mslxdff 0.1.155 → 0.1.158
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.
- package/bin/mslxdff.js +4 -4
- package/docs/ARCHITECTURE.md +426 -0
- package/docs/FEATURE_TREE.md +164 -0
- package/docs/MOBILE.md +82 -0
- package/docs/adr/0001-reasoning-content-injection.md +14 -0
- package/docs/adr/0002-models-free-filter.md +12 -0
- package/docs/adr/0003-zero-state-no-auth.md +10 -0
- package/docs/adr/0004-bearer-token.md +18 -0
- package/docs/adr/0005-peer-mesh.md +53 -0
- package/docs/adr/0006-broadband-member.md +103 -0
- package/docs/adr/0007-multi-provider-prefix.md +25 -0
- package/docs/adr/0008-share-keys-to-peers.md +52 -0
- package/docs/adr/0009-chat-repl.md +37 -0
- package/docs/adr/0010-allowlist.md +36 -0
- package/docs/adr/0011-broadband-stream.md +30 -0
- package/docs/adr/0012-responses-endpoint-codex-sync.md +48 -0
- package/docs/adr/0013-node16-compat.md +41 -0
- package/docs/adr/0014-deepseek-provider.md +52 -0
- package/docs/adr/0015-upstream-probe-routing.md +50 -0
- package/docs/adr/0016-model-capabilities.md +27 -0
- package/docs/adr/0017-ai-sdk-upstream-engine.md +59 -0
- package/docs/adr/0018-zen-client-identity.md +52 -0
- package/docs/adr/0019-share-keys-always-lend.md +59 -0
- package/docs/adr/0020-zen-free-lane-agent-shape.md +52 -0
- package/docs/adr/0021-usage-report-jsonl.md +47 -0
- package/docs/adr/0022-models-capability-merge.md +61 -0
- package/docs/adr/0023-key-provider-default-direct.md +58 -0
- package/docs/adr/0024-node18-baseline.md +63 -0
- package/docs/adr/0025-workbuddy-authdir-follows-state.md +71 -0
- package/docs/adr/0026-cline-provider-id-unify.md +67 -0
- package/docs/adr/0027-codearts-provider.md +48 -0
- package/docs/adr/0028-traework-provider.md +34 -0
- package/docs/adr/0029-qoder-native-provider.md +60 -0
- package/docs/adr/0030-models-list-scoped-by-picks.md +53 -0
- package/docs/adr/0031-qoder-true-streaming.md +50 -0
- package/docs/adr/0032-generic-responses-channel.md +72 -0
- package/docs/adr/0033-cline-allowlist-auto-sync.md +75 -0
- package/docs/adr/0034-request-level-human-readable-observability.md +60 -0
- package/docs/adr/0035-sdk-channel-headers-timeout.md +49 -0
- package/docs/adr/0036-qoder-per-request-sticky-account.md +82 -0
- package/docs/adr/0037-qwenwork-independent-provider.md +82 -0
- package/docs/adr/0038-zcode-provider.md +140 -0
- package/docs/agents/domain.md +51 -0
- package/docs/agents/issue-tracker.md +30 -0
- package/docs/agents/triage-labels.md +15 -0
- package/docs/cli_help.md +1391 -0
- package/docs/cli_help_mini.md +133 -0
- package/docs/plans/bench-via-latency-2026-09-01.md +215 -0
- package/docs/plugins.md +187 -0
- package/package.json +1 -1
- package/src/auto.js +254 -254
- package/src/bench/cline-bench.js +42 -42
- package/src/bench/probe.js +70 -70
- package/src/bench/report.js +162 -162
- package/src/bench/runner.js +77 -77
- package/src/bench/via-probe.js +124 -124
- package/src/bench/via-routes.js +87 -87
- package/src/bench/workbuddy-bench.js +54 -54
- package/src/chat/engine.js +160 -160
- package/src/chat/gateway.js +163 -163
- package/src/chat/orchestrator.js +234 -234
- package/src/chat/prompt.js +70 -70
- package/src/chat/repl.js +88 -88
- package/src/chat/terminal.js +135 -135
- package/src/chat/tools.js +306 -306
- package/src/chat-pipeline/index.js +123 -123
- package/src/chat-pipeline/policy.js +76 -76
- package/src/chat-pipeline/serial-trial.js +210 -210
- package/src/cli/commands/group.js +249 -249
- package/src/cli/commands/model/list-providers.js +4 -3
- package/src/cli/commands/model/list-render.js +7 -9
- package/src/cli/commands/model/list-sort.js +52 -0
- package/src/cli/commands/model/list-tty.js +76 -0
- package/src/cli/commands/model/list.js +5 -62
- package/src/cli/commands/model/picks.js +50 -50
- package/src/cli/commands/provider/bench-via.js +247 -247
- package/src/cli/commands/provider/bench.js +141 -141
- package/src/cli/commands/provider/index.js +124 -116
- package/src/cli/commands/provider/models.js +139 -128
- package/src/cli/commands/provider/qwenwork-login.js +119 -0
- package/src/cli/commands/provider/zcode-login.js +77 -0
- package/src/cli/commands/provider/zcode-quota.js +55 -0
- package/src/cli/commands/sync.js +232 -232
- package/src/cli/provider-row.js +2 -2
- package/src/cli/status.js +279 -279
- package/src/daemon.js +96 -96
- package/src/model-capabilities/enrich.js +86 -86
- package/src/model-capabilities/index.js +183 -183
- package/src/model-capabilities/parse.js +70 -70
- package/src/model-trace.js +1 -0
- package/src/models.js +225 -225
- package/src/providers/classify.js +1 -1
- package/src/providers/cline/auth.js +228 -228
- package/src/providers/cline/chat.js +307 -307
- package/src/providers/cline.js +2 -2
- package/src/providers/keyring.js +60 -56
- package/src/providers/qoder/chat.js +183 -100
- package/src/providers/qoder/index.js +230 -184
- package/src/providers/qoder/sse.js +103 -46
- package/src/providers/qoder/sticky.js +1 -0
- package/src/providers/qoder/stream.js +32 -9
- package/src/providers/qwenwork/account-store.js +133 -0
- package/src/providers/qwenwork/constants.js +67 -0
- package/src/providers/qwenwork/cosy.js +120 -0
- package/src/providers/qwenwork/crypto.js +218 -0
- package/src/providers/qwenwork/http.js +20 -0
- package/src/providers/qwenwork/index.js +327 -0
- package/src/providers/qwenwork/payload.js +142 -0
- package/src/providers/qwenwork/rsa.js +54 -0
- package/src/providers/qwenwork/sse.js +268 -0
- package/src/providers/qwenwork/stream.js +130 -0
- package/src/providers/qwenwork/upstream.js +120 -0
- package/src/providers/qwenwork.js +1 -0
- package/src/providers/registry.js +66 -56
- package/src/providers/share-keys.js +2 -2
- package/src/providers/workbuddy/chat.js +248 -248
- package/src/providers/workbuddy/reshape.js +152 -152
- package/src/providers/workbuddy.js +2 -2
- package/src/providers/zcode/account-store.js +129 -0
- package/src/providers/zcode/auth.js +28 -0
- package/src/providers/zcode/chat.js +171 -0
- package/src/providers/zcode/const.js +54 -0
- package/src/providers/zcode/headers.js +55 -0
- package/src/providers/zcode/index.js +124 -0
- package/src/providers/zcode/models.js +65 -0
- package/src/providers/zcode/oauth.js +120 -0
- package/src/providers/zcode/quota.js +176 -0
- package/src/providers/zcode/sse.js +179 -0
- package/src/reasoning.js +32 -32
- package/src/routes/chat/gateway.js +46 -46
- package/src/routes/chat/relay-pipeline.js +250 -246
- package/src/routes/chat/via-route-handler.js +144 -144
- package/src/routes/hedge.js +255 -255
- package/src/routes/models-route.js +167 -167
- package/src/routes/peers.js +273 -273
- package/src/routes/stream.js +438 -399
- package/src/runtime/bootstrap.js +45 -45
- package/src/runtime/provider-gate.js +33 -30
- package/src/runtime/providers-setup.js +165 -165
- package/src/server.js +64 -64
- package/src/state/schemas/allowlist.js +92 -92
- package/src/sync-opencode.js +280 -280
- package/src/transport/index.js +244 -244
- package/src/transport/pool.js +56 -56
- package/src/transport/retry.js +24 -24
- package/src/transport/sse.js +93 -93
- package/src/upstream-probe/display.js +52 -52
- package/src/upstream-probe/probe.js +49 -49
- package/src/upstream-probe/rotate.js +110 -110
- package/src/upstream-probe/start.js +45 -45
- package/src/upstream.js +289 -289
package/src/routes/hedge.js
CHANGED
|
@@ -1,255 +1,255 @@
|
|
|
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 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 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
|
+
}
|