llm-simple-router 0.5.4 → 0.5.6

package/dist/admin/metrics.js

@@ -33,13 +33,13 @@ const TimeseriesQuerySchema = Type.Object({
     end_time: Type.Optional(Type.String()),
 });
 const DASHBOARD_PERIODS = new Set(["window", "weekly", "monthly"]);
-function resolveMetricsTime(query, db, routerKeyId) {
+function resolveMetricsTime(query, db, routerKeyId, providerId) {
     if (query.start_time && query.end_time) {
         return { startTime: query.start_time, endTime: query.end_time, legacyPeriod: "30d" };
     }
     const period = query.period ?? "weekly";
     if (DASHBOARD_PERIODS.has(period)) {
-        const range = resolveTimeRange(period, db, routerKeyId);
+        const range = resolveTimeRange(period, db, routerKeyId, providerId);
         return { startTime: range.startTime, endTime: range.endTime, legacyPeriod: "5h" };
     }
     return { legacyPeriod: period };
@@ -48,14 +48,14 @@ export const adminMetricsRoutes = (app, options, done) => {
     const { db } = options;
     app.get("/admin/api/metrics/summary", { schema: { querystring: SummaryQuerySchema } }, async (request, reply) => {
         const query = request.query;
-        const { startTime, endTime, legacyPeriod } = resolveMetricsTime(query, db, query.router_key_id);
+        const { startTime, endTime, legacyPeriod } = resolveMetricsTime(query, db, query.router_key_id, query.provider_id);
         const summary = getMetricsSummary(db, legacyPeriod, query.provider_id, query.backend_model, query.router_key_id, startTime, endTime);
         return reply.send(summary);
     });
     app.get("/admin/api/metrics/timeseries", { schema: { querystring: TimeseriesQuerySchema } }, async (request, reply) => {
         const query = request.query;
         const metric = query.metric;
-        const { startTime, endTime, legacyPeriod } = resolveMetricsTime(query, db, query.router_key_id);
+        const { startTime, endTime, legacyPeriod } = resolveMetricsTime(query, db, query.router_key_id, query.provider_id);
         const timeseries = getMetricsTimeseries(db, legacyPeriod, metric, query.provider_id, query.backend_model, query.router_key_id, startTime, endTime);
         return reply.send(timeseries);
     });

package/dist/admin/stats.js

@@ -10,6 +10,8 @@ const StatsQuerySchema = Type.Object({
     start_time: Type.Optional(Type.String()),
     end_time: Type.Optional(Type.String()),
     router_key_id: Type.Optional(Type.String()),
+    provider_id: Type.Optional(Type.String()),
+    backend_model: Type.Optional(Type.String()),
 });
 export const adminStatsRoutes = (app, options, done) => {
     app.get("/admin/api/stats", { schema: { querystring: StatsQuerySchema } }, async (request, reply) => {
@@ -21,12 +23,12 @@ export const adminStatsRoutes = (app, options, done) => {
             endTime = query.end_time;
         }
         else {
-            const range = resolveTimeRange((query.period ?? "weekly"), options.db, query.router_key_id);
+            const range = resolveTimeRange((query.period ?? "weekly"), options.db, query.router_key_id, query.provider_id);
             startTime = range.startTime;
             endTime = range.endTime;
         }
-        const stats = getStats(options.db, startTime, endTime, query.router_key_id);
-        return reply.send(stats);
+        const stats = getStats(options.db, startTime, endTime, query.router_key_id, query.provider_id, query.backend_model);
+        return reply.send({ ...stats, startTime, endTime });
     });
     done();
 };

package/dist/admin/usage.js

@@ -1,16 +1,26 @@
 import { Type } from "@sinclair/typebox";
 import { getWindowsInRange, getWindowUsage } from "../db/usage-windows.js";
+import { getProviderById } from "../db/providers.js";
 import { resolveTimeRange } from "../utils/time-range.js";
 const UsageQuerySchema = Type.Object({
     router_key_id: Type.Optional(Type.String()),
+    provider_id: Type.Optional(Type.String()),
 });
-function getDailyUsage(db, startTime, endTime, routerKeyId) {
-    const routerKeyFilter = routerKeyId
-        ? " AND rl.router_key_id = ?"
-        : "";
-    const params = routerKeyId
-        ? [startTime, endTime, routerKeyId]
-        : [startTime, endTime];
+function getDailyUsage(db, startTime, endTime, routerKeyId, providerId) {
+    const conditions = [
+        "rm.is_complete = 1",
+        "rm.created_at >= datetime(?)",
+        "rm.created_at < datetime(?)",
+    ];
+    const params = [startTime, endTime];
+    if (routerKeyId) {
+        conditions.push("rm.router_key_id = ?");
+        params.push(routerKeyId);
+    }
+    if (providerId) {
+        conditions.push("rm.provider_id = ?");
+        params.push(providerId);
+    }
     return db.prepare(`
     SELECT
       date(rm.created_at) AS date,
@@ -18,37 +28,48 @@ function getDailyUsage(db, startTime, endTime, routerKeyId) {
       COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens,
       COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens
     FROM request_metrics rm
-    JOIN request_logs rl ON rl.id = rm.request_log_id
-    WHERE rm.is_complete = 1
-      AND rm.created_at >= datetime(?)
-      AND rm.created_at < datetime(?)
-      ${routerKeyFilter}
+    WHERE ${conditions.join(" AND ")}
     GROUP BY date(rm.created_at)
     ORDER BY date ASC
   `).all(...params);
 }
+function resolveProviderName(db, providerId) {
+    if (!providerId)
+        return null;
+    return getProviderById(db, providerId)?.name ?? null;
+}
 export const adminUsageRoutes = (app, options, done) => {
     const { db } = options;
     app.get("/admin/api/usage/windows", { schema: { querystring: UsageQuerySchema } }, async (request) => {
         const query = request.query;
-        const range = resolveTimeRange("window", db, query.router_key_id);
-        const windows = getWindowsInRange(db, range.startTime, range.endTime, query.router_key_id);
-        if (windows.length === 0)
+        if (query.provider_id) {
+            const range = resolveTimeRange("window", db, query.router_key_id, query.provider_id);
+            const windows = getWindowsInRange(db, range.startTime, range.endTime, query.router_key_id, query.provider_id);
+            if (windows.length === 0)
+                return [];
+            return windows.map(w => ({
+                window: { ...w, provider_name: resolveProviderName(db, w.provider_id) },
+                usage: getWindowUsage(db, w.start_time, w.end_time, query.router_key_id, query.provider_id),
+            }));
+        }
+        const allWindows = getWindowsInRange(db, "1970-01-01", "2099-12-31", query.router_key_id)
+            .filter((w) => w.provider_id !== null);
+        if (allWindows.length === 0)
             return [];
-        return windows.map(w => ({
-            window: w,
+        return allWindows.map(w => ({
+            window: { ...w, provider_name: resolveProviderName(db, w.provider_id) },
             usage: getWindowUsage(db, w.start_time, w.end_time, query.router_key_id),
         }));
     });
     app.get("/admin/api/usage/weekly", { schema: { querystring: UsageQuerySchema } }, async (request) => {
         const query = request.query;
         const range = resolveTimeRange("weekly", db, query.router_key_id);
-        return getDailyUsage(db, range.startTime, range.endTime, query.router_key_id);
+        return getDailyUsage(db, range.startTime, range.endTime, query.router_key_id, query.provider_id);
     });
     app.get("/admin/api/usage/monthly", { schema: { querystring: UsageQuerySchema } }, async (request) => {
         const query = request.query;
         const range = resolveTimeRange("monthly", db, query.router_key_id);
-        return getDailyUsage(db, range.startTime, range.endTime, query.router_key_id);
+        return getDailyUsage(db, range.startTime, range.endTime, query.router_key_id, query.provider_id);
     });
     done();
 };

package/dist/db/metrics.d.ts

@@ -23,6 +23,8 @@ export type MetricsInsert = {
     provider_id: string;
     backend_model: string;
     api_type: string;
+    router_key_id?: string | null;
+    status_code?: number | null;
     input_tokens?: number | null;
     output_tokens?: number | null;
     cache_creation_tokens?: number | null;

package/dist/db/metrics.js

@@ -2,8 +2,8 @@ import { randomUUID } from "crypto";
 import { MS_PER_SECOND } from "../constants.js";
 export function insertMetrics(db, m) {
     const id = randomUUID();
-    db.prepare(`INSERT INTO request_metrics (id, request_log_id, provider_id, backend_model, api_type, input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens, ttft_ms, total_duration_ms, tokens_per_second, stop_reason, is_complete)
-     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(id, m.request_log_id, m.provider_id, m.backend_model, m.api_type, m.input_tokens ?? null, m.output_tokens ?? null, m.cache_creation_tokens ?? null, m.cache_read_tokens ?? null, m.ttft_ms ?? null, m.total_duration_ms ?? null, m.tokens_per_second ?? null, m.stop_reason ?? null, m.is_complete ?? 1);
+    db.prepare(`INSERT INTO request_metrics (id, request_log_id, provider_id, backend_model, api_type, router_key_id, status_code, input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens, ttft_ms, total_duration_ms, tokens_per_second, stop_reason, is_complete)
+     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(id, m.request_log_id, m.provider_id, m.backend_model, m.api_type, m.router_key_id ?? null, m.status_code ?? null, m.input_tokens ?? null, m.output_tokens ?? null, m.cache_creation_tokens ?? null, m.cache_read_tokens ?? null, m.ttft_ms ?? null, m.total_duration_ms ?? null, m.tokens_per_second ?? null, m.stop_reason ?? null, m.is_complete ?? 1);
     return id;
 }
 const PERIOD_OFFSET = {
@@ -64,8 +64,7 @@ export function getMetricsSummary(db, period, providerId, backendModel, routerKe
         params.push(backendModel);
     }
     if (routerKeyId) {
-        joins.push("LEFT JOIN request_logs rl ON rl.id = rm.request_log_id");
-        conditions.push("rl.router_key_id = ?");
+        conditions.push("rm.router_key_id = ?");
         params.push(routerKeyId);
     }
     return db.prepare(`
@@ -108,19 +107,17 @@ export function getMetricsTimeseries(db, period, metric, providerId, backendMode
         params.push(backendModel);
     }
     if (routerKeyId) {
-        conditions.push("rl.router_key_id = ?");
+        conditions.push("rm.router_key_id = ?");
         params.push(routerKeyId);
     }
     const where = conditions.join(" AND ");
     const expr = METRIC_EXPR[metric];
-    const joinClause = routerKeyId ? "LEFT JOIN request_logs rl ON rl.id = rm.request_log_id" : "";
     const rows = db.prepare(`
     SELECT
       (unixepoch(rm.created_at) / ?) * ? AS bucket_key,
       ${expr} AS avg_value,
       COUNT(*) AS count
     FROM request_metrics rm
-    ${joinClause}
     WHERE ${where}
     GROUP BY bucket_key
     ORDER BY bucket_key ASC

package/dist/db/migrations/026_metrics_independent.sql

@@ -0,0 +1,54 @@
+-- Metrics 独立化：request_metrics 增加路由维度列，解除级联删除依赖
+-- usage_windows 增加 provider_id 支持按 provider 维度追踪使用量
+
+-- 1. 重建 request_metrics：CASCADE -> SET NULL，同时新增 router_key_id / status_code
+CREATE TABLE request_metrics_new (
+  id TEXT PRIMARY KEY,
+  request_log_id TEXT UNIQUE REFERENCES request_logs(id) ON DELETE SET NULL,
+  provider_id TEXT NOT NULL,
+  backend_model TEXT NOT NULL,
+  api_type TEXT NOT NULL,
+  input_tokens INTEGER,
+  output_tokens INTEGER,
+  cache_creation_tokens INTEGER,
+  cache_read_tokens INTEGER,
+  ttft_ms INTEGER,
+  total_duration_ms INTEGER,
+  tokens_per_second REAL,
+  stop_reason TEXT,
+  is_complete INTEGER NOT NULL DEFAULT 1,
+  router_key_id TEXT,
+  status_code INTEGER,
+  created_at TEXT NOT NULL DEFAULT (datetime('now'))
+);
+
+INSERT INTO request_metrics_new
+  (id, request_log_id, provider_id, backend_model, api_type,
+   input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens,
+   ttft_ms, total_duration_ms, tokens_per_second, stop_reason,
+   is_complete, created_at)
+SELECT
+  id, request_log_id, provider_id, backend_model, api_type,
+  input_tokens, output_tokens, cache_creation_tokens, cache_read_tokens,
+  ttft_ms, total_duration_ms, tokens_per_second, stop_reason,
+  is_complete, created_at
+FROM request_metrics;
+
+-- 回填 router_key_id 和 status_code 从 request_logs
+UPDATE request_metrics_new
+SET
+  router_key_id = rl.router_key_id,
+  status_code = rl.status_code
+FROM request_logs rl
+WHERE rl.id = request_metrics_new.request_log_id;
+
+DROP TABLE request_metrics;
+ALTER TABLE request_metrics_new RENAME TO request_metrics;
+
+-- 重建原有索引
+CREATE INDEX idx_metrics_time_provider_model ON request_metrics(created_at, provider_id, backend_model);
+CREATE INDEX idx_metrics_api_type_created_at ON request_metrics(api_type, created_at);
+
+-- 2. usage_windows 增加 provider_id 列
+ALTER TABLE usage_windows ADD COLUMN provider_id TEXT;
+CREATE INDEX IF NOT EXISTS idx_usage_windows_provider_id ON usage_windows(provider_id);

package/dist/db/stats.d.ts

@@ -3,6 +3,7 @@ export interface Stats {
     totalRequests: number;
     successRate: number;
     avgTps: number;
-    totalTokens: number;
+    totalInputTokens: number;
+    totalOutputTokens: number;
 }
-export declare function getStats(db: Database.Database, startTime: string, endTime: string, routerKeyId?: string): Stats;
+export declare function getStats(db: Database.Database, startTime: string, endTime: string, routerKeyId?: string, providerId?: string, backendModel?: string): Stats;

package/dist/db/stats.js

@@ -1,4 +1,4 @@
-export function getStats(db, startTime, endTime, routerKeyId) {
+export function getStats(db, startTime, endTime, routerKeyId, providerId, backendModel) {
     const conditions = [
         "rm.is_complete = 1",
         "rm.created_at >= datetime(?)",
@@ -6,18 +6,26 @@ export function getStats(db, startTime, endTime, routerKeyId) {
     ];
     const params = [startTime, endTime];
     if (routerKeyId) {
-        conditions.push("rl.router_key_id = ?");
+        conditions.push("rm.router_key_id = ?");
         params.push(routerKeyId);
     }
+    if (providerId) {
+        conditions.push("rm.provider_id = ?");
+        params.push(providerId);
+    }
+    if (backendModel) {
+        conditions.push("rm.backend_model = ?");
+        params.push(backendModel);
+    }
     const where = conditions.join(" AND ");
     const row = db.prepare(`
     SELECT
       COUNT(*) AS total_requests,
-      SUM(CASE WHEN rl.status_code >= 200 AND rl.status_code < 300 THEN 1 ELSE 0 END) AS success_count,
+      SUM(CASE WHEN rm.status_code >= 200 AND rm.status_code < 300 THEN 1 ELSE 0 END) AS success_count,
       AVG(rm.tokens_per_second) AS avg_tps,
-      COALESCE(SUM(rm.input_tokens), 0) + COALESCE(SUM(rm.output_tokens), 0) AS total_tokens
+      COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens,
+      COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens
     FROM request_metrics rm
-    JOIN request_logs rl ON rl.id = rm.request_log_id
     WHERE ${where}
   `).get(...params);
     const total = row?.total_requests ?? 0;
@@ -25,6 +33,7 @@ export function getStats(db, startTime, endTime, routerKeyId) {
         totalRequests: total,
         successRate: total > 0 ? (row?.success_count ?? 0) / total : 0,
         avgTps: row?.avg_tps ?? 0,
-        totalTokens: row?.total_tokens ?? 0,
+        totalInputTokens: row?.total_input_tokens ?? 0,
+        totalOutputTokens: row?.total_output_tokens ?? 0,
     };
 }

package/dist/db/usage-windows.d.ts

@@ -2,6 +2,7 @@ import Database from "better-sqlite3";
 export interface UsageWindow {
     id: string;
     router_key_id: string | null;
+    provider_id: string | null;
     start_time: string;
     end_time: string;
     created_at: string;
@@ -12,8 +13,8 @@ export interface WindowUsage {
     total_output_tokens: number;
 }
 export declare function insertWindow(db: Database.Database, w: Omit<UsageWindow, "created_at">): string;
-export declare function getLatestWindow(db: Database.Database, routerKeyId?: string): UsageWindow | null;
-/** 返回与 [start, end) 区间有重叠的窗口 */
-export declare function getWindowsInRange(db: Database.Database, start: string, end: string, routerKeyId?: string): UsageWindow[];
+export declare function getLatestWindow(db: Database.Database, routerKeyId?: string, providerId?: string): UsageWindow | null;
+/** 返回与 [start, end) 区间有重叠的窗口。可选参数不传表示不过滤该维度（与 getLatestWindow 的 IS NULL 语义不同） */
+export declare function getWindowsInRange(db: Database.Database, start: string, end: string, routerKeyId?: string, providerId?: string): UsageWindow[];
 /** 聚合指定时间窗口内的请求计数和 token 用量 */
-export declare function getWindowUsage(db: Database.Database, startTime: string, endTime: string, routerKeyId?: string): WindowUsage;
+export declare function getWindowUsage(db: Database.Database, startTime: string, endTime: string, routerKeyId?: string, providerId?: string): WindowUsage;

package/dist/db/usage-windows.js

@@ -1,37 +1,65 @@
 import { randomUUID } from "crypto";
 export function insertWindow(db, w) {
     const id = w.id || randomUUID();
-    db.prepare("INSERT INTO usage_windows (id, router_key_id, start_time, end_time) VALUES (?, ?, ?, ?)").run(id, w.router_key_id ?? null, w.start_time, w.end_time);
+    db.prepare("INSERT INTO usage_windows (id, router_key_id, provider_id, start_time, end_time) VALUES (?, ?, ?, ?, ?)").run(id, w.router_key_id ?? null, w.provider_id ?? null, w.start_time, w.end_time);
     return id;
 }
-export function getLatestWindow(db, routerKeyId) {
-    const sql = routerKeyId
-        ? "SELECT * FROM usage_windows WHERE router_key_id = ? ORDER BY start_time DESC LIMIT 1"
-        : "SELECT * FROM usage_windows ORDER BY start_time DESC LIMIT 1";
-    const params = routerKeyId ? [routerKeyId] : [];
+export function getLatestWindow(db, routerKeyId, providerId) {
+    const conditions = [];
+    const params = [];
+    if (routerKeyId) {
+        conditions.push("router_key_id = ?");
+        params.push(routerKeyId);
+    }
+    else {
+        conditions.push("router_key_id IS NULL");
+    }
+    if (providerId) {
+        conditions.push("provider_id = ?");
+        params.push(providerId);
+    }
+    else {
+        conditions.push("provider_id IS NULL");
+    }
+    const sql = `SELECT * FROM usage_windows WHERE ${conditions.join(" AND ")} ORDER BY start_time DESC LIMIT 1`;
     return db.prepare(sql).get(...params) ?? null;
 }
-/** 返回与 [start, end) 区间有重叠的窗口 */
-export function getWindowsInRange(db, start, end, routerKeyId) {
+/** 返回与 [start, end) 区间有重叠的窗口。可选参数不传表示不过滤该维度（与 getLatestWindow 的 IS NULL 语义不同） */
+export function getWindowsInRange(db, start, end, routerKeyId, providerId) {
+    const conditions = ["start_time < ?", "end_time > ?"];
+    const params = [end, start];
     if (routerKeyId) {
-        return db.prepare("SELECT * FROM usage_windows WHERE start_time < ? AND end_time > ? AND router_key_id = ? ORDER BY start_time ASC").all(end, start, routerKeyId);
+        conditions.push("router_key_id = ?");
+        params.push(routerKeyId);
+    }
+    if (providerId) {
+        conditions.push("provider_id = ?");
+        params.push(providerId);
     }
-    return db.prepare("SELECT * FROM usage_windows WHERE start_time < ? AND end_time > ? ORDER BY start_time ASC").all(end, start);
+    return db.prepare(`SELECT * FROM usage_windows WHERE ${conditions.join(" AND ")} ORDER BY start_time ASC`).all(...params);
 }
 /** 聚合指定时间窗口内的请求计数和 token 用量 */
-export function getWindowUsage(db, startTime, endTime, routerKeyId) {
-    const baseSql = `
+export function getWindowUsage(db, startTime, endTime, routerKeyId, providerId) {
+    const conditions = [
+        "rm.is_complete = 1",
+        "rm.created_at >= datetime(?)",
+        "rm.created_at < datetime(?)",
+    ];
+    const params = [startTime, endTime];
+    if (routerKeyId) {
+        conditions.push("rm.router_key_id = ?");
+        params.push(routerKeyId);
+    }
+    if (providerId) {
+        conditions.push("rm.provider_id = ?");
+        params.push(providerId);
+    }
+    return db.prepare(`
     SELECT
       COUNT(*) AS request_count,
       COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens,
       COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens
     FROM request_metrics rm
-    JOIN request_logs rl ON rl.id = rm.request_log_id
-    WHERE rm.is_complete = 1
-      AND rm.created_at >= datetime(?)
-      AND rm.created_at < datetime(?)`;
-    if (routerKeyId) {
-        return db.prepare(`${baseSql} AND rl.router_key_id = ?`).get(startTime, endTime, routerKeyId);
-    }
-    return db.prepare(baseSql).get(startTime, endTime);
+    WHERE ${conditions.join(" AND ")}
+  `).get(...params);
 }

package/dist/middleware/auth.js

@@ -53,15 +53,25 @@ const authMiddlewareRaw = (app, options, done) => {
             reply.code(HTTP_SERVICE_UNAVAILABLE).send({ error: { message: "Service not initialized" } });
             return reply;
         }
+        let token;
         const authHeader = request.headers.authorization;
-        if (!authHeader || !authHeader.startsWith("Bearer ")) {
+        if (authHeader && authHeader.startsWith("Bearer ")) {
+            token = authHeader.slice(BEARER_PREFIX_LENGTH);
+        }
+        else {
+            // Fallback: Anthropic SDK sends API key via x-api-key header
+            const apiKeyHeader = request.headers["x-api-key"];
+            if (apiKeyHeader) {
+                token = apiKeyHeader;
+            }
+        }
+        if (!token) {
             if (proxyApiType) {
                 logRejectedAuth(options.db, proxyApiType, HTTP_UNAUTHORIZED, "Invalid API key", request);
             }
             unauthorizedReply(reply);
             return reply;
         }
-        const token = authHeader.slice(BEARER_PREFIX_LENGTH);
         const hash = createHash("sha256").update(token).digest("hex");
         const row = stmt.get(hash);
         if (!row) {

package/dist/monitor/request-tracker.d.ts

@@ -2,7 +2,7 @@ import type { ServerResponse } from "node:http";
 import { StatsAggregator } from "./stats-aggregator.js";
 import { RuntimeCollector } from "./runtime-collector.js";
 import type { ProviderSemaphoreManager } from "../proxy/semaphore.js";
-import type { ActiveRequest, ProviderConcurrencySnapshot, RuntimeMetrics, StatsSnapshot } from "./types.js";
+import type { ActiveRequest, AttemptSnapshot, ProviderConcurrencySnapshot, RuntimeMetrics, StatsSnapshot } from "./types.js";
 export interface TrackerLogger {
     debug(obj: Record<string, unknown>, msg: string): void;
     warn(obj: Record<string, unknown>, msg: string): void;
@@ -31,6 +31,7 @@ export declare class RequestTracker {
     complete(id: string, result: {
         status: "completed" | "failed";
         statusCode?: number;
+        attempts?: AttemptSnapshot[];
     }): void;
     getActive(): ActiveRequest[];
     getRecent(limit?: number): ActiveRequest[];

package/dist/monitor/request-tracker.js

@@ -73,6 +73,7 @@ export class RequestTracker {
             ...req,
             status: result.status,
             completedAt: now,
+            attempts: result.attempts ?? req.attempts,
         };
         this.activeMap.delete(id);
         this.streamAccumulators.delete(id);
@@ -161,6 +162,7 @@ export class RequestTracker {
         if (this.pushTimer)
             return;
         this.tickCount = 0;
+        this.runtimeCollector.start();
         this.pushTimer = setInterval(() => {
             this.tickCount++;
             this.cleanupRecent();
@@ -179,6 +181,7 @@ export class RequestTracker {
             clearInterval(this.pushTimer);
             this.pushTimer = null;
         }
+        this.runtimeCollector.stop();
     }
     broadcast(event, data) {
         // Strip clientRequest from broadcasts to reduce bandwidth;

package/dist/monitor/stats-aggregator.js

@@ -34,13 +34,13 @@ function emptyAccumulator() {
         successCount: 0,
         errorCount: 0,
         retryCount: 0,
-        latencySum: 0,
-        latencyCount: 0,
+        latencyBuffer: new RingBuffer(PROVIDER_LATENCY_CAPACITY),
         errorsByCode: new Map(),
     };
 }
 const TOP_ERRORS_LIMIT = 5;
 const DEFAULT_CAPACITY = 1000;
+const PROVIDER_LATENCY_CAPACITY = 200;
 const HTTP_SUCCESS_RANGE_MIN = 200;
 const HTTP_SUCCESS_RANGE_MAX = 400;
 const PERCENTILE_P50 = 0.5;
@@ -103,8 +103,7 @@ export class StatsAggregator {
             acc = emptyAccumulator();
             this.providers.set(providerId, acc);
         }
-        acc.latencySum += ms;
-        acc.latencyCount++;
+        acc.latencyBuffer.push(ms);
     }
     getStats() {
         const sorted = this.latencyBuffer.sorted();
@@ -123,9 +122,7 @@ export class StatsAggregator {
                 totalRequests: acc.totalRequests,
                 successCount: acc.successCount,
                 errorCount: acc.errorCount,
-                avgLatencyMs: acc.latencyCount > 0
-                    ? acc.latencySum / acc.latencyCount
-                    : 0,
+                avgLatencyMs: avgFromBuffer(acc.latencyBuffer),
                 retryCount: acc.retryCount,
                 topErrors,
             };
@@ -164,3 +161,7 @@ function percentile(sorted, p) {
     const idx = Math.ceil(p * sorted.length) - 1;
     return sorted[Math.max(0, Math.min(idx, sorted.length - 1))];
 }
+function avgFromBuffer(buf) {
+    const sorted = buf.sorted();
+    return sorted.length > 0 ? sorted.reduce((s, v) => s + v, 0) / sorted.length : 0;
+}

package/dist/proxy/orchestrator.js

@@ -30,7 +30,12 @@ export class ProxyOrchestrator {
                 this.deps.trackerScope.markQueued(trackerReq.id, false);
             }
             return this.executeResilience(config, ctx);
-        }), (result) => this.extractTrackStatus(result));
+        }), (result) => this.extractTrackStatus(result), (result) => result.attempts.map(a => ({
+            statusCode: a.statusCode,
+            error: a.error,
+            latencyMs: a.latencyMs,
+            providerId: a.target.provider_id,
+        })));
         this.sendResponse(reply, result.result, ctx);
         return result;
     }

package/dist/proxy/proxy-handler.js

@@ -16,6 +16,15 @@ import { applyProviderPatches } from "./patch/index.js";
 const HTTP_ERROR_THRESHOLD = 400;
 const MAX_LOG_FIELD_LENGTH = 80;
 const UPSTREAM_ERROR_STATUS = 502;
+/** 从 TransportResult 中提取最终 HTTP status code */
+function getTransportStatusCode(result) {
+    if (result.kind === "success" || result.kind === "error" || result.kind === "stream_error")
+        return result.statusCode;
+    if (result.kind === "stream_success" || result.kind === "stream_abort")
+        return result.statusCode;
+    // kind === "throw"：无 HTTP 状态码
+    return null;
+}
 function rejectAndReply(reply, params, error, errorMessage, providerId) {
     insertRejectedLog({
         db: params.db, logId: params.logId, apiType: params.apiType, model: params.model,
@@ -150,11 +159,11 @@ async function executeFailoverLoop(ctx) {
                 clientReq, upstreamReqBase, logId, routerKeyId, originalModel, sessionId,
                 failover: { isFailoverIteration, rootLogId: rootLogId },
             }, resilienceResult.attempts, resilienceResult.result, startTime);
-            collectTransportMetrics(deps.db, apiType, resilienceResult.result, isStream, lastLogId, provider.id, resolved.backend_model, request);
+            collectTransportMetrics(deps.db, apiType, resilienceResult.result, isStream, lastLogId, provider.id, resolved.backend_model, request, routerKeyId, getTransportStatusCode(resilienceResult.result));
             const tr = resilienceResult.result;
             const succeeded = tr.kind === "success" || tr.kind === "stream_success" || tr.kind === "stream_abort";
             if (succeeded)
-                deps.usageWindowTracker?.recordRequest(routerKeyId ?? undefined);
+                deps.usageWindowTracker?.recordRequest(provider.id, routerKeyId ?? undefined);
             if (isStream && deps.tracker) {
                 const sc = deps.tracker.get(logId)?.streamContent;
                 const blocks = sc?.blocks;

package/dist/proxy/proxy-logging.d.ts

@@ -23,4 +23,4 @@ export declare function logResilienceResult(db: Database.Database, params: {
     sessionId?: string | null;
     failover?: FailoverContext;
 }, attempts: ResilienceAttempt[], result: TransportResult, startTime: number): string;
-export declare function collectTransportMetrics(db: Database.Database, apiType: "openai" | "anthropic", result: TransportResult, isStream: boolean, lastSuccessLogId: string, providerId: string, backendModel: string, request: FastifyRequest): void;
+export declare function collectTransportMetrics(db: Database.Database, apiType: "openai" | "anthropic", result: TransportResult, isStream: boolean, lastSuccessLogId: string, providerId: string, backendModel: string, request: FastifyRequest, routerKeyId?: string | null, statusCode?: number | null): void;

package/dist/proxy/proxy-logging.js

@@ -116,8 +116,11 @@ export function logResilienceResult(db, params, attempts, result, startTime) {
     }
     return lastSuccessLogId;
 }
-export function collectTransportMetrics(db, apiType, result, isStream, lastSuccessLogId, providerId, backendModel, request) {
-    const base = { request_log_id: lastSuccessLogId, provider_id: providerId, backend_model: backendModel, api_type: apiType };
+export function collectTransportMetrics(db, apiType, result, isStream, lastSuccessLogId, providerId, backendModel, request, routerKeyId, statusCode) {
+    const base = {
+        request_log_id: lastSuccessLogId, provider_id: providerId, backend_model: backendModel, api_type: apiType,
+        router_key_id: routerKeyId ?? null, status_code: statusCode ?? null,
+    };
     try {
         if (isStream && (result.kind === "stream_success" || result.kind === "stream_abort")) {
             if (result.metrics) {

package/dist/proxy/resilience.d.ts

@@ -25,8 +25,6 @@ export interface ResilienceConfig {
     failoverThreshold: number;
     ruleMatcher?: RetryRuleMatcher;
     isFailover: boolean;
-    /** DB 规则 max_retries 的全局安全阀，防止单规则配置导致过多重试 */
-    globalRetryCap?: number;
     /** 全局迭代上限，防止极端配置导致 while(true) 循环过多 */
     iterationCap?: number;
 }

package/dist/proxy/resilience.js

@@ -29,7 +29,6 @@ export function createStrategy(rule) {
 const RETRYABLE_THROW_CODES = new Set(["ETIMEDOUT", "ECONNRESET", "ECONNREFUSED"]);
 const HTTP_TOO_MANY_REQUESTS = 429;
 const DEFAULT_THROW_MAX_RETRIES = 3;
-const DEFAULT_RETRY_CAP = 3;
 const DEFAULT_ITERATION_CAP = 50;
 // ---------- Internal helpers ----------
 function isRetryableThrow(err) {
@@ -94,7 +93,7 @@ export class ResilienceLayer {
             const matchedRule = body && config.ruleMatcher
                 ? config.ruleMatcher.match(result.statusCode, body)
                 : null;
-            if (matchedRule && state.attemptCount < Math.min(matchedRule.max_retries, config.globalRetryCap ?? DEFAULT_RETRY_CAP)) {
+            if (matchedRule && state.attemptCount < matchedRule.max_retries) {
                 const strategy = createStrategy(matchedRule);
                 const headers = extractHeaders(result);
                 const retryAfterMs = result.statusCode === HTTP_TOO_MANY_REQUESTS
@@ -110,7 +109,7 @@ export class ResilienceLayer {
         const body = extractBody(result);
         if (body && config.ruleMatcher) {
             const matchedRule = config.ruleMatcher.match(result.statusCode, body);
-            if (matchedRule && state.attemptCount < Math.min(matchedRule.max_retries, config.globalRetryCap ?? DEFAULT_RETRY_CAP)) {
+            if (matchedRule && state.attemptCount < matchedRule.max_retries) {
                 const strategy = createStrategy(matchedRule);
                 return { action: "retry", delayMs: strategy.getDelay(state.attemptCount) };
             }