame)) sawDone = true;
-						ctrl.enqueue(enc.encode(fixFrame(frame)));
+		stream: new ReadableStream({
+			start: async (ctrl) => {
+				reader = stream.getReader();
+				try {
+					for (;;) {
+						const { done, value } = await reader.read();
+						if (done) break;
+						buf += dec.decode(value, { stream: true });
+						let sep;
+						while ((sep = buf.indexOf("\n\n")) >= 0) {
+							const frame = buf.slice(0, sep + 2);
+							buf = buf.slice(sep + 2);
+							if (isDoneFrame(frame)) sawDone = true;
+							ctrl.enqueue(enc.encode(fixFrame(frame)));
+						}
 					}
+					if (buf !== "") {
+						if (isDoneFrame(buf)) sawDone = true;
+						ctrl.enqueue(enc.encode(fixFrame(buf)));
+					}
+				} catch {} finally {
+					try {
+						reader.releaseLock();
+					} catch {}
 				}
-				if (buf !== "") {
-					if (isDoneFrame(buf)) sawDone = true;
-					ctrl.enqueue(enc.encode(fixFrame(buf)));
-				}
-			} finally {
-				reader.releaseLock();
-			}
-			ctrl.close();
-		} }),
+				try {
+					ctrl.close();
+				} catch {}
+			},
+			/**
+			* **取消必须往下传**：ReadableStream 的取消不会自动穿过 wrapper。
+			* 不实现这个分支，客户端断开的取消就停在这里，上游 fetch body 永不 cancel
+			* ——那条 TCP 连接和 in-flight 请求原样挂着（长跑一天就会累积到耗尽）。
+			* 用 reader.cancel() 而不是 stream.cancel()：流已被 getReader() 锁住，
+			* 直接 cancel 会抛 `Invalid state: ReadableStream is locked`。
+			*/
+			cancel: (reason) => reader?.cancel(reason)
+		}),
 		sawDone: () => sawDone
 	};
 }
@@ -1599,6 +1635,7 @@ async function writeChatResult(res, r, wantsStream, probe, startedAt = 0) {
 		try {
 			body = await aggregateSSE(r.stream);
 		} catch {
+			await r.stream.cancel().catch(() => {});
 			return false;
 		}
 		if (probe !== void 0) probe.tokens = usageFromJsonBody(body);
@@ -1613,6 +1650,18 @@ async function writeChatResult(res, r, wantsStream, probe, startedAt = 0) {
 	let wroteAny = false;
 	/** 响应已死（客户端断开/res 已销毁）：等 drain 的写入要立刻放弃，别吊死。 */
 	let dead = false;
+	/**
+	* 唤醒「等 drain 的挂起 write」。客户端断开时必须调它：那时 'drain' 永不到来，
+	* 不唤醒就永远卡在 write 里，pipeTo 不 settle，finally 不执行，
+	* **上游流也就永不 cancel**（= 每次客户端放弃漏一条连接）。
+	*/
+	let releaseWait;
+	/**
+	* 中止 pipeTo 的信号源（客户端断开时 abort）。
+	* 单靠 `dead` 标志只能让**下一次** write 立刻返回，唤醒不了**已经挂起**的那次；
+	* abort 才能让 pipeTo 立刻收尾并走它的取消路径（进而 cancel 上游）。
+	*/
+	const abort = new AbortController();
 	const writeChunk = (chunk) => new Promise((resolve) => {
 		if (dead) {
 			resolve();
@@ -1635,6 +1684,11 @@ async function writeChatResult(res, r, wantsStream, probe, startedAt = 0) {
 				resolve();
 			};
 			res.once("drain", onDrain);
+			/** 让这个挂起的 write 立刻收尾（客户端断开时调用）。 */
+			releaseWait = () => {
+				res.removeListener("drain", onDrain);
+				resolve();
+			};
 		}
 	});
 	const onError = () => {
@@ -1642,13 +1696,28 @@ async function writeChatResult(res, r, wantsStream, probe, startedAt = 0) {
 		res.destroy();
 	};
 	res.once("error", onError);
+	/**
+	* **客户端断开必须被当成取消信号**：只监听 'error' 是不够的 —— 对端关闭
+	* （pi-ai 的流空闲超时、用户取消、消费方提前退出）触发的是 'close'，
+	* 不是 'error'。此时若不收手，pipeTo 会一直等着 drain（永不到来），
+	* finally 永不执行，上游 fetch body **永不 cancel** → 那条 TCP 连接和
+	* in-flight 请求原样挂着，长跑累积到耗尽（"用久了就报 connection error"）。
+	*/
+	const onClose = () => {
+		if (res.writableEnded) return;
+		dead = true;
+		releaseWait?.();
+		abort.abort();
+	};
+	res.once("close", onClose);
 	const norm = normalizeSSEStream(tapped?.stream ?? r.stream);
 	try {
-		await norm.stream.pipeTo(new WritableStream({ write: writeChunk }));
+		await norm.stream.pipeTo(new WritableStream({ write: writeChunk }), { signal: abort.signal });
 	} catch {} finally {
 		res.removeListener("error", onError);
+		res.removeListener("close", onClose);
 		if (wroteAny && !norm.sawDone() && !dead && !res.writableEnded && !res.destroyed) await writeChunk(DONE_FRAME);
-		if (wroteAny) res.end();
+		if (wroteAny && !dead) res.end();
 		if (probe !== void 0 && tapped !== void 0) {
 			const got = await tapped.done();
 			probe.tokens = got.usage;
@@ -2315,6 +2384,16 @@ var Router = class {
 */
 /** tool-result 里图片挂到 user 消息时用的说明文字（与官方适配器一致）。 */
 const TOOL_RESULT_IMAGE_TEXT = "Tool result images";
+/**
+* 流空闲超时：上游连上后这么久没吐**任何**字节就判失败（抛 TIMEOUT，可重试）。
+*
+* 对齐 `dsh-llm-pi-ai` 的 `DEFAULT_STREAM_IDLE_TIMEOUT_MS = 300_000`：同一个
+* dsh-router `/v1`，走自定义供应商（pi-ai）时有这个兜底、走内置 adapter 时没有
+* —— 差异导致「上游停摆」在内置路径上永久挂住。
+*
+* 只守空闲、不封总时长：长生成只要持续吐块就不该被打断。
+*/
+const STREAM_IDLE_TIMEOUT_MS = 3e5;
 async function wireMessages(options, system, attachments) {
 	const out = [];
 	/**
@@ -2756,11 +2835,18 @@ var RouterAdapter = class RouterAdapter extends LlmAdapter {
 	baseURL;
 	source;
 	resolveAttachments;
-	constructor(baseURL, source, resolveAttachments) {
+	/** 流空闲超时（ms）：上游这么久不吐字节就判 TIMEOUT。 */
+	idleTimeoutMs;
+	/**
+	* @param idleTimeoutMs 流空闲超时；缺省用生产值（300s，对齐 pi-ai）。
+	*   可注入只为测试能压到几百毫秒，生产不传。
+	*/
+	constructor(baseURL, source, resolveAttachments, idleTimeoutMs = STREAM_IDLE_TIMEOUT_MS) {
 		super();
 		this.baseURL = baseURL;
 		this.source = source;
 		this.resolveAttachments = resolveAttachments ?? (() => void 0);
+		this.idleTimeoutMs = idleTimeoutMs;
 	}
 	providerInfo(provider) {
 		return {
@@ -2824,16 +2910,19 @@ var RouterAdapter = class RouterAdapter extends LlmAdapter {
 				signal: controller.signal
 			});
 		} catch (error) {
+			options.signal?.removeEventListener("abort", onAbort);
 			throw new LlmError(`dsh-router upstream call failed: ${error.message}`, "TRANSPORT", { cause: error });
+		}
+		try {
+			if (!resp.ok) {
+				const detail = await resp.text().catch(() => "");
+				throw new LlmError(`dsh-router /v1 returned ${resp.status}: ${detail.slice(0, 200)}`, "TRANSPORT");
+			}
+			if (resp.body === null) return;
+			yield* translateSse(ssePayloads(resp.body, this.idleTimeoutMs));
 		} finally {
 			options.signal?.removeEventListener("abort", onAbort);
 		}
-		if (!resp.ok) {
-			const detail = await resp.text().catch(() => "");
-			throw new LlmError(`dsh-router /v1 returned ${resp.status}: ${detail.slice(0, 200)}`, "TRANSPORT");
-		}
-		if (resp.body === null) return;
-		yield* translateSse(ssePayloads(resp.body));
 	}
 };
 /**
@@ -2861,16 +2950,41 @@ function looksComplete(payload) {
 * 跨块边界：一个 `data:` 行可能横跨两次 read()，所以留 `buffer` 接住
 * 最后一段未完整的行（9router 的 pipeTransformedSSE 同样处理）。
 */
-async function* ssePayloads(body) {
+async function* ssePayloads(body, idleTimeoutMs) {
 	const reader = body.getReader();
 	const decoder = new TextDecoder("utf-8", { fatal: false });
 	let buffer = "";
 	/** 攒一帧内的多行 data（SSE 规范：多行用 \n 连起来）。**必须**是局部
 	*  状态 —— 放模块级会让并发请求互相串数据。 */
 	let payload = "";
+	/**
+	* 带**空闲超时**的读：上游连上后若长时间一个字节都不吐，就读穿不了 ——
+	* 没有超时的话这里会永久挂着，连接和 in-flight 请求都收不回。
+	*
+	* 为什么必须有（issue #6 的另一半）：`dsh-llm-pi-ai`（自定义供应商走的那条
+	* openai-completions adapter）有 `idleWatchdog`，默认 **300s** 空闲即
+	* `LLM_STREAM_IDLE_TIMEOUT` 中止；而本 adapter 此前**没有任何超时**，只靠
+	* 调用方的 `options.signal`。于是同一个组合，走内置 Router 时上游一停摆就
+	* 永久挂住（占用连接、且该请求永不收尾），走自定义供应商时 300s 后被放弃
+	* —— 放弃又会触发服务端那条「客户端断开不回收」的泄漏路径，两边叠加就是
+	* 「用久了报 connection error」。
+	*
+	* 只守**空闲**、不设总时长：长生成可以跑很久，只要还在吐块就不该被打断
+	* （总时长封顶会让长回答写到一半被砍）。
+	*/
+	const readWithIdleTimeout = () => new Promise((resolve, reject) => {
+		const timer = setTimeout(() => reject(new LlmError(`dsh-router upstream stream idle for ${idleTimeoutMs}ms`, "TIMEOUT")), idleTimeoutMs);
+		reader.read().then((r) => {
+			clearTimeout(timer);
+			resolve(r);
+		}, (e) => {
+			clearTimeout(timer);
+			reject(e);
+		});
+	});
 	try {
 		for (;;) {
-			const { done, value } = await reader.read();
+			const { done, value } = await readWithIdleTimeout();
 			if (done) break;
 			buffer += decoder.decode(value, { stream: true });
 			const lines = buffer.split("\n");
@@ -4210,6 +4324,7 @@ function apply(rawContext) {
 					id: e.id,
 					name: e.name ?? e.id,
 					...e.description !== void 0 ? { description: e.description } : {},
+					...e.icon !== void 0 ? { icon: e.icon } : {},
 					enabled: extStore.isEnabled(e.id),
 					ready: st?.ready === true,
 					...st?.detail !== void 0 ? { detail: st.detail } : {}
@@ -4217,6 +4332,32 @@ function apply(rawContext) {
 			}).filter((e) => e !== void 0)
 		});
 	});
+	const extDispatch = ctx.webServer.register({
+		kind: "prefix",
+		path: `${ROUTER_API_BASE}/ext/`,
+		handler: async (req, res) => {
+			const pathname = new URL(req.url ?? "/", "http://localhost").pathname;
+			const mount = `${ROUTER_API_BASE}/ext/`;
+			const rest = pathname.startsWith(mount) ? pathname.slice(mount.length) : "";
+			const slash = rest.indexOf("/");
+			const extId = slash < 0 ? rest : rest.slice(0, slash);
+			const subpath = slash < 0 ? "" : rest.slice(slash + 1);
+			if (extId === "") {
+				writeJson(res, 404, {
+					ok: false,
+					error: "extension not found"
+				});
+				return;
+			}
+			const ext = exts[extId];
+			if (ext !== void 0 && typeof ext.api === "function" && await ext.api(subpath, req, res)) return;
+			writeJson(res, 404, {
+				ok: false,
+				error: "extension api not found"
+			});
+		}
+	});
+	disposers.push(extDispatch);
 	/** 解析并校验周期参数（不接受任意字符串）。 */
 	const periodOf = (raw) => {
 		return raw === "24h" || raw === "7d" || raw === "30d" ? raw : "today";