@jeffusion/bungee-core 6.1.1 → 6.2.0

package/dist/plugins/llm-tps/control.js

@@ -0,0 +1,212 @@
+// @bun
+// plugins/llm-tps/contract.ts
+var TPS_RETENTION_MS = 31 * 24 * 60 * 60 * 1000;
+var TPS_REPORTING_INCOMPLETE_KEY = "reporting:incomplete";
+
+// plugins/llm-tps/server/metrics.ts
+function isUrlIdentity(value) {
+  return /^(?:[a-z][a-z\d+.-]*:)?\/\//i.test(value);
+}
+function safeCount(value, name) {
+  if (typeof value !== "number" || !Number.isSafeInteger(value) || value < 0)
+    throw new RangeError(`invalid llm-tps ${name}`);
+  return value;
+}
+function identifier(value, name) {
+  if (typeof value !== "string" || !value.trim())
+    throw new TypeError(`invalid llm-tps ${name}`);
+}
+function validateSample(sample) {
+  for (const key of ["attemptId", "requestId", "routeId", "endpointId"])
+    identifier(sample[key], key);
+  if (isUrlIdentity(sample.endpointId))
+    throw new TypeError("llm-tps endpoint must be opaque");
+  safeCount(sample.finishedAtMs, "finishedAtMs");
+  safeCount(sample.outputTokens, "outputTokens");
+  if (!Number.isFinite(sample.generationDurationMs) || sample.generationDurationMs <= 0)
+    throw new RangeError("invalid llm-tps duration");
+  if (sample.outputSource !== "official" && sample.outputSource !== "estimated")
+    throw new TypeError("invalid llm-tps output source");
+  if (!Number.isFinite(sample.outputTokens / sample.generationDurationMs * 1000))
+    throw new RangeError("invalid llm-tps rate");
+}
+function rangeDuration(range) {
+  if (range === "1h")
+    return 3600000;
+  if (range === "12h")
+    return 43200000;
+  if (range === "24h")
+    return 86400000;
+  throw new RangeError("invalid llm-tps range");
+}
+function aggregate(row) {
+  const outputTokens = safeCount(row.outputTokens, "aggregate tokens");
+  const sampleCount = safeCount(row.sampleCount, "sample count");
+  const estimatedSampleCount = safeCount(row.estimatedSampleCount, "estimated sample count");
+  const duration = row.generationDurationMs;
+  if (typeof duration !== "number" || !Number.isFinite(duration) || duration < 0 || sampleCount > 0 && duration <= 0 || estimatedSampleCount > sampleCount || sampleCount === 0 && (outputTokens !== 0 || duration !== 0 || estimatedSampleCount !== 0))
+    throw new RangeError("invalid llm-tps aggregate");
+  const tps = sampleCount === 0 ? null : outputTokens / duration * 1000;
+  if (tps !== null && (!Number.isFinite(tps) || tps < 0))
+    throw new RangeError("invalid llm-tps aggregate rate");
+  return { tps, outputTokens, generationDurationMs: duration, sampleCount, estimatedSampleCount };
+}
+
+// plugins/llm-tps/server/storage.ts
+function observationClient(storage) {
+  if (typeof storage.invokeObservation !== "function")
+    throw new Error("llm-tps observation storage is unavailable");
+  const invoke = storage.invokeObservation.bind(storage);
+  return {
+    recordSample: (sample) => invoke("recordSample", [sample]),
+    queryStats: (input) => invoke("queryStats", [input])
+  };
+}
+var metrics = `COALESCE(SUM(output_tokens), 0) AS outputTokens,
+  COALESCE(SUM(generation_duration_ms), 0) AS generationDurationMs,
+  COUNT(*) AS sampleCount, COALESCE(SUM(output_source = 'estimated'), 0) AS estimatedSampleCount,
+  COALESCE(SUM(CASE WHEN typeof(output_tokens) != 'integer' OR output_tokens < 0 OR output_tokens > 9007199254740991
+    OR typeof(generation_duration_ms) NOT IN ('integer', 'real') OR NOT(generation_duration_ms > 0)
+    OR generation_duration_ms > 1.7976931348623157e308 OR output_source NOT IN ('official','estimated')
+    OR typeof(finished_at_ms) != 'integer' OR finished_at_ms < 0 OR finished_at_ms > 9007199254740991
+    THEN 1 ELSE 0 END), 0) AS invalidSamples`;
+function createObservationAdapter(observation) {
+  return Object.freeze({
+    async recordSample(sample) {
+      validateSample(sample);
+      observation.withDatabase((db) => {
+        db.transaction(() => {
+          db.query(`INSERT OR IGNORE INTO llm_tps_samples
+            (attempt_id, request_id, route_id, endpoint_id, finished_at_ms, output_tokens, generation_duration_ms, output_source)
+            VALUES (?, ?, ?, ?, ?, ?, ?, ?)`).run(sample.attemptId, sample.requestId, sample.routeId, sample.endpointId, sample.finishedAtMs, sample.outputTokens, sample.generationDurationMs, sample.outputSource);
+          db.query(`DELETE FROM llm_tps_samples WHERE attempt_id IN (
+            SELECT attempt_id FROM llm_tps_samples INDEXED BY idx_llm_tps_samples_finished
+            WHERE finished_at_ms < ? ORDER BY finished_at_ms LIMIT 500
+          )`).run(Date.now() - TPS_RETENTION_MS);
+        }).immediate();
+      });
+    },
+    async queryStats(input) {
+      safeCount(input.asOfMs, "asOfMs");
+      const start = input.asOfMs - rangeDuration(input.range);
+      return observation.withDatabase((db) => db.transaction(() => {
+        const where = "WHERE finished_at_ms >= ? AND finished_at_ms <= ?";
+        const totalRow = db.query(`SELECT ${metrics} FROM llm_tps_samples ${where}`).get(start, input.asOfMs);
+        const rows = db.query(`SELECT route_id AS routeId, endpoint_id AS endpointId, ${metrics}
+          FROM llm_tps_samples ${where} GROUP BY route_id, endpoint_id`).all(start, input.asOfMs);
+        const checked = (row) => {
+          if (safeCount(row.invalidSamples, "invalid sample count") !== 0)
+            throw new RangeError("invalid persisted llm-tps sample");
+          return aggregate(row);
+        };
+        const total = checked(totalRow);
+        const endpoints = rows.map((row) => {
+          identifier(row.routeId, "routeId");
+          identifier(row.endpointId, "endpointId");
+          if (isUrlIdentity(row.endpointId))
+            throw new TypeError("invalid persisted llm-tps endpoint");
+          return { routeId: row.routeId, endpointId: row.endpointId, ...checked(row) };
+        }).sort((a, b) => b.tps - a.tps || compare(a.routeId, b.routeId) || compare(a.endpointId, b.endpointId));
+        return { range: input.range, asOfMs: input.asOfMs, total, endpoints, reportingIncomplete: false };
+      })());
+    }
+  });
+}
+function compare(a, b) {
+  return a < b ? -1 : a > b ? 1 : 0;
+}
+
+// plugins/llm-tps/server/control.ts
+class ControlError extends Error {
+  code;
+  constructor(code) {
+    super(code);
+    this.code = code;
+  }
+}
+function abortable(operation, signals) {
+  return new Promise((resolve, reject) => {
+    const listeners = [];
+    const cleanup = () => signals.forEach((signal, index) => signal.removeEventListener("abort", listeners[index]));
+    const fail = (index) => {
+      cleanup();
+      reject(new ControlError(index < 3 ? "disposed" : "request_cancelled"));
+    };
+    signals.forEach((signal, index) => {
+      const listener = () => fail(index);
+      listeners.push(listener);
+      signal.addEventListener("abort", listener, { once: true });
+    });
+    const cancelled = signals.findIndex((signal) => signal.aborted);
+    if (cancelled !== -1)
+      fail(cancelled);
+    operation.then((value) => {
+      cleanup();
+      resolve(value);
+    }, (error) => {
+      cleanup();
+      reject(error);
+    });
+  });
+}
+function createControl(host) {
+  const client = observationClient(host.storage);
+  const disposal = new AbortController;
+  let disposed = false;
+  const assertAlive = (context) => {
+    if (disposed || host.signal.aborted || context?.signal.aborted)
+      throw new ControlError("disposed");
+    if (context?.requestSignal.aborted || context?.request.signal.aborted)
+      throw new ControlError("request_cancelled");
+  };
+  const dispose = () => {
+    disposed = true;
+    disposal.abort();
+    host.signal.removeEventListener("abort", dispose);
+  };
+  host.signal.addEventListener("abort", dispose, { once: true });
+  return {
+    api: [{ path: "/stats", methods: ["GET"], handler: "getStats", async invoke(context) {
+      try {
+        assertAlive(context);
+        if (context.request.method !== "GET")
+          return Response.json({ error: "method_not_allowed" }, { status: 405, headers: { allow: "GET" } });
+        const params = new URL(context.request.url).searchParams;
+        const range = params.get("range") ?? "24h";
+        if ([...params.keys()].some((key) => key !== "range") || params.getAll("range").length > 1 || !["1h", "12h", "24h"].includes(range))
+          throw new ControlError("invalid_input");
+        const readIncomplete = async () => {
+          const storage = host.storage;
+          const strict = storage.readStrict ? await storage.readStrict(TPS_REPORTING_INCOMPLETE_KEY) : null;
+          const value = strict === null ? await storage.get(TPS_REPORTING_INCOMPLETE_KEY) : strict.found ? strict.value : null;
+          if (value !== null && value !== false && value !== true)
+            throw new Error("invalid reporting status");
+          return value === true;
+        };
+        const [stats, incomplete] = await abortable(Promise.all([
+          client.queryStats({ range, asOfMs: Date.now() }),
+          readIncomplete()
+        ]), [disposal.signal, host.signal, context.signal, context.requestSignal, context.request.signal]);
+        assertAlive(context);
+        return Response.json({ ...stats, reportingIncomplete: stats.reportingIncomplete || incomplete });
+      } catch (error) {
+        const code = error instanceof ControlError ? error.code : "storage_unavailable";
+        const status = code === "invalid_input" || code === "request_cancelled" ? 400 : code === "disposed" ? 409 : 503;
+        return Response.json({ error: code }, { status });
+      }
+    } }],
+    rpc: [],
+    start() {
+      assertAlive();
+    },
+    dispose
+  };
+}
+var control_default = { createControl };
+export {
+  createControl,
+  createObservationAdapter,
+  control_default as default
+};
+
+//# debugId=6D69678B8F3EC90164756E2164756E21

package/dist/plugins/llm-tps/control.js.map

@@ -0,0 +1,13 @@
+{
+  "version": 3,
+  "sources": ["../../../../../plugins/llm-tps/contract.ts", "../../../../../plugins/llm-tps/server/metrics.ts", "../../../../../plugins/llm-tps/server/storage.ts", "../../../../../plugins/llm-tps/server/control.ts"],
+  "sourcesContent": [
+    "export type TpsRange = '1h' | '12h' | '24h';\n\nexport interface TpsSample {\n  readonly attemptId: string;\n  readonly requestId: string;\n  readonly routeId: string;\n  /** Configured endpoint ID, or an opaque hash when the host identity is a URL. */\n  readonly endpointId: string;\n  readonly finishedAtMs: number;\n  readonly outputTokens: number;\n  readonly generationDurationMs: number;\n  readonly outputSource: 'official' | 'estimated';\n}\n\nexport interface TpsAggregate {\n  readonly tps: number | null;\n  readonly outputTokens: number;\n  readonly generationDurationMs: number;\n  readonly sampleCount: number;\n  readonly estimatedSampleCount: number;\n}\n\nexport interface TpsEndpointStats extends TpsAggregate {\n  readonly routeId: string;\n  readonly endpointId: string;\n}\n\nexport interface TpsStatsResponse {\n  readonly range: TpsRange;\n  readonly asOfMs: number;\n  readonly total: TpsAggregate;\n  readonly endpoints: readonly TpsEndpointStats[];\n  readonly reportingIncomplete: boolean;\n}\n\nexport const TPS_RETENTION_MS = 31 * 24 * 60 * 60 * 1000;\nexport const TPS_REPORTING_INCOMPLETE_KEY = 'reporting:incomplete';\n",
+    "import { createHash } from 'node:crypto';\nimport type { TokenMeteringResult } from '@jeffusion/bungee-core/plugin';\nimport type { TpsAggregate, TpsRange, TpsSample } from '../contract';\n\nexport function isUrlIdentity(value: string): boolean {\n  return /^(?:[a-z][a-z\\d+.-]*:)?\\/\\//i.test(value);\n}\n\nexport function endpointIdentity(upstreamId: string): string {\n  return isUrlIdentity(upstreamId)\n    ? `endpoint-${createHash('sha256').update(upstreamId).digest('hex').slice(0, 16)}`\n    : upstreamId;\n}\n\nexport function safeCount(value: unknown, name: string): number {\n  if (typeof value !== 'number' || !Number.isSafeInteger(value) || value < 0) throw new RangeError(`invalid llm-tps ${name}`);\n  return value;\n}\n\nexport function identifier(value: unknown, name: string): asserts value is string {\n  if (typeof value !== 'string' || !value.trim()) throw new TypeError(`invalid llm-tps ${name}`);\n}\n\nexport function validateSample(sample: TpsSample): void {\n  for (const key of ['attemptId', 'requestId', 'routeId', 'endpointId'] as const) identifier(sample[key], key);\n  if (isUrlIdentity(sample.endpointId)) throw new TypeError('llm-tps endpoint must be opaque');\n  safeCount(sample.finishedAtMs, 'finishedAtMs');\n  safeCount(sample.outputTokens, 'outputTokens');\n  if (!Number.isFinite(sample.generationDurationMs) || sample.generationDurationMs <= 0) throw new RangeError('invalid llm-tps duration');\n  if (sample.outputSource !== 'official' && sample.outputSource !== 'estimated') throw new TypeError('invalid llm-tps output source');\n  if (!Number.isFinite(sample.outputTokens / sample.generationDurationMs * 1000)) throw new RangeError('invalid llm-tps rate');\n}\n\n/** Unknown input usage does not invalidate observable output usage. */\nexport function sampleFromResult(result: TokenMeteringResult, routeId: string): TpsSample | null {\n  const output = result.generationOutput;\n  if (result.routeId !== routeId || !output\n    || (output.source !== 'official' && output.source !== 'estimated')\n    || !Number.isSafeInteger(output.tokens) || output.tokens < 0\n    || !Number.isFinite(result.generationDurationMs) || result.generationDurationMs! <= 0) return null;\n  const sample: TpsSample = {\n    attemptId: result.attemptId, requestId: result.requestId, routeId: result.routeId,\n    endpointId: endpointIdentity(result.upstreamId), finishedAtMs: result.finishedAtMs,\n    outputTokens: output.tokens, generationDurationMs: result.generationDurationMs!, outputSource: output.source,\n  };\n  validateSample(sample);\n  return sample;\n}\n\nexport function rangeDuration(range: TpsRange): number {\n  if (range === '1h') return 3_600_000;\n  if (range === '12h') return 43_200_000;\n  if (range === '24h') return 86_400_000;\n  throw new RangeError('invalid llm-tps range');\n}\n\nexport function aggregate(row: Record<string, unknown>): TpsAggregate {\n  const outputTokens = safeCount(row.outputTokens, 'aggregate tokens');\n  const sampleCount = safeCount(row.sampleCount, 'sample count');\n  const estimatedSampleCount = safeCount(row.estimatedSampleCount, 'estimated sample count');\n  const duration = row.generationDurationMs;\n  if (typeof duration !== 'number' || !Number.isFinite(duration) || duration < 0\n    || (sampleCount > 0 && duration <= 0) || estimatedSampleCount > sampleCount\n    || (sampleCount === 0 && (outputTokens !== 0 || duration !== 0 || estimatedSampleCount !== 0))) throw new RangeError('invalid llm-tps aggregate');\n  const tps = sampleCount === 0 ? null : outputTokens / duration * 1000;\n  if (tps !== null && (!Number.isFinite(tps) || tps < 0)) throw new RangeError('invalid llm-tps aggregate rate');\n  return { tps, outputTokens, generationDurationMs: duration, sampleCount, estimatedSampleCount };\n}\n",
+    "import type { PluginObservationStorage, PluginStorage } from '@jeffusion/bungee-core/plugin';\nimport { TPS_RETENTION_MS, type TpsRange, type TpsSample, type TpsStatsResponse } from '../contract';\nimport { aggregate, identifier, isUrlIdentity, rangeDuration, safeCount, validateSample } from './metrics';\n\nexport interface TpsObservationClient {\n  recordSample(sample: TpsSample): Promise<void>;\n  queryStats(input: { range: TpsRange; asOfMs: number }): Promise<TpsStatsResponse>;\n}\n\n/** Runtime and control use only the host's namespaced observation adapter. */\nexport function observationClient(storage: PluginStorage): TpsObservationClient {\n  if (typeof storage.invokeObservation !== 'function') throw new Error('llm-tps observation storage is unavailable');\n  const invoke = storage.invokeObservation.bind(storage);\n  return {\n    recordSample: sample => invoke<void>('recordSample', [sample]),\n    queryStats: input => invoke<TpsStatsResponse>('queryStats', [input]),\n  };\n}\n\nconst metrics = `COALESCE(SUM(output_tokens), 0) AS outputTokens,\n  COALESCE(SUM(generation_duration_ms), 0) AS generationDurationMs,\n  COUNT(*) AS sampleCount, COALESCE(SUM(output_source = 'estimated'), 0) AS estimatedSampleCount,\n  COALESCE(SUM(CASE WHEN typeof(output_tokens) != 'integer' OR output_tokens < 0 OR output_tokens > 9007199254740991\n    OR typeof(generation_duration_ms) NOT IN ('integer', 'real') OR NOT(generation_duration_ms > 0)\n    OR generation_duration_ms > 1.7976931348623157e308 OR output_source NOT IN ('official','estimated')\n    OR typeof(finished_at_ms) != 'integer' OR finished_at_ms < 0 OR finished_at_ms > 9007199254740991\n    THEN 1 ELSE 0 END), 0) AS invalidSamples`;\n\n/** SQL is confined to the trusted access.db adapter callback. */\nexport function createObservationAdapter(observation: PluginObservationStorage): TpsObservationClient {\n  return Object.freeze({\n    async recordSample(sample: TpsSample): Promise<void> {\n      validateSample(sample);\n      observation.withDatabase(db => {\n        db.transaction(() => {\n          db.query(`INSERT OR IGNORE INTO llm_tps_samples\n            (attempt_id, request_id, route_id, endpoint_id, finished_at_ms, output_tokens, generation_duration_ms, output_source)\n            VALUES (?, ?, ?, ?, ?, ?, ?, ?)`).run(sample.attemptId, sample.requestId, sample.routeId, sample.endpointId,\n            sample.finishedAtMs, sample.outputTokens, sample.generationDurationMs, sample.outputSource);\n          db.query(`DELETE FROM llm_tps_samples WHERE attempt_id IN (\n            SELECT attempt_id FROM llm_tps_samples INDEXED BY idx_llm_tps_samples_finished\n            WHERE finished_at_ms < ? ORDER BY finished_at_ms LIMIT 500\n          )`).run(Date.now() - TPS_RETENTION_MS);\n        }).immediate();\n      });\n    },\n    async queryStats(input: { range: TpsRange; asOfMs: number }): Promise<TpsStatsResponse> {\n      safeCount(input.asOfMs, 'asOfMs');\n      const start = input.asOfMs - rangeDuration(input.range);\n      return observation.withDatabase(db => db.transaction(() => {\n        const where = 'WHERE finished_at_ms >= ? AND finished_at_ms <= ?';\n        const totalRow = db.query<Record<string, unknown>, [number, number]>(`SELECT ${metrics} FROM llm_tps_samples ${where}`).get(start, input.asOfMs)!;\n        const rows = db.query<Record<string, unknown>, [number, number]>(`SELECT route_id AS routeId, endpoint_id AS endpointId, ${metrics}\n          FROM llm_tps_samples ${where} GROUP BY route_id, endpoint_id`).all(start, input.asOfMs);\n        const checked = (row: Record<string, unknown>) => {\n          if (safeCount(row.invalidSamples, 'invalid sample count') !== 0) throw new RangeError('invalid persisted llm-tps sample');\n          return aggregate(row);\n        };\n        const total = checked(totalRow);\n        const endpoints = rows.map(row => {\n          identifier(row.routeId, 'routeId'); identifier(row.endpointId, 'endpointId');\n          if (isUrlIdentity(row.endpointId)) throw new TypeError('invalid persisted llm-tps endpoint');\n          return { routeId: row.routeId, endpointId: row.endpointId, ...checked(row) };\n        }).sort((a, b) => (b.tps! - a.tps!) || compare(a.routeId, b.routeId) || compare(a.endpointId, b.endpointId));\n        return { range: input.range, asOfMs: input.asOfMs, total, endpoints, reportingIncomplete: false };\n      })());\n    },\n  });\n}\n\nfunction compare(a: string, b: string): number { return a < b ? -1 : a > b ? 1 : 0; }\n",
+    "import type { ControlApiHandlerContext, ControlHostContext, ControlPlugin, PluginControl } from '@jeffusion/bungee-core/plugin';\nimport { TPS_REPORTING_INCOMPLETE_KEY, type TpsRange } from '../contract';\nimport { observationClient } from './storage';\nexport { createObservationAdapter } from './storage';\n\nclass ControlError extends Error {\n  constructor(readonly code: 'disposed' | 'request_cancelled' | 'invalid_input') { super(code); }\n}\n\nfunction abortable<T>(operation: Promise<T>, signals: readonly AbortSignal[]): Promise<T> {\n  return new Promise<T>((resolve, reject) => {\n    const listeners: Array<() => void> = [];\n    const cleanup = () => signals.forEach((signal, index) => signal.removeEventListener('abort', listeners[index]!));\n    const fail = (index: number) => { cleanup(); reject(new ControlError(index < 3 ? 'disposed' : 'request_cancelled')); };\n    signals.forEach((signal, index) => {\n      const listener = () => fail(index); listeners.push(listener);\n      signal.addEventListener('abort', listener, { once: true });\n    });\n    const cancelled = signals.findIndex(signal => signal.aborted);\n    if (cancelled !== -1) fail(cancelled);\n    operation.then(value => { cleanup(); resolve(value); }, error => { cleanup(); reject(error); });\n  });\n}\n\nexport function createControl(host: ControlHostContext): PluginControl {\n  const client = observationClient(host.storage);\n  const disposal = new AbortController();\n  let disposed = false;\n  const assertAlive = (context?: ControlApiHandlerContext) => {\n    if (disposed || host.signal.aborted || context?.signal.aborted) throw new ControlError('disposed');\n    if (context?.requestSignal.aborted || context?.request.signal.aborted) throw new ControlError('request_cancelled');\n  };\n  const dispose = () => { disposed = true; disposal.abort(); host.signal.removeEventListener('abort', dispose); };\n  host.signal.addEventListener('abort', dispose, { once: true });\n  return {\n    api: [{ path: '/stats', methods: ['GET'], handler: 'getStats', async invoke(context) {\n      try {\n        assertAlive(context);\n        if (context.request.method !== 'GET') return Response.json({ error: 'method_not_allowed' }, { status: 405, headers: { allow: 'GET' } });\n        const params = new URL(context.request.url).searchParams;\n        const range = params.get('range') ?? '24h';\n        if ([...params.keys()].some(key => key !== 'range') || params.getAll('range').length > 1\n          || !['1h', '12h', '24h'].includes(range)) throw new ControlError('invalid_input');\n        const readIncomplete = async () => {\n          const storage = host.storage;\n          const strict = storage.readStrict ? await storage.readStrict<unknown>(TPS_REPORTING_INCOMPLETE_KEY) : null;\n          const value = strict === null ? await storage.get<unknown>(TPS_REPORTING_INCOMPLETE_KEY) : strict.found ? strict.value : null;\n          if (value !== null && value !== false && value !== true) throw new Error('invalid reporting status');\n          return value === true;\n        };\n        const [stats, incomplete] = await abortable(Promise.all([\n          client.queryStats({ range: range as TpsRange, asOfMs: Date.now() }), readIncomplete(),\n        ]), [disposal.signal, host.signal, context.signal, context.requestSignal, context.request.signal]);\n        assertAlive(context);\n        return Response.json({ ...stats, reportingIncomplete: stats.reportingIncomplete || incomplete });\n      } catch (error) {\n        const code = error instanceof ControlError ? error.code : 'storage_unavailable';\n        const status = code === 'invalid_input' || code === 'request_cancelled' ? 400 : code === 'disposed' ? 409 : 503;\n        return Response.json({ error: code }, { status });\n      }\n    } }],\n    rpc: [],\n    start() { assertAlive(); },\n    dispose,\n  };\n}\n\nexport default { createControl } satisfies ControlPlugin;\n"
+  ],
+  "mappings": ";;AAmCO,IAAM,mBAAmB,KAAK,KAAK,KAAK,KAAK;AAC7C,IAAM,+BAA+B;;;AChCrC,SAAS,aAAa,CAAC,OAAwB;AAAA,EACpD,OAAO,+BAA+B,KAAK,KAAK;AAAA;AAS3C,SAAS,SAAS,CAAC,OAAgB,MAAsB;AAAA,EAC9D,IAAI,OAAO,UAAU,YAAY,CAAC,OAAO,cAAc,KAAK,KAAK,QAAQ;AAAA,IAAG,MAAM,IAAI,WAAW,mBAAmB,MAAM;AAAA,EAC1H,OAAO;AAAA;AAGF,SAAS,UAAU,CAAC,OAAgB,MAAuC;AAAA,EAChF,IAAI,OAAO,UAAU,YAAY,CAAC,MAAM,KAAK;AAAA,IAAG,MAAM,IAAI,UAAU,mBAAmB,MAAM;AAAA;AAGxF,SAAS,cAAc,CAAC,QAAyB;AAAA,EACtD,WAAW,OAAO,CAAC,aAAa,aAAa,WAAW,YAAY;AAAA,IAAY,WAAW,OAAO,MAAM,GAAG;AAAA,EAC3G,IAAI,cAAc,OAAO,UAAU;AAAA,IAAG,MAAM,IAAI,UAAU,iCAAiC;AAAA,EAC3F,UAAU,OAAO,cAAc,cAAc;AAAA,EAC7C,UAAU,OAAO,cAAc,cAAc;AAAA,EAC7C,IAAI,CAAC,OAAO,SAAS,OAAO,oBAAoB,KAAK,OAAO,wBAAwB;AAAA,IAAG,MAAM,IAAI,WAAW,0BAA0B;AAAA,EACtI,IAAI,OAAO,iBAAiB,cAAc,OAAO,iBAAiB;AAAA,IAAa,MAAM,IAAI,UAAU,+BAA+B;AAAA,EAClI,IAAI,CAAC,OAAO,SAAS,OAAO,eAAe,OAAO,uBAAuB,IAAI;AAAA,IAAG,MAAM,IAAI,WAAW,sBAAsB;AAAA;AAmBtH,SAAS,aAAa,CAAC,OAAyB;AAAA,EACrD,IAAI,UAAU;AAAA,IAAM,OAAO;AAAA,EAC3B,IAAI,UAAU;AAAA,IAAO,OAAO;AAAA,EAC5B,IAAI,UAAU;AAAA,IAAO,OAAO;AAAA,EAC5B,MAAM,IAAI,WAAW,uBAAuB;AAAA;AAGvC,SAAS,SAAS,CAAC,KAA4C;AAAA,EACpE,MAAM,eAAe,UAAU,IAAI,cAAc,kBAAkB;AAAA,EACnE,MAAM,cAAc,UAAU,IAAI,aAAa,cAAc;AAAA,EAC7D,MAAM,uBAAuB,UAAU,IAAI,sBAAsB,wBAAwB;AAAA,EACzF,MAAM,WAAW,IAAI;AAAA,EACrB,IAAI,OAAO,aAAa,YAAY,CAAC,OAAO,SAAS,QAAQ,KAAK,WAAW,KACvE,cAAc,KAAK,YAAY,KAAM,uBAAuB,eAC5D,gBAAgB,MAAM,iBAAiB,KAAK,aAAa,KAAK,yBAAyB;AAAA,IAAK,MAAM,IAAI,WAAW,2BAA2B;AAAA,EAClJ,MAAM,MAAM,gBAAgB,IAAI,OAAO,eAAe,WAAW;AAAA,EACjE,IAAI,QAAQ,SAAS,CAAC,OAAO,SAAS,GAAG,KAAK,MAAM;AAAA,IAAI,MAAM,IAAI,WAAW,gCAAgC;AAAA,EAC7G,OAAO,EAAE,KAAK,cAAc,sBAAsB,UAAU,aAAa,qBAAqB;AAAA;;;ACxDzF,SAAS,iBAAiB,CAAC,SAA8C;AAAA,EAC9E,IAAI,OAAO,QAAQ,sBAAsB;AAAA,IAAY,MAAM,IAAI,MAAM,4CAA4C;AAAA,EACjH,MAAM,SAAS,QAAQ,kBAAkB,KAAK,OAAO;AAAA,EACrD,OAAO;AAAA,IACL,cAAc,YAAU,OAAa,gBAAgB,CAAC,MAAM,CAAC;AAAA,IAC7D,YAAY,WAAS,OAAyB,cAAc,CAAC,KAAK,CAAC;AAAA,EACrE;AAAA;AAGF,IAAM,UAAU;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAUT,SAAS,wBAAwB,CAAC,aAA6D;AAAA,EACpG,OAAO,OAAO,OAAO;AAAA,SACb,aAAY,CAAC,QAAkC;AAAA,MACnD,eAAe,MAAM;AAAA,MACrB,YAAY,aAAa,QAAM;AAAA,QAC7B,GAAG,YAAY,MAAM;AAAA,UACnB,GAAG,MAAM;AAAA;AAAA,4CAEyB,EAAE,IAAI,OAAO,WAAW,OAAO,WAAW,OAAO,SAAS,OAAO,YACjG,OAAO,cAAc,OAAO,cAAc,OAAO,sBAAsB,OAAO,YAAY;AAAA,UAC5F,GAAG,MAAM;AAAA;AAAA;AAAA,YAGP,EAAE,IAAI,KAAK,IAAI,IAAI,gBAAgB;AAAA,SACtC,EAAE,UAAU;AAAA,OACd;AAAA;AAAA,SAEG,WAAU,CAAC,OAAuE;AAAA,MACtF,UAAU,MAAM,QAAQ,QAAQ;AAAA,MAChC,MAAM,QAAQ,MAAM,SAAS,cAAc,MAAM,KAAK;AAAA,MACtD,OAAO,YAAY,aAAa,QAAM,GAAG,YAAY,MAAM;AAAA,QACzD,MAAM,QAAQ;AAAA,QACd,MAAM,WAAW,GAAG,MAAiD,UAAU,gCAAgC,OAAO,EAAE,IAAI,OAAO,MAAM,MAAM;AAAA,QAC/I,MAAM,OAAO,GAAG,MAAiD,0DAA0D;AAAA,iCAClG,sCAAsC,EAAE,IAAI,OAAO,MAAM,MAAM;AAAA,QACxF,MAAM,UAAU,CAAC,QAAiC;AAAA,UAChD,IAAI,UAAU,IAAI,gBAAgB,sBAAsB,MAAM;AAAA,YAAG,MAAM,IAAI,WAAW,kCAAkC;AAAA,UACxH,OAAO,UAAU,GAAG;AAAA;AAAA,QAEtB,MAAM,QAAQ,QAAQ,QAAQ;AAAA,QAC9B,MAAM,YAAY,KAAK,IAAI,SAAO;AAAA,UAChC,WAAW,IAAI,SAAS,SAAS;AAAA,UAAG,WAAW,IAAI,YAAY,YAAY;AAAA,UAC3E,IAAI,cAAc,IAAI,UAAU;AAAA,YAAG,MAAM,IAAI,UAAU,oCAAoC;AAAA,UAC3F,OAAO,EAAE,SAAS,IAAI,SAAS,YAAY,IAAI,eAAe,QAAQ,GAAG,EAAE;AAAA,SAC5E,EAAE,KAAK,CAAC,GAAG,MAAO,EAAE,MAAO,EAAE,OAAS,QAAQ,EAAE,SAAS,EAAE,OAAO,KAAK,QAAQ,EAAE,YAAY,EAAE,UAAU,CAAC;AAAA,QAC3G,OAAO,EAAE,OAAO,MAAM,OAAO,QAAQ,MAAM,QAAQ,OAAO,WAAW,qBAAqB,MAAM;AAAA,OACjG,EAAE,CAAC;AAAA;AAAA,EAER,CAAC;AAAA;AAGH,SAAS,OAAO,CAAC,GAAW,GAAmB;AAAA,EAAE,OAAO,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI;AAAA;;;ACjEjF,MAAM,qBAAqB,MAAM;AAAA,EACV;AAAA,EAArB,WAAW,CAAU,MAA0D;AAAA,IAAE,MAAM,IAAI;AAAA,IAAtE;AAAA;AACvB;AAEA,SAAS,SAAY,CAAC,WAAuB,SAA6C;AAAA,EACxF,OAAO,IAAI,QAAW,CAAC,SAAS,WAAW;AAAA,IACzC,MAAM,YAA+B,CAAC;AAAA,IACtC,MAAM,UAAU,MAAM,QAAQ,QAAQ,CAAC,QAAQ,UAAU,OAAO,oBAAoB,SAAS,UAAU,MAAO,CAAC;AAAA,IAC/G,MAAM,OAAO,CAAC,UAAkB;AAAA,MAAE,QAAQ;AAAA,MAAG,OAAO,IAAI,aAAa,QAAQ,IAAI,aAAa,mBAAmB,CAAC;AAAA;AAAA,IAClH,QAAQ,QAAQ,CAAC,QAAQ,UAAU;AAAA,MACjC,MAAM,WAAW,MAAM,KAAK,KAAK;AAAA,MAAG,UAAU,KAAK,QAAQ;AAAA,MAC3D,OAAO,iBAAiB,SAAS,UAAU,EAAE,MAAM,KAAK,CAAC;AAAA,KAC1D;AAAA,IACD,MAAM,YAAY,QAAQ,UAAU,YAAU,OAAO,OAAO;AAAA,IAC5D,IAAI,cAAc;AAAA,MAAI,KAAK,SAAS;AAAA,IACpC,UAAU,KAAK,WAAS;AAAA,MAAE,QAAQ;AAAA,MAAG,QAAQ,KAAK;AAAA,OAAM,WAAS;AAAA,MAAE,QAAQ;AAAA,MAAG,OAAO,KAAK;AAAA,KAAI;AAAA,GAC/F;AAAA;AAGI,SAAS,aAAa,CAAC,MAAyC;AAAA,EACrE,MAAM,SAAS,kBAAkB,KAAK,OAAO;AAAA,EAC7C,MAAM,WAAW,IAAI;AAAA,EACrB,IAAI,WAAW;AAAA,EACf,MAAM,cAAc,CAAC,YAAuC;AAAA,IAC1D,IAAI,YAAY,KAAK,OAAO,WAAW,SAAS,OAAO;AAAA,MAAS,MAAM,IAAI,aAAa,UAAU;AAAA,IACjG,IAAI,SAAS,cAAc,WAAW,SAAS,QAAQ,OAAO;AAAA,MAAS,MAAM,IAAI,aAAa,mBAAmB;AAAA;AAAA,EAEnH,MAAM,UAAU,MAAM;AAAA,IAAE,WAAW;AAAA,IAAM,SAAS,MAAM;AAAA,IAAG,KAAK,OAAO,oBAAoB,SAAS,OAAO;AAAA;AAAA,EAC3G,KAAK,OAAO,iBAAiB,SAAS,SAAS,EAAE,MAAM,KAAK,CAAC;AAAA,EAC7D,OAAO;AAAA,IACL,KAAK,CAAC,EAAE,MAAM,UAAU,SAAS,CAAC,KAAK,GAAG,SAAS,iBAAkB,OAAM,CAAC,SAAS;AAAA,MACnF,IAAI;AAAA,QACF,YAAY,OAAO;AAAA,QACnB,IAAI,QAAQ,QAAQ,WAAW;AAAA,UAAO,OAAO,SAAS,KAAK,EAAE,OAAO,qBAAqB,GAAG,EAAE,QAAQ,KAAK,SAAS,EAAE,OAAO,MAAM,EAAE,CAAC;AAAA,QACtI,MAAM,SAAS,IAAI,IAAI,QAAQ,QAAQ,GAAG,EAAE;AAAA,QAC5C,MAAM,QAAQ,OAAO,IAAI,OAAO,KAAK;AAAA,QACrC,IAAI,CAAC,GAAG,OAAO,KAAK,CAAC,EAAE,KAAK,SAAO,QAAQ,OAAO,KAAK,OAAO,OAAO,OAAO,EAAE,SAAS,KAClF,CAAC,CAAC,MAAM,OAAO,KAAK,EAAE,SAAS,KAAK;AAAA,UAAG,MAAM,IAAI,aAAa,eAAe;AAAA,QAClF,MAAM,iBAAiB,YAAY;AAAA,UACjC,MAAM,UAAU,KAAK;AAAA,UACrB,MAAM,SAAS,QAAQ,aAAa,MAAM,QAAQ,WAAoB,4BAA4B,IAAI;AAAA,UACtG,MAAM,QAAQ,WAAW,OAAO,MAAM,QAAQ,IAAa,4BAA4B,IAAI,OAAO,QAAQ,OAAO,QAAQ;AAAA,UACzH,IAAI,UAAU,QAAQ,UAAU,SAAS,UAAU;AAAA,YAAM,MAAM,IAAI,MAAM,0BAA0B;AAAA,UACnG,OAAO,UAAU;AAAA;AAAA,QAEnB,OAAO,OAAO,cAAc,MAAM,UAAU,QAAQ,IAAI;AAAA,UACtD,OAAO,WAAW,EAAE,OAA0B,QAAQ,KAAK,IAAI,EAAE,CAAC;AAAA,UAAG,eAAe;AAAA,QACtF,CAAC,GAAG,CAAC,SAAS,QAAQ,KAAK,QAAQ,QAAQ,QAAQ,QAAQ,eAAe,QAAQ,QAAQ,MAAM,CAAC;AAAA,QACjG,YAAY,OAAO;AAAA,QACnB,OAAO,SAAS,KAAK,KAAK,OAAO,qBAAqB,MAAM,uBAAuB,WAAW,CAAC;AAAA,QAC/F,OAAO,OAAO;AAAA,QACd,MAAM,OAAO,iBAAiB,eAAe,MAAM,OAAO;AAAA,QAC1D,MAAM,SAAS,SAAS,mBAAmB,SAAS,sBAAsB,MAAM,SAAS,aAAa,MAAM;AAAA,QAC5G,OAAO,SAAS,KAAK,EAAE,OAAO,KAAK,GAAG,EAAE,OAAO,CAAC;AAAA;AAAA,MAElD,CAAC;AAAA,IACH,KAAK,CAAC;AAAA,IACN,KAAK,GAAG;AAAA,MAAE,YAAY;AAAA;AAAA,IACtB;AAAA,EACF;AAAA;AAGF,IAAe,oBAAE,cAAc;",
+  "debugId": "6D69678B8F3EC90164756E2164756E21",
+  "names": []
+}

package/dist/plugins/llm-tps/index.js

@@ -0,0 +1,286 @@
+// @bun
+// packages/core/src/plugin-logger.ts
+var sinkKey = Symbol.for("@jeffusion/bungee/plugin-log-sink");
+function createLogger(bindings = {}) {
+  const method = (level) => (value, message) => {
+    const sink = globalThis[sinkKey];
+    if (sink) {
+      if (typeof value === "string")
+        sink[level](bindings, value);
+      else
+        sink[level]({ ...bindings, ...value }, message);
+    } else if (level !== "debug" || process.env.LOG_LEVEL === "debug") {
+      const write = level === "fatal" ? console.error : console[level];
+      write(message ?? (typeof value === "string" ? value : ""), typeof value === "string" ? bindings : { ...bindings, ...value });
+    }
+  };
+  return {
+    info: method("info"),
+    warn: method("warn"),
+    error: method("error"),
+    debug: method("debug"),
+    fatal: method("fatal"),
+    child: (extra) => createLogger({ ...bindings, ...extra })
+  };
+}
+var logger = createLogger();
+
+// packages/core/src/hooks/impl.ts
+var WATERFALL_VIEW_TRANSFORM = Symbol("waterfall-view-transform");
+// packages/core/src/plugin.types.ts
+function definePlugin(plugin) {
+  return plugin;
+}
+// packages/core/src/plugin-extensions.ts
+var ANONYMOUS_PRINCIPAL = Object.freeze({ domain: "anonymous", keyId: "", credentialVersion: 0 });
+// packages/core/src/plugin-durable-state.ts
+var MAX_BYTES = 1024 * 1024;
+// packages/core/src/plugin-services/host-rpc.ts
+import { AsyncLocalStorage } from "async_hooks";
+
+// packages/core/src/plugin-services/wire-contract.ts
+var RPC_JSON_MAX_BYTES = 64 * 1024;
+var RPC_CONTRACT_MAX_BYTES = 256 * 1024;
+var RPC_CALL_PURPOSES = new Set(["bootstrap", "background", "management", "request", "attempt"]);
+var RPC_COMMAND_DEDUPLICATION = new Set(["local-transaction", "external-contract", "none"]);
+var RPC_COMMAND_POLICY_KEYS = new Set(["deduplication", "resultRetentionMs", "quotaBytes", "maxResultBytes"]);
+
+// packages/core/src/plugin-services/rpc-runtime.ts
+var PROCESSES = new Set(["control", "worker", "ingress"]);
+var SCOPES = new Set(["global", "binding"]);
+var PURPOSES = new Set(["bootstrap", "background", "management", "request", "attempt"]);
+var ENDPOINTS = new WeakMap;
+var REVOKED = new WeakMap;
+
+// packages/core/src/plugin-services/host-rpc.ts
+var SERVICE_ERROR_MESSAGES = {
+  unauthorized: "RPC call is not authorized",
+  not_ready: "RPC endpoint is not ready",
+  retired: "RPC endpoint is retiring",
+  revoked: "RPC endpoint generation was revoked",
+  unsupported_method: "RPC method is not published",
+  wrong_purpose: "RPC method is not published for this purpose",
+  invalid_input: "RPC input is not valid",
+  invalid_output: "RPC output is not valid",
+  invalid_operation_id: "RPC command operation id is invalid",
+  overloaded: "RPC runtime is at capacity",
+  cancelled: "RPC call was cancelled",
+  timeout: "RPC call deadline exceeded",
+  unknown: "RPC command outcome is unknown",
+  capability_unavailable: "RPC command execution is unavailable",
+  closed: "RPC runtime is closed",
+  failed: "RPC call failed",
+  conflict: "RPC command identity conflicts with an existing record",
+  expired: "RPC command result has expired",
+  pending: "RPC command is still pending",
+  rejected: "RPC command was rejected",
+  missing: "RPC command has no stored result",
+  storage_failure: "RPC command storage failed",
+  deadlock: "RPC call chain would deadlock",
+  call_depth_exceeded: "RPC call depth exceeded",
+  undeclared: "RPC service is not declared by the manifest",
+  unavailable: "RPC service is unavailable",
+  ambiguous_placement: "RPC service placement is ambiguous",
+  invalid_registration: "RPC registration is invalid"
+};
+var SERVICE_ERROR_CODES = new Set(Object.keys(SERVICE_ERROR_MESSAGES));
+var EMPTY_OPTIONS = Object.freeze({});
+var FRAME_STORE = new AsyncLocalStorage;
+var CALLEE_INFO = new WeakMap;
+// packages/core/src/plugin-services/peer-channel-protocol.ts
+var CHANNEL_REQUEST_MAX_BYTES = 4 * 1024;
+var CHANNEL_RESULT_MAX_BYTES = 60 * 1024;
+var CHANNEL_MAX_CHUNK_BYTES = 60 * 1024;
+var PLUGIN_CHANNEL_LANES = Object.freeze(["event", "snapshot", "stream"]);
+var ERROR_CODES = new Set([
+  "unavailable",
+  "ambiguous",
+  "unauthorized",
+  "invalid",
+  "overloaded",
+  "expired",
+  "conflict",
+  "gap",
+  "truncated",
+  "rejected",
+  "cancelled",
+  "closed",
+  "unknown",
+  "failed"
+]);
+var RPC_CALL_PURPOSES2 = new Set(["bootstrap", "background", "management", "request", "attempt"]);
+
+// packages/core/src/plugin-services/peer-channel-hub.ts
+var CHANNEL_MAX_TRANSFER_BUFFER_BYTES = 4 * CHANNEL_MAX_CHUNK_BYTES;
+var CHANNEL_MAX_TOTAL_BUFFER_BYTES = 16 * CHANNEL_MAX_CHUNK_BYTES;
+var CHANNEL_STREAM_MAX_TOTAL_BYTES = 64 * 1024 * 1024;
+var CHANNEL_STREAM_MAX_CREDIT_BYTES = 8 * CHANNEL_MAX_CHUNK_BYTES;
+var CHANNEL_STREAM_DEFAULT_CREDIT_BYTES = 4 * CHANNEL_MAX_CHUNK_BYTES;
+var CHANNEL_SNAPSHOT_MAX_BYTES = 64 * 1024 * 1024;
+var CHANNEL_EVENT_MAX_PAYLOAD_BYTES = 60 * 1024;
+var EMPTY = new Uint8Array(0);
+var HANDLE_REGISTRATIONS = new WeakMap;
+
+// packages/core/src/plugin-services/snapshot-store.ts
+var DEFAULT_CHUNK_BYTES = 60 * 1024;
+var SNAPSHOT_FAMILY_REFS = new Map;
+
+// packages/core/src/plugin-services.ts
+var CHANNEL_KINDS = new Set(["events", "snapshot", "stream"]);
+var TOKEN_METERING_SERVICE_ID = "token-metering.v1";
+var TOKEN_METERING_CONTRACT_VERSION = 1;
+// packages/core/src/gateway/retry-action.ts
+var retryActionBrand = Symbol.for("@jeffusion/bungee/request-retry-action");
+// packages/core/src/token-stats-window.ts
+var HOUR = 3600000;
+var TOKEN_STATS_RETENTION_MS = 31 * 24 * HOUR;
+// plugins/llm-tps/contract.ts
+var TPS_RETENTION_MS = 31 * 24 * 60 * 60 * 1000;
+var TPS_REPORTING_INCOMPLETE_KEY = "reporting:incomplete";
+
+// plugins/llm-tps/server/metrics.ts
+import { createHash } from "crypto";
+function isUrlIdentity(value) {
+  return /^(?:[a-z][a-z\d+.-]*:)?\/\//i.test(value);
+}
+function endpointIdentity(upstreamId) {
+  return isUrlIdentity(upstreamId) ? `endpoint-${createHash("sha256").update(upstreamId).digest("hex").slice(0, 16)}` : upstreamId;
+}
+function safeCount(value, name) {
+  if (typeof value !== "number" || !Number.isSafeInteger(value) || value < 0)
+    throw new RangeError(`invalid llm-tps ${name}`);
+  return value;
+}
+function identifier(value, name) {
+  if (typeof value !== "string" || !value.trim())
+    throw new TypeError(`invalid llm-tps ${name}`);
+}
+function validateSample(sample) {
+  for (const key of ["attemptId", "requestId", "routeId", "endpointId"])
+    identifier(sample[key], key);
+  if (isUrlIdentity(sample.endpointId))
+    throw new TypeError("llm-tps endpoint must be opaque");
+  safeCount(sample.finishedAtMs, "finishedAtMs");
+  safeCount(sample.outputTokens, "outputTokens");
+  if (!Number.isFinite(sample.generationDurationMs) || sample.generationDurationMs <= 0)
+    throw new RangeError("invalid llm-tps duration");
+  if (sample.outputSource !== "official" && sample.outputSource !== "estimated")
+    throw new TypeError("invalid llm-tps output source");
+  if (!Number.isFinite(sample.outputTokens / sample.generationDurationMs * 1000))
+    throw new RangeError("invalid llm-tps rate");
+}
+function sampleFromResult(result, routeId) {
+  const output = result.generationOutput;
+  if (result.routeId !== routeId || !output || output.source !== "official" && output.source !== "estimated" || !Number.isSafeInteger(output.tokens) || output.tokens < 0 || !Number.isFinite(result.generationDurationMs) || result.generationDurationMs <= 0)
+    return null;
+  const sample = {
+    attemptId: result.attemptId,
+    requestId: result.requestId,
+    routeId: result.routeId,
+    endpointId: endpointIdentity(result.upstreamId),
+    finishedAtMs: result.finishedAtMs,
+    outputTokens: output.tokens,
+    generationDurationMs: result.generationDurationMs,
+    outputSource: output.source
+  };
+  validateSample(sample);
+  return sample;
+}
+
+// plugins/llm-tps/server/storage.ts
+function observationClient(storage) {
+  if (typeof storage.invokeObservation !== "function")
+    throw new Error("llm-tps observation storage is unavailable");
+  const invoke = storage.invokeObservation.bind(storage);
+  return {
+    recordSample: (sample) => invoke("recordSample", [sample]),
+    queryStats: (input) => invoke("queryStats", [input])
+  };
+}
+
+// plugins/llm-tps/server/index.ts
+var MAX_PENDING = 256;
+var LlmTpsPlugin = definePlugin(class {
+  static name = "llm-tps";
+  static version = "1.0.0";
+  unsubscribe;
+  accepting = false;
+  pending = 0;
+  writes = Promise.resolve();
+  reportingWrite;
+  reportingPersisted = false;
+  bodyRequirements() {
+    return { request: "none" };
+  }
+  register(_hooks) {}
+  async init(context) {
+    if (context.scope?.type !== "route" || !context.scope.routeId?.trim())
+      throw new Error("llm-tps requires a route binding");
+    if (!context.services)
+      throw new Error("llm-tps requires token-metering service");
+    const client = observationClient(context.storage);
+    const storage = context.storage.uncached?.() ?? context.storage;
+    const metering = context.services.consume("token-metering", TOKEN_METERING_SERVICE_ID, TOKEN_METERING_CONTRACT_VERSION);
+    const routeId = context.scope.routeId;
+    const incomplete = () => {
+      if (this.reportingPersisted || this.reportingWrite)
+        return;
+      this.reportingWrite = Promise.resolve().then(() => storage.set(TPS_REPORTING_INCOMPLETE_KEY, true)).then(() => {
+        this.reportingPersisted = true;
+      }).catch(() => {
+        try {
+          context.logger.error("llm-tps could not persist incomplete reporting status");
+        } catch {}
+      }).finally(() => {
+        this.reportingWrite = undefined;
+      });
+    };
+    this.accepting = true;
+    this.unsubscribe = metering.subscribe({
+      onResult: (result) => {
+        if (!this.accepting || result.routeId !== routeId)
+          return;
+        try {
+          if (result.observationIncomplete && result.generationDurationMs === undefined)
+            incomplete();
+          const sample = sampleFromResult(result, routeId);
+          if (!sample)
+            return;
+          if (this.pending >= MAX_PENDING) {
+            incomplete();
+            return;
+          }
+          this.pending++;
+          this.writes = this.writes.then(() => client.recordSample(sample)).catch(incomplete).finally(() => {
+            this.pending--;
+          });
+        } catch {
+          incomplete();
+        }
+      },
+      onFailure: (_error, result) => {
+        if (this.accepting && result.routeId === routeId)
+          incomplete();
+      }
+    });
+    context.services.onDispose(() => this.stopAccepting());
+  }
+  stopAccepting() {
+    this.accepting = false;
+    const unsubscribe = this.unsubscribe;
+    this.unsubscribe = undefined;
+    unsubscribe?.();
+  }
+  async onDestroy() {
+    this.stopAccepting();
+    await this.writes;
+    await this.reportingWrite;
+  }
+});
+var server_default = LlmTpsPlugin;
+export {
+  LlmTpsPlugin,
+  server_default as default
+};
+
+//# debugId=32378BFF08D1414E64756E2164756E21