mslxdff 0.1.187 → 0.1.188

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (111) hide show
  1. package/bin/mslxdff.js +3 -3
  2. package/docs/cli_help_mini.md +145 -0
  3. package/package.json +1 -1
  4. package/src/auto.js +254 -254
  5. package/src/autostart.js +5 -2
  6. package/src/bench/cline-bench.js +42 -42
  7. package/src/bench/probe.js +70 -70
  8. package/src/bench/report.js +162 -162
  9. package/src/bench/runner.js +77 -77
  10. package/src/bench/via-probe.js +124 -124
  11. package/src/bench/via-routes.js +87 -87
  12. package/src/bench/workbuddy-bench.js +54 -54
  13. package/src/chat/engine.js +160 -160
  14. package/src/chat/gateway.js +163 -163
  15. package/src/chat/orchestrator.js +234 -234
  16. package/src/chat/prompt.js +70 -70
  17. package/src/chat/repl.js +88 -88
  18. package/src/chat/terminal.js +135 -135
  19. package/src/chat/tools.js +306 -306
  20. package/src/chat-pipeline/index.js +173 -131
  21. package/src/chat-pipeline/policy.js +84 -84
  22. package/src/chat-pipeline/serial-trial.js +6 -3
  23. package/src/chat-pipeline/terminal-verdict.js +29 -0
  24. package/src/cli/commands/daemon.js +55 -18
  25. package/src/cli/commands/group.js +249 -249
  26. package/src/cli/commands/model/picks.js +50 -50
  27. package/src/cli/commands/provider/bench-via.js +247 -247
  28. package/src/cli/commands/provider/bench.js +141 -141
  29. package/src/cli/commands/provider/index.js +136 -136
  30. package/src/cli/commands/provider/models.js +145 -145
  31. package/src/cli/commands/provider/qwenwork-login.js +119 -119
  32. package/src/cli/commands/sync.js +310 -310
  33. package/src/cli/commands/system.js +23 -7
  34. package/src/cli/join-wizard.js +8 -4
  35. package/src/cli/policy.js +14 -2
  36. package/src/cli/status.js +279 -279
  37. package/src/daemon.js +232 -195
  38. package/src/logs.js +27 -11
  39. package/src/model-capabilities/enrich.js +86 -86
  40. package/src/model-capabilities/index.js +183 -183
  41. package/src/model-capabilities/parse.js +70 -70
  42. package/src/model-trace.js +35 -4
  43. package/src/models.js +225 -225
  44. package/src/peers.js +20 -1
  45. package/src/providers/AGENTS.md +49 -0
  46. package/src/providers/cline/auth.js +228 -228
  47. package/src/providers/cline/chat.js +307 -307
  48. package/src/providers/cline.js +2 -2
  49. package/src/providers/codearts/chat.js +182 -182
  50. package/src/providers/globalqwenwork/cosy.js +123 -123
  51. package/src/providers/globalqwenwork/crypto.js +218 -218
  52. package/src/providers/globalqwenwork/http.js +20 -20
  53. package/src/providers/globalqwenwork/rsa.js +54 -54
  54. package/src/providers/globalqwenwork/sse.js +268 -268
  55. package/src/providers/globalqwenwork/stream.js +130 -130
  56. package/src/providers/globalqwenwork/upstream.js +122 -122
  57. package/src/providers/keyring.js +60 -60
  58. package/src/providers/qoder/chat.js +203 -203
  59. package/src/providers/qoder/index.js +249 -249
  60. package/src/providers/qoder/sse.js +145 -145
  61. package/src/providers/qwenwork/account-store.js +133 -133
  62. package/src/providers/qwenwork/constants.js +67 -67
  63. package/src/providers/qwenwork/cosy.js +120 -120
  64. package/src/providers/qwenwork/crypto.js +218 -218
  65. package/src/providers/qwenwork/http.js +20 -20
  66. package/src/providers/qwenwork/index.js +337 -327
  67. package/src/providers/qwenwork/payload.js +144 -144
  68. package/src/providers/qwenwork/rsa.js +54 -54
  69. package/src/providers/qwenwork/sse.js +268 -268
  70. package/src/providers/qwenwork/stream.js +130 -130
  71. package/src/providers/qwenwork/upstream.js +120 -120
  72. package/src/providers/qwenwork.js +1 -1
  73. package/src/providers/registry.js +79 -79
  74. package/src/providers/traework/chat.js +7 -1
  75. package/src/providers/workbuddy/chat.js +248 -248
  76. package/src/providers/workbuddy/reshape.js +152 -152
  77. package/src/providers/workbuddy.js +2 -2
  78. package/src/reasoning.js +32 -32
  79. package/src/routes/AGENTS.md +37 -0
  80. package/src/routes/chat/empty-turn-report.js +31 -0
  81. package/src/routes/chat/gateway.js +46 -46
  82. package/src/routes/chat/peer-handler.js +3 -2
  83. package/src/routes/chat/relay-pipeline.js +282 -294
  84. package/src/routes/chat/via-route-handler.js +146 -146
  85. package/src/routes/hedge.js +253 -255
  86. package/src/routes/index.js +14 -6
  87. package/src/routes/models-route.js +167 -167
  88. package/src/routes/peers.js +335 -273
  89. package/src/routes/responses-route.js +44 -2
  90. package/src/routes/stream-body.js +84 -0
  91. package/src/routes/stream-hold.js +152 -4
  92. package/src/routes/stream-scan.js +27 -2
  93. package/src/routes/stream.js +343 -385
  94. package/src/runtime/bootstrap.js +67 -67
  95. package/src/runtime/group-sync.js +70 -17
  96. package/src/runtime/provider-gate.js +36 -36
  97. package/src/runtime/providers-setup.js +168 -168
  98. package/src/runtime/server-lifecycle.js +1 -1
  99. package/src/server.js +78 -78
  100. package/src/state/schemas/allowlist.js +92 -92
  101. package/src/sync-opencode.js +280 -280
  102. package/src/transport/index.js +244 -244
  103. package/src/transport/pool.js +56 -56
  104. package/src/transport/retry.js +24 -24
  105. package/src/transport/sse.js +93 -93
  106. package/src/upstream-engine/sdk/attempt.js +9 -1
  107. package/src/upstream-probe/display.js +52 -52
  108. package/src/upstream-probe/probe.js +49 -49
  109. package/src/upstream-probe/rotate.js +110 -110
  110. package/src/upstream-probe/start.js +45 -45
  111. package/src/upstream.js +293 -293
@@ -1,255 +1,253 @@
1
- function hedgeDelayMs() {
2
- const raw = process.env.MSLXDFF_HEDGE_DELAY_MS;
3
- if (raw === undefined || raw === null || raw === "") return 1000;
4
- const s = String(raw).trim().toLowerCase();
5
- if (s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled") return 0;
6
- const n = Number(s);
7
- if (Number.isInteger(n) && n >= 0) return n;
8
- return 1000;
9
- }
10
-
11
- function isFastFailStatus(status) {
12
- const s = Number(status);
13
- return s === 429 || s === 502 || s === 503 || s === 504;
14
- }
15
-
16
- function shouldHedge({ isStream, canForwardPeers, hedgeDelayMs: d, hasPeers, model }) {
17
- if (!isStream) return false;
18
- if (!canForwardPeers) return false;
19
- if (!hasPeers) return false;
20
- if (!d || d <= 0) return false;
21
- // 供应商级禁对冲在 shouldUseGroupForModel 层收敛(workbuddy local-only 硬禁,见 ADR-0015);
22
- // 这里只留 muse-spark 特判:走 /responses 流式(event: 包装 + 加密 reasoning),组员旧版无此整形且聚合 JSON 带错 header,必抢赢本地慢首块,需本地直出
23
- if (String(model || "").toLowerCase().startsWith("muse-spark")) return false;
24
- return true;
25
- }
26
-
27
- function withTimeout(promise, ms) {
28
- if (ms <= 0) return promise;
29
- let timer;
30
- const timeout = new Promise((_, rej) => {
31
- timer = setTimeout(() => rej(new Error("hedge-timeout")), ms);
32
- });
33
- return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
34
- }
35
-
36
- function makeBufferedBody(chunk, handle, kind) {
37
- if (kind === "web") {
38
- const reader = handle;
39
- return (async function* () {
40
- yield chunk;
41
- while (true) {
42
- const { value, done } = await reader.read();
43
- if (done) break;
44
- yield value;
45
- }
46
- })();
47
- }
48
- const it = handle;
49
- return (async function* () {
50
- yield chunk;
51
- for await (const c of { [Symbol.asyncIterator]: () => it }) {
52
- yield c;
53
- }
54
- })();
55
- }
56
-
57
- async function cancelBody(body, handle, kind) {
58
- try {
59
- if (body && typeof body.cancel === "function") await body.cancel().catch(() => {});
60
- if (kind === "web" && handle && typeof handle.cancel === "function") await handle.cancel().catch(() => {});
61
- else if (handle && typeof handle.return === "function") await handle.return().catch(() => {});
62
- } catch {}
63
- }
64
-
65
- function getFirstChunkPromise(body) {
66
- if (!body) throw new Error("no-body");
67
- if (typeof body.getReader === "function") {
68
- const reader = body.getReader();
69
- const p = reader.read().then((r) => {
70
- if (r.done) throw new Error("no-chunk");
71
- return { chunk: r.value, handle: reader, kind: "web", reader };
72
- });
73
- return { promise: p, handle: reader, kind: "web" };
74
- }
75
- const it = body[Symbol.asyncIterator]?.();
76
- if (!it) throw new Error("no-iterator");
77
- const p = it.next().then((r) => {
78
- if (r.done) throw new Error("no-chunk");
79
- return { chunk: r.value, handle: it, kind: "node", it };
80
- });
81
- return { promise: p, handle: it, kind: "node" };
82
- }
83
-
84
- export { hedgeDelayMs, isFastFailStatus, shouldHedge };
85
-
86
- export async function hedgedFirstChunkRace({
87
- localUpRes,
88
- peers,
89
- handlerCtx,
90
- hedgeDelayMs: hedgeDelay,
91
- evt,
92
- }) {
93
- const t0 = Date.now();
94
- const debugHedge = process.env.MSLXDFF_DEBUG === "1";
95
- if (debugHedge) try { console.log(`[hedge] start delay=${hedgeDelay} peers=${peers?.ordered?.().length ?? 0} hasBody=${Boolean(localUpRes.body)} getReader=${typeof localUpRes.body?.getReader}`); } catch {}
96
- let localHandle = null;
97
- let localKind = null;
98
- let localBody = localUpRes.body;
99
- let localFirstPromise = null;
100
- try {
101
- const { promise, handle, kind } = getFirstChunkPromise(localBody);
102
- localHandle = handle;
103
- localKind = kind;
104
- localFirstPromise = promise;
105
- } catch (e) {
106
- evt?.("hedge-trigger", { reason: "fast-fail-no-body", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0 });
107
- return { needsPeer: true };
108
- }
109
-
110
- try {
111
- const local = await withTimeout(localFirstPromise, hedgeDelay);
112
- if (debugHedge) try { console.log(`[hedge] local win before hedge ttf=${Date.now() - t0}ms`); } catch {}
113
- evt?.("hedge-win", { via: "local", reason: "no-hedge-needed", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay });
114
- return {
115
- winner: "local",
116
- upRes: localUpRes,
117
- bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
118
- peerInfo: null,
119
- ttfMs: Date.now() - t0,
120
- hedged: false,
121
- };
122
- } catch (err) {
123
- if (err.message !== "hedge-timeout") {
124
- if (debugHedge) try { console.log(`[hedge] local fast-fail ${err.message}`); } catch {}
125
- evt?.("hedge-trigger", { reason: "fast-fail", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0, error: String(err.message).slice(0, 200) });
126
- return { needsPeer: true, localError: err };
127
- }
128
- }
129
-
130
- if (debugHedge) try { console.log(`[hedge] trigger slow hedge after ${hedgeDelay}ms`); } catch {}
131
- evt?.("hedge-trigger", { reason: "slow", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0 });
132
-
133
- const { racePeerCandidates } = await import("./peers.js");
134
- const candidates = handlerCtx.peers.ordered();
135
- let peerWin = null;
136
- try {
137
- peerWin =
138
- (await racePeerCandidates(candidates, handlerCtx)) ||
139
- (await racePeerCandidates(handlerCtx.peers.orderedByLastError(), handlerCtx)) ||
140
- (await racePeerCandidates(handlerCtx.peers.coolingByLastError(), handlerCtx));
141
- } catch (_) {
142
- peerWin = null;
143
- }
144
- if (debugHedge) try { console.log(`[hedge] peerWin=${peerWin ? peerWin.peer.url : "none"}`); } catch {}
145
-
146
- if (!peerWin) {
147
- const remaining = 25_000 - hedgeDelay;
148
- try {
149
- const local = await withTimeout(localFirstPromise, Math.max(0, remaining));
150
- evt?.("hedge-win", { via: "local", reason: "peer-miss", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay });
151
- if (debugHedge) try { console.log(`[hedge] local win after peer miss`); } catch {}
152
- return {
153
- winner: "local",
154
- upRes: localUpRes,
155
- bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
156
- peerInfo: null,
157
- ttfMs: Date.now() - t0,
158
- hedged: true,
159
- };
160
- } catch (e) {
161
- evt?.("hedge-lose", { via: "local", reason: "both-fail", error: String(e.message).slice(0, 200) });
162
- await cancelBody(localBody, localHandle, localKind);
163
- return null;
164
- }
165
- }
166
-
167
- let peerHandle = null;
168
- let peerKind = null;
169
- let peerBody = peerWin.res.body;
170
- let peerFirstPromise = null;
171
- try {
172
- const { promise, handle, kind } = getFirstChunkPromise(peerBody);
173
- peerHandle = handle;
174
- peerKind = kind;
175
- peerFirstPromise = promise.then((r) => ({ ...r, peer: peerWin }));
176
- } catch (e) {
177
- evt?.("hedge-lose", { via: "peer", reason: "peer-no-body", error: String(e.message).slice(0, 200) });
178
- try {
179
- const local = await localFirstPromise;
180
- return {
181
- winner: "local",
182
- upRes: localUpRes,
183
- bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
184
- peerInfo: null,
185
- ttfMs: Date.now() - t0,
186
- hedged: true,
187
- };
188
- } catch {
189
- return null;
190
- }
191
- }
192
-
193
- try {
194
- const winner = await Promise.race([
195
- localFirstPromise.then((local) => ({ via: "local", data: { ...local, kind: localKind, handle: localHandle } })),
196
- peerFirstPromise.then((peer) => ({ via: "peer", data: peer })),
197
- ]);
198
- if (debugHedge) try { console.log(`[hedge] race winner=${winner.via}`); } catch {}
199
-
200
- if (winner.via === "local") {
201
- await cancelBody(peerBody, peerHandle, peerKind);
202
- evt?.("hedge-win", { via: "local", reason: "race", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay, peer: peerWin.peer.url });
203
- evt?.("hedge-loser-cancel", { via: "peer", peer: peerWin.peer.url });
204
- return {
205
- winner: "local",
206
- upRes: localUpRes,
207
- bufferedBody: makeBufferedBody(winner.data.chunk, winner.data.handle, winner.data.kind),
208
- peerInfo: null,
209
- ttfMs: Date.now() - t0,
210
- hedged: true,
211
- };
212
- } else {
213
- await cancelBody(localBody, localHandle, localKind);
214
- evt?.("hedge-win", { via: "peer", reason: "race", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay, peer: winner.data.peer.peer.url });
215
- evt?.("hedge-loser-cancel", { via: "local" });
216
- return {
217
- winner: "peer",
218
- upRes: winner.data.peer.res,
219
- bufferedBody: makeBufferedBody(winner.data.chunk, winner.data.handle, winner.data.kind),
220
- peerInfo: winner.data.peer,
221
- ttfMs: Date.now() - t0,
222
- hedged: true,
223
- };
224
- }
225
- } catch (e) {
226
- if (debugHedge) try { console.log(`[hedge] race both failed, fallback sequential`); } catch {}
227
- try {
228
- const local = await localFirstPromise;
229
- await cancelBody(peerBody, peerHandle, peerKind);
230
- return {
231
- winner: "local",
232
- upRes: localUpRes,
233
- bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
234
- peerInfo: null,
235
- ttfMs: Date.now() - t0,
236
- hedged: true,
237
- };
238
- } catch {
239
- try {
240
- const peer = await peerFirstPromise;
241
- await cancelBody(localBody, localHandle, localKind);
242
- return {
243
- winner: "peer",
244
- upRes: peer.peer.res,
245
- bufferedBody: makeBufferedBody(peer.chunk, peer.handle, peer.kind),
246
- peerInfo: peer.peer,
247
- ttfMs: Date.now() - t0,
248
- hedged: true,
249
- };
250
- } catch {
251
- return null;
252
- }
253
- }
254
- }
255
- }
1
+ function hedgeDelayMs() {
2
+ const raw = process.env.MSLXDFF_HEDGE_DELAY_MS;
3
+ if (raw === undefined || raw === null || raw === "") return 1000;
4
+ const s = String(raw).trim().toLowerCase();
5
+ if (s === "0" || s === "off" || s === "false" || s === "no" || s === "disable" || s === "disabled") return 0;
6
+ const n = Number(s);
7
+ if (Number.isInteger(n) && n >= 0) return n;
8
+ return 1000;
9
+ }
10
+
11
+ function shouldHedge({ isStream, canForwardPeers, hedgeDelayMs: d, hasPeers, model }) {
12
+ if (!isStream) return false;
13
+ if (!canForwardPeers) return false;
14
+ if (!hasPeers) return false;
15
+ if (!d || d <= 0) return false;
16
+ // 供应商级禁对冲在 shouldUseGroupForModel 层收敛(workbuddy local-only 硬禁,见 ADR-0015);
17
+ // 这里只留 muse-spark 特判:走 /responses 流式(event: 包装 + 加密 reasoning),组员旧版无此整形且聚合 JSON 带错 header,必抢赢本地慢首块,需本地直出
18
+ if (String(model || "").toLowerCase().startsWith("muse-spark")) return false;
19
+ return true;
20
+ }
21
+
22
+ function withTimeout(promise, ms) {
23
+ if (ms <= 0) return promise;
24
+ let timer;
25
+ const timeout = new Promise((_, rej) => {
26
+ timer = setTimeout(() => rej(new Error("hedge-timeout")), ms);
27
+ });
28
+ return Promise.race([promise, timeout]).finally(() => clearTimeout(timer));
29
+ }
30
+
31
+ function makeBufferedBody(chunk, handle, kind) {
32
+ if (kind === "web") {
33
+ const reader = handle;
34
+ return (async function* () {
35
+ yield chunk;
36
+ while (true) {
37
+ const { value, done } = await reader.read();
38
+ if (done) break;
39
+ yield value;
40
+ }
41
+ })();
42
+ }
43
+ const it = handle;
44
+ return (async function* () {
45
+ yield chunk;
46
+ for await (const c of { [Symbol.asyncIterator]: () => it }) {
47
+ yield c;
48
+ }
49
+ })();
50
+ }
51
+
52
+ async function cancelBody(body, handle, kind) {
53
+ try {
54
+ if (body && typeof body.cancel === "function") await body.cancel().catch(() => {});
55
+ if (kind === "web" && handle && typeof handle.cancel === "function") await handle.cancel().catch(() => {});
56
+ else if (handle && typeof handle.return === "function") await handle.return().catch(() => {});
57
+ } catch {}
58
+ }
59
+
60
+ function getFirstChunkPromise(body) {
61
+ if (!body) throw new Error("no-body");
62
+ if (typeof body.getReader === "function") {
63
+ const reader = body.getReader();
64
+ const p = reader.read().then((r) => {
65
+ if (r.done) throw new Error("no-chunk");
66
+ return { chunk: r.value, handle: reader, kind: "web", reader };
67
+ });
68
+ return { promise: p, handle: reader, kind: "web" };
69
+ }
70
+ const it = body[Symbol.asyncIterator]?.();
71
+ if (!it) throw new Error("no-iterator");
72
+ const p = it.next().then((r) => {
73
+ if (r.done) throw new Error("no-chunk");
74
+ return { chunk: r.value, handle: it, kind: "node", it };
75
+ });
76
+ return { promise: p, handle: it, kind: "node" };
77
+ }
78
+
79
+ export { hedgeDelayMs, shouldHedge };
80
+
81
+ export async function hedgedFirstChunkRace({
82
+ localUpRes,
83
+ peers,
84
+ handlerCtx,
85
+ hedgeDelayMs: hedgeDelay,
86
+ evt,
87
+ }) {
88
+ const t0 = Date.now();
89
+ const debugHedge = process.env.MSLXDFF_DEBUG === "1";
90
+ if (debugHedge) try { console.log(`[hedge] start delay=${hedgeDelay} peers=${peers?.ordered?.().length ?? 0} hasBody=${Boolean(localUpRes.body)} getReader=${typeof localUpRes.body?.getReader}`); } catch {}
91
+ let localHandle = null;
92
+ let localKind = null;
93
+ let localBody = localUpRes.body;
94
+ let localFirstPromise = null;
95
+ try {
96
+ const { promise, handle, kind } = getFirstChunkPromise(localBody);
97
+ localHandle = handle;
98
+ localKind = kind;
99
+ localFirstPromise = promise;
100
+ } catch (e) {
101
+ evt?.("hedge-trigger", { reason: "fast-fail-no-body", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0 });
102
+ return { needsPeer: true };
103
+ }
104
+
105
+ try {
106
+ const local = await withTimeout(localFirstPromise, hedgeDelay);
107
+ if (debugHedge) try { console.log(`[hedge] local win before hedge ttf=${Date.now() - t0}ms`); } catch {}
108
+ evt?.("hedge-win", { via: "local", reason: "no-hedge-needed", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay });
109
+ return {
110
+ winner: "local",
111
+ upRes: localUpRes,
112
+ bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
113
+ peerInfo: null,
114
+ ttfMs: Date.now() - t0,
115
+ hedged: false,
116
+ };
117
+ } catch (err) {
118
+ if (err.message !== "hedge-timeout") {
119
+ if (debugHedge) try { console.log(`[hedge] local fast-fail ${err.message}`); } catch {}
120
+ evt?.("hedge-trigger", { reason: "fast-fail", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0, error: String(err.message).slice(0, 200) });
121
+ return { needsPeer: true, localError: err };
122
+ }
123
+ }
124
+
125
+ if (debugHedge) try { console.log(`[hedge] trigger slow hedge after ${hedgeDelay}ms`); } catch {}
126
+ evt?.("hedge-trigger", { reason: "slow", hedgeDelayMs: hedgeDelay, triggerAtMs: Date.now() - t0 });
127
+
128
+ const { racePeerCandidates, peerStageCtx } = await import("./peers.js");
129
+ const candidates = handlerCtx.peers.ordered();
130
+ // 三批共享同一「peer 阶段总预算」(peerStageCtx 设 deadline,racePeerCandidates 按剩余量裁剪等待):
131
+ // 旧行为是每批各吃满一次等头超时,串行叠加可到 90s+(实测 peer 阶段 63s、total 137s)。
132
+ const stage = peerStageCtx(handlerCtx);
133
+ let peerWin = null;
134
+ try {
135
+ peerWin =
136
+ (await racePeerCandidates(candidates, stage)) ||
137
+ (await racePeerCandidates(stage.peers.orderedByLastError(), stage)) ||
138
+ (await racePeerCandidates(stage.peers.coolingByLastError(), stage));
139
+ } catch (_) {
140
+ peerWin = null;
141
+ }
142
+ if (debugHedge) try { console.log(`[hedge] peerWin=${peerWin ? peerWin.peer.url : "none"}`); } catch {}
143
+
144
+ if (!peerWin) {
145
+ const remaining = 25_000 - hedgeDelay;
146
+ try {
147
+ const local = await withTimeout(localFirstPromise, Math.max(0, remaining));
148
+ evt?.("hedge-win", { via: "local", reason: "peer-miss", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay });
149
+ if (debugHedge) try { console.log(`[hedge] local win after peer miss`); } catch {}
150
+ return {
151
+ winner: "local",
152
+ upRes: localUpRes,
153
+ bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
154
+ peerInfo: null,
155
+ ttfMs: Date.now() - t0,
156
+ hedged: true,
157
+ };
158
+ } catch (e) {
159
+ evt?.("hedge-lose", { via: "local", reason: "both-fail", error: String(e.message).slice(0, 200) });
160
+ await cancelBody(localBody, localHandle, localKind);
161
+ return null;
162
+ }
163
+ }
164
+
165
+ let peerHandle = null;
166
+ let peerKind = null;
167
+ let peerBody = peerWin.res.body;
168
+ let peerFirstPromise = null;
169
+ try {
170
+ const { promise, handle, kind } = getFirstChunkPromise(peerBody);
171
+ peerHandle = handle;
172
+ peerKind = kind;
173
+ peerFirstPromise = promise.then((r) => ({ ...r, peer: peerWin }));
174
+ } catch (e) {
175
+ evt?.("hedge-lose", { via: "peer", reason: "peer-no-body", error: String(e.message).slice(0, 200) });
176
+ try {
177
+ const local = await localFirstPromise;
178
+ return {
179
+ winner: "local",
180
+ upRes: localUpRes,
181
+ bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
182
+ peerInfo: null,
183
+ ttfMs: Date.now() - t0,
184
+ hedged: true,
185
+ };
186
+ } catch {
187
+ return null;
188
+ }
189
+ }
190
+
191
+ try {
192
+ const winner = await Promise.race([
193
+ localFirstPromise.then((local) => ({ via: "local", data: { ...local, kind: localKind, handle: localHandle } })),
194
+ peerFirstPromise.then((peer) => ({ via: "peer", data: peer })),
195
+ ]);
196
+ if (debugHedge) try { console.log(`[hedge] race winner=${winner.via}`); } catch {}
197
+
198
+ if (winner.via === "local") {
199
+ await cancelBody(peerBody, peerHandle, peerKind);
200
+ evt?.("hedge-win", { via: "local", reason: "race", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay, peer: peerWin.peer.url });
201
+ evt?.("hedge-loser-cancel", { via: "peer", peer: peerWin.peer.url });
202
+ return {
203
+ winner: "local",
204
+ upRes: localUpRes,
205
+ bufferedBody: makeBufferedBody(winner.data.chunk, winner.data.handle, winner.data.kind),
206
+ peerInfo: null,
207
+ ttfMs: Date.now() - t0,
208
+ hedged: true,
209
+ };
210
+ } else {
211
+ await cancelBody(localBody, localHandle, localKind);
212
+ evt?.("hedge-win", { via: "peer", reason: "race", ttfMs: Date.now() - t0, hedgeDelayMs: hedgeDelay, peer: winner.data.peer.peer.url });
213
+ evt?.("hedge-loser-cancel", { via: "local" });
214
+ return {
215
+ winner: "peer",
216
+ upRes: winner.data.peer.res,
217
+ bufferedBody: makeBufferedBody(winner.data.chunk, winner.data.handle, winner.data.kind),
218
+ peerInfo: winner.data.peer,
219
+ ttfMs: Date.now() - t0,
220
+ hedged: true,
221
+ };
222
+ }
223
+ } catch (e) {
224
+ if (debugHedge) try { console.log(`[hedge] race both failed, fallback sequential`); } catch {}
225
+ try {
226
+ const local = await localFirstPromise;
227
+ await cancelBody(peerBody, peerHandle, peerKind);
228
+ return {
229
+ winner: "local",
230
+ upRes: localUpRes,
231
+ bufferedBody: makeBufferedBody(local.chunk, local.handle, local.kind),
232
+ peerInfo: null,
233
+ ttfMs: Date.now() - t0,
234
+ hedged: true,
235
+ };
236
+ } catch {
237
+ try {
238
+ const peer = await peerFirstPromise;
239
+ await cancelBody(localBody, localHandle, localKind);
240
+ return {
241
+ winner: "peer",
242
+ upRes: peer.peer.res,
243
+ bufferedBody: makeBufferedBody(peer.chunk, peer.handle, peer.kind),
244
+ peerInfo: peer.peer,
245
+ ttfMs: Date.now() - t0,
246
+ hedged: true,
247
+ };
248
+ } catch {
249
+ return null;
250
+ }
251
+ }
252
+ }
253
+ }
@@ -1,4 +1,5 @@
1
1
  import { timingSafeEqual, createHash } from "node:crypto";
2
+ import { readFileSync } from "node:fs";
2
3
  import { DEFAULT_MAX_HOPS } from "../peers.js";
3
4
  import { json, notFound, authorized } from "./helpers.js";
4
5
  import { chatHandler } from "./chat.js";
@@ -9,10 +10,22 @@ import { relayHandler } from "./relay.js";
9
10
  import { responsesHandler } from "./responses-route.js";
10
11
  import { messagesHandler, countTokensHandler } from "./messages-route.js";
11
12
 
12
- export function createRouter({ token, upstream, models, auto, logs, peers, maxHops = DEFAULT_MAX_HOPS, groups, bans, bus, plugins }) {
13
+ export function createRouter({ token, upstream, models, auto, logs, peers, maxHops = DEFAULT_MAX_HOPS, groups, bans, bus, plugins, version = null }) {
14
+ // /health 要报版本:调用方(waitForHealth)拿它判断「应答的是不是刚起的那个」。
15
+ // 未显式传入时读 package.json 兜底 —— 测试直接 createRouter() 也能拿到真版本,不必逐个补参数。
16
+ const healthVersion = (() => {
17
+ if (version) return version;
18
+ try { return JSON.parse(readFileSync(new URL("../../package.json", import.meta.url), "utf8")).version ?? null; } catch { return null; }
19
+ })();
13
20
  return async function router(req, res) {
14
21
  const method = req.method || "GET";
15
22
  const path = (req.url || "").split("?")[0];
23
+ // pid/version 不是装饰:调用方(`-restart`/`-port`/`-update` 的 waitForHealth)需要它判断
24
+ // 「应答的到底是不是我刚起的那个」。只回 {status:"ok"} 时,旧 daemon 占着端口也能让健康检查通过,
25
+ // 于是「新进程 EADDRINUSE 秒退、旧版仍在服务」会被谎报成重启成功(BUG-29)。
26
+ if (method === "GET" && path === "/health") {
27
+ return json(res, 200, { status: "ok", pid: process.pid, version: healthVersion });
28
+ }
16
29
  // dynamic provider models route: GET /v1/providers/:id/models
17
30
  if (method === "GET" && /^\/v1\/providers\/[^/]+\/models\/?$/.test(path)) {
18
31
  if (!authorized(req, token)) {
@@ -34,11 +47,6 @@ export function createRouter({ token, upstream, models, auto, logs, peers, maxHo
34
47
  }
35
48
 
36
49
  const ROUTES = [
37
- {
38
- method: "GET",
39
- path: "/health",
40
- handler: ({ res }) => json(res, 200, { status: "ok" }),
41
- },
42
50
  {
43
51
  method: "POST",
44
52
  path: "/v1/chat/completions",