llm-simple-router 0.4.0 → 0.4.2

package/dist/db/log-cleaner.js

@@ -0,0 +1,42 @@
+import { deleteLogsBefore } from "./logs.js";
+import { getLogRetentionDays } from "./settings.js";
+const MS_PER_DAY = 86_400_000;
+const CLEANUP_INTERVAL_MS = 3_600_000; // 1 小时
+/** 运行一次清理，返回删除条数 */
+export function runLogCleanup(db) {
+    const days = getLogRetentionDays(db);
+    if (days <= 0)
+        return 0;
+    const cutoff = new Date(Date.now() - days * MS_PER_DAY).toISOString();
+    return deleteLogsBefore(db, cutoff);
+}
+/** 启动定时清理，返回 handle 用于停止 */
+export function scheduleLogCleanup(db, log) {
+    let cleaning = false;
+    let timer = null;
+    const doCleanup = () => {
+        if (cleaning)
+            return;
+        cleaning = true;
+        try {
+            const deleted = runLogCleanup(db);
+            if (deleted > 0)
+                log.info(`Log cleanup: deleted ${deleted} records`);
+        }
+        finally {
+            cleaning = false;
+        }
+    };
+    // 启动时立即执行一次
+    doCleanup();
+    // 定时执行
+    timer = setInterval(doCleanup, CLEANUP_INTERVAL_MS);
+    return {
+        stop: () => {
+            if (timer) {
+                clearInterval(timer);
+                timer = null;
+            }
+        },
+    };
+}

package/dist/db/logs.d.ts

@@ -9,21 +9,27 @@ export interface RequestLog {
     is_stream: number;
     error_message: string | null;
     created_at: string;
-    request_body: string | null;
-    response_body: string | null;
     client_request: string | null;
     upstream_request: string | null;
     upstream_response: string | null;
-    client_response: string | null;
     is_retry: number;
     is_failover: number;
     original_request_id: string | null;
     original_model: string | null;
+    input_tokens: number | null;
+    output_tokens: number | null;
+    cache_read_tokens: number | null;
+    ttft_ms: number | null;
+    tokens_per_second: number | null;
+    stop_reason: string | null;
+    backend_model: string | null;
+    metrics_complete: number;
+    stream_text_content: string | null;
 }
-/** 列表查询扩展字段：JOIN request_metrics + providers 获得 */
+/** 列表查询扩展字段：JOIN providers 获得 provider_name */
 export interface RequestLogListRow extends RequestLog {
-    backend_model: string | null;
     provider_name: string | null;
+    child_count?: number;
 }
 export interface RequestLogInsert {
     id: string;
@@ -35,12 +41,9 @@ export interface RequestLogInsert {
     is_stream: number;
     error_message: string | null;
     created_at: string;
-    request_body?: string | null;
-    response_body?: string | null;
     client_request?: string | null;
     upstream_request?: string | null;
     upstream_response?: string | null;
-    client_response?: string | null;
     is_retry?: number;
     is_failover?: number;
     original_request_id?: string | null;
@@ -54,11 +57,29 @@ export declare function getRequestLogs(db: Database.Database, options: {
     api_type?: string;
     model?: string;
     router_key_id?: string;
+    provider_id?: string;
+    start_time?: string;
+    end_time?: string;
 }): {
     data: RequestLogListRow[];
     total: number;
 };
 export declare function getRequestLogById(db: Database.Database, id: string): RequestLog | undefined;
+type MetricsUpdate = {
+    input_tokens?: number | null;
+    output_tokens?: number | null;
+    cache_read_tokens?: number | null;
+    ttft_ms?: number | null;
+    tokens_per_second?: number | null;
+    stop_reason?: string | null;
+    is_complete?: number;
+};
+/** 双写：collectTransportMetrics 写 request_metrics 的同时，更新 request_logs 的冗余列 */
+export declare function updateLogMetrics(db: Database.Database, logId: string, m: MetricsUpdate): void;
+/** 流式请求完成后，将 tracker 中累积的文本内容写入 request_logs */
+export declare function updateLogStreamContent(db: Database.Database, logId: string, textContent: string): void;
+/** 启动时回填：从 request_metrics 补齐 metrics_complete = 0 但实际有指标的行 */
+export declare function backfillMetricsFromRequestMetrics(db: Database.Database): number;
 export declare function deleteLogsBefore(db: Database.Database, beforeDate: string): number;
 /** 查询某条日志的子请求（retry/failover 关联），上限 100 条 */
 export declare function getRequestLogChildren(db: Database.Database, parentId: string): RequestLogListRow[];
@@ -72,7 +93,11 @@ export declare function getRequestLogsGrouped(db: Database.Database, options: {
     api_type?: string;
     model?: string;
     router_key_id?: string;
+    provider_id?: string;
+    start_time?: string;
+    end_time?: string;
 }): {
     data: RequestLogGroupedRow[];
     total: number;
 };
+export {};

package/dist/db/logs.js

@@ -1,14 +1,17 @@
 // --- request_logs ---
-/** 三处日志列表查询共享的 SELECT 列 + JOIN 子句 */
+/** 日志列表查询共享的 SELECT 列 + JOIN 子句（metrics 已冗余到 request_logs，无需 JOIN request_metrics） */
 const LOG_LIST_SELECT = `rl.id, rl.api_type, rl.model, rl.provider_id, rl.status_code, rl.latency_ms,
             rl.is_stream, rl.error_message, rl.created_at, rl.is_retry, rl.is_failover, rl.original_request_id, rl.original_model,
             CASE WHEN rl.provider_id = 'router' THEN rl.upstream_request ELSE NULL END AS upstream_request,
-            rm.backend_model, COALESCE(p.name, rl.provider_id) AS provider_name`;
-const LOG_LIST_JOIN = `LEFT JOIN request_metrics rm ON rm.request_log_id = rl.id
-     LEFT JOIN providers p ON p.id = rl.provider_id`;
+            rl.input_tokens, rl.output_tokens, rl.cache_read_tokens, rl.ttft_ms, rl.tokens_per_second, rl.stop_reason,
+            rl.backend_model, rl.metrics_complete,
+            COALESCE(p.name, rl.provider_id) AS provider_name`;
+const LOG_LIST_JOIN = `LEFT JOIN providers p ON p.id = rl.provider_id`;
 export function insertRequestLog(db, log) {
-    db.prepare(`INSERT INTO request_logs (id, api_type, model, provider_id, status_code, latency_ms, is_stream, error_message, created_at, request_body, response_body, client_request, upstream_request, upstream_response, client_response, is_retry, is_failover, original_request_id, router_key_id, original_model)
-     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(log.id, log.api_type, log.model, log.provider_id, log.status_code, log.latency_ms, log.is_stream, log.error_message, log.created_at, log.request_body ?? null, log.response_body ?? null, log.client_request ?? null, log.upstream_request ?? null, log.upstream_response ?? null, log.client_response ?? null, log.is_retry ?? 0, log.is_failover ?? 0, log.original_request_id ?? null, log.router_key_id ?? null, log.original_model ?? null);
+    db.prepare(`INSERT INTO request_logs (id, api_type, model, provider_id, status_code, latency_ms,
+      is_stream, error_message, created_at, client_request, upstream_request, upstream_response,
+      is_retry, is_failover, original_request_id, router_key_id, original_model)
+     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(log.id, log.api_type, log.model, log.provider_id, log.status_code, log.latency_ms, log.is_stream, log.error_message, log.created_at, log.client_request ?? null, log.upstream_request ?? null, log.upstream_response ?? null, log.is_retry ?? 0, log.is_failover ?? 0, log.original_request_id ?? null, log.router_key_id ?? null, log.original_model ?? null);
 }
 function buildLogWhereClause(options, baseCondition) {
     let where = baseCondition;
@@ -25,6 +28,18 @@ function buildLogWhereClause(options, baseCondition) {
         where += " AND rl.router_key_id = ?";
         params.push(options.router_key_id);
     }
+    if (options.provider_id) {
+        where += " AND rl.provider_id = ?";
+        params.push(options.provider_id);
+    }
+    if (options.start_time) {
+        where += " AND rl.created_at >= ?";
+        params.push(options.start_time);
+    }
+    if (options.end_time) {
+        where += " AND rl.created_at <= ?";
+        params.push(options.end_time);
+    }
     return { where, params };
 }
 export function getRequestLogs(db, options) {
@@ -42,6 +57,37 @@ export function getRequestLogs(db, options) {
 export function getRequestLogById(db, id) {
     return db.prepare("SELECT * FROM request_logs WHERE id = ?").get(id);
 }
+/** 双写：collectTransportMetrics 写 request_metrics 的同时，更新 request_logs 的冗余列 */
+export function updateLogMetrics(db, logId, m) {
+    db.prepare(`UPDATE request_logs SET
+       input_tokens = ?, output_tokens = ?, cache_read_tokens = ?,
+       ttft_ms = ?, tokens_per_second = ?, stop_reason = ?,
+       backend_model = (SELECT backend_model FROM request_metrics WHERE request_log_id = ?),
+       metrics_complete = ?
+     WHERE id = ?`).run(m.input_tokens ?? null, m.output_tokens ?? null, m.cache_read_tokens ?? null, m.ttft_ms ?? null, m.tokens_per_second ?? null, m.stop_reason ?? null, logId, m.is_complete ?? 1, logId);
+}
+/** 流式请求完成后，将 tracker 中累积的文本内容写入 request_logs */
+export function updateLogStreamContent(db, logId, textContent) {
+    db.prepare("UPDATE request_logs SET stream_text_content = ? WHERE id = ?").run(textContent, logId);
+}
+/** 启动时回填：从 request_metrics 补齐 metrics_complete = 0 但实际有指标的行 */
+export function backfillMetricsFromRequestMetrics(db) {
+    return db.prepare(`
+    UPDATE request_logs
+    SET
+      input_tokens = rm.input_tokens,
+      output_tokens = rm.output_tokens,
+      cache_read_tokens = rm.cache_read_tokens,
+      ttft_ms = rm.ttft_ms,
+      tokens_per_second = rm.tokens_per_second,
+      stop_reason = rm.stop_reason,
+      backend_model = rm.backend_model,
+      metrics_complete = rm.is_complete
+    FROM request_metrics rm
+    WHERE rm.request_log_id = request_logs.id
+      AND request_logs.metrics_complete = 0
+  `).run().changes;
+}
 export function deleteLogsBefore(db, beforeDate) {
     return db.prepare("DELETE FROM request_logs WHERE created_at < ?").run(beforeDate).changes;
 }

package/dist/db/metrics.d.ts

@@ -1,5 +1,5 @@
 import Database from "better-sqlite3";
-export type MetricsPeriod = "1h" | "6h" | "24h" | "7d" | "30d";
+export type MetricsPeriod = "1h" | "5h" | "6h" | "24h" | "7d" | "30d";
 export type MetricsMetric = "ttft" | "tps" | "tokens" | "cache_rate" | "request_count" | "input_tokens" | "output_tokens" | "cache_hit_tokens";
 export interface MetricsRow {
     id: string;
@@ -48,10 +48,10 @@ export interface MetricsSummaryRow {
     total_cache_hit_tokens: number;
     cache_hit_rate: number | null;
 }
-export declare function getMetricsSummary(db: Database.Database, period: MetricsPeriod, providerId?: string, backendModel?: string, routerKeyId?: string): MetricsSummaryRow[];
+export declare function getMetricsSummary(db: Database.Database, period: MetricsPeriod, providerId?: string, backendModel?: string, routerKeyId?: string, startTime?: string, endTime?: string): MetricsSummaryRow[];
 export interface MetricsTimeseriesRow {
     time_bucket: string;
     avg_value: number | null;
     count: number;
 }
-export declare function getMetricsTimeseries(db: Database.Database, period: MetricsPeriod, metric: MetricsMetric, providerId?: string, backendModel?: string, routerKeyId?: string): MetricsTimeseriesRow[];
+export declare function getMetricsTimeseries(db: Database.Database, period: MetricsPeriod, metric: MetricsMetric, providerId?: string, backendModel?: string, routerKeyId?: string, startTime?: string, endTime?: string): MetricsTimeseriesRow[];

package/dist/db/metrics.js

@@ -7,6 +7,7 @@ export function insertMetrics(db, m) {
 }
 const PERIOD_OFFSET = {
     "1h": "-1 hours",
+    "5h": "-5 hours",
     "6h": "-6 hours",
     "24h": "-1 day",
     "7d": "-7 days",
@@ -14,6 +15,7 @@ const PERIOD_OFFSET = {
 };
 const BUCKET_SECONDS = {
     "1h": 60,
+    "5h": 300,
     "6h": 300,
     "24h": 900,
     "7d": 3600,
@@ -21,10 +23,39 @@ const BUCKET_SECONDS = {
 };
 // unix epoch 秒转毫秒的乘数
 const MS_PER_SEC = 1000;
-export function getMetricsSummary(db, period, providerId, backendModel, routerKeyId) {
-    const offset = PERIOD_OFFSET[period];
-    const conditions = ["rm.is_complete = 1", "rm.created_at >= datetime('now', ?)"];
-    const params = [offset];
+// 时间跨度（秒）→ 桶大小（秒）的阶梯映射，与 BUCKET_SECONDS 保持对齐
+const BUCKET_THRESHOLDS = [
+    { maxSec: 3600, bucketSec: 60 }, // ≤1h: 1min
+    { maxSec: 21600, bucketSec: 300 }, // ≤6h: 5min
+    { maxSec: 86400, bucketSec: 900 }, // ≤1d: 15min
+    { maxSec: 604800, bucketSec: 3600 }, // ≤7d: 1h
+];
+const FALLBACK_BUCKET_SEC = 14400; // >7d: 4h
+function calculateBucketSeconds(startTime, endTime) {
+    const ms = new Date(endTime).getTime() - new Date(startTime).getTime();
+    const sec = ms / MS_PER_SEC;
+    const match = BUCKET_THRESHOLDS.find((t) => sec <= t.maxSec);
+    return match ? match.bucketSec : FALLBACK_BUCKET_SEC;
+}
+function buildTimeCondition(period, startTime, endTime) {
+    if (startTime && endTime) {
+        // request_metrics.created_at 用 datetime('now') 格式 (YYYY-MM-DD HH:MM:SS)，
+        // 前端传入 ISO 8601，需要转换格式以匹配字符串比较
+        return {
+            timeWhere: "rm.created_at >= datetime(?) AND rm.created_at < datetime(?)",
+            timeParams: [startTime, endTime],
+        };
+    }
+    return {
+        timeWhere: "rm.created_at >= datetime('now', ?)",
+        timeParams: [PERIOD_OFFSET[period]],
+    };
+}
+export function getMetricsSummary(db, period, providerId, backendModel, routerKeyId, startTime, endTime) {
+    const { timeWhere, timeParams } = buildTimeCondition(period, startTime, endTime);
+    const conditions = ["rm.is_complete = 1", timeWhere];
+    const params = [...timeParams];
+    const joins = ["LEFT JOIN providers p ON p.id = rm.provider_id"];
     if (providerId) {
         conditions.push("rm.provider_id = ?");
         params.push(providerId);
@@ -33,48 +64,23 @@ export function getMetricsSummary(db, period, providerId, backendModel, routerKe
         conditions.push("rm.backend_model = ?");
         params.push(backendModel);
     }
-    const joins = "LEFT JOIN providers p ON p.id = rm.provider_id";
     if (routerKeyId) {
+        joins.push("LEFT JOIN request_logs rl ON rl.id = rm.request_log_id");
         conditions.push("rl.router_key_id = ?");
         params.push(routerKeyId);
-        return db.prepare(`
-      SELECT
-        rm.provider_id, COALESCE(p.name, rm.provider_id) AS provider_name, rm.backend_model,
-        COUNT(*) AS request_count, AVG(rm.ttft_ms) AS avg_ttft_ms, NULL AS p50_ttft_ms, NULL AS p95_ttft_ms,
-        AVG(rm.tokens_per_second) AS avg_tps,
-        COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens, COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens,
-        COALESCE(SUM(rm.cache_read_tokens), 0) AS total_cache_hit_tokens,
-        CASE WHEN SUM(rm.input_tokens) > 0 THEN SUM(rm.cache_read_tokens) * 1.0 / SUM(rm.input_tokens) ELSE NULL END AS cache_hit_rate
-      FROM request_metrics rm
-      LEFT JOIN providers p ON p.id = rm.provider_id
-      LEFT JOIN request_logs rl ON rl.id = rm.request_log_id
-      WHERE ${conditions.join(" AND ")}
-      GROUP BY rm.provider_id, rm.backend_model ORDER BY request_count DESC
-    `).all(...params);
     }
-    const where = conditions.join(" AND ");
     return db.prepare(`
     SELECT
-      rm.provider_id,
-      COALESCE(p.name, rm.provider_id) AS provider_name,
-      rm.backend_model,
-      COUNT(*) AS request_count,
-      AVG(rm.ttft_ms) AS avg_ttft_ms,
-      NULL AS p50_ttft_ms,
-      NULL AS p95_ttft_ms,
+      rm.provider_id, COALESCE(p.name, rm.provider_id) AS provider_name, rm.backend_model,
+      COUNT(*) AS request_count, AVG(rm.ttft_ms) AS avg_ttft_ms, NULL AS p50_ttft_ms, NULL AS p95_ttft_ms,
       AVG(rm.tokens_per_second) AS avg_tps,
-      COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens,
-      COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens,
+      COALESCE(SUM(rm.input_tokens), 0) AS total_input_tokens, COALESCE(SUM(rm.output_tokens), 0) AS total_output_tokens,
       COALESCE(SUM(rm.cache_read_tokens), 0) AS total_cache_hit_tokens,
-      CASE WHEN SUM(rm.input_tokens) > 0
-        THEN SUM(rm.cache_read_tokens) * 1.0 / SUM(rm.input_tokens)
-        ELSE NULL
-      END AS cache_hit_rate
+      CASE WHEN SUM(rm.input_tokens) > 0 THEN SUM(rm.cache_read_tokens) * 1.0 / SUM(rm.input_tokens) ELSE NULL END AS cache_hit_rate
     FROM request_metrics rm
-    ${joins}
-    WHERE ${where}
-    GROUP BY rm.provider_id, rm.backend_model
-    ORDER BY request_count DESC
+    ${joins.join(" ")}
+    WHERE ${conditions.join(" AND ")}
+    GROUP BY rm.provider_id, rm.backend_model ORDER BY request_count DESC
   `).all(...params);
 }
 const METRIC_EXPR = {
@@ -87,11 +93,13 @@ const METRIC_EXPR = {
     output_tokens: "SUM(rm.output_tokens)",
     cache_hit_tokens: "SUM(rm.cache_read_tokens)",
 };
-export function getMetricsTimeseries(db, period, metric, providerId, backendModel, routerKeyId) {
-    const offset = PERIOD_OFFSET[period];
-    const bucketSec = BUCKET_SECONDS[period];
-    const conditions = ["rm.is_complete = 1", "rm.created_at >= datetime('now', ?)"];
-    const params = [offset];
+export function getMetricsTimeseries(db, period, metric, providerId, backendModel, routerKeyId, startTime, endTime) {
+    const bucketSec = (startTime && endTime)
+        ? calculateBucketSeconds(startTime, endTime)
+        : BUCKET_SECONDS[period];
+    const { timeWhere, timeParams } = buildTimeCondition(period, startTime, endTime);
+    const conditions = ["rm.is_complete = 1", timeWhere];
+    const params = [...timeParams];
     if (providerId) {
         conditions.push("rm.provider_id = ?");
         params.push(providerId);

package/dist/db/migrations/019_create_usage_windows.sql

@@ -0,0 +1,11 @@
+CREATE TABLE IF NOT EXISTS usage_windows (
+  id TEXT PRIMARY KEY,
+  router_key_id TEXT,
+  start_time TEXT NOT NULL,
+  end_time TEXT NOT NULL,
+  created_at TEXT DEFAULT (datetime('now'))
+);
+
+CREATE INDEX IF NOT EXISTS idx_usage_windows_start ON usage_windows(start_time);
+CREATE INDEX IF NOT EXISTS idx_usage_windows_router_key ON usage_windows(router_key_id);
+CREATE INDEX IF NOT EXISTS idx_usage_windows_end_time ON usage_windows(end_time);

package/dist/db/migrations/020_drop_log_redundancy.sql

@@ -0,0 +1,8 @@
+-- 019_drop_log_redundancy.sql
+-- 删除冗余大文本字段：request_body = client_request.body, response_body = upstream_response.body, client_response ≈ upstream_response
+ALTER TABLE request_logs DROP COLUMN request_body;
+ALTER TABLE request_logs DROP COLUMN response_body;
+ALTER TABLE request_logs DROP COLUMN client_response;
+
+-- 日志自动清理保留天数（默认 3 天，0 = 不自动清理）
+INSERT OR IGNORE INTO settings (key, value) VALUES ('log_retention_days', '3');

package/dist/db/migrations/021_merge_metrics_columns.sql

@@ -0,0 +1,28 @@
+-- V1 双写：将 request_metrics 列冗余到 request_logs，消除日志查询的 JOIN
+-- request_metrics 表保留不动，聚合查询仍查它
+
+ALTER TABLE request_logs ADD COLUMN input_tokens INTEGER;
+ALTER TABLE request_logs ADD COLUMN output_tokens INTEGER;
+ALTER TABLE request_logs ADD COLUMN cache_read_tokens INTEGER;
+ALTER TABLE request_logs ADD COLUMN ttft_ms INTEGER;
+ALTER TABLE request_logs ADD COLUMN tokens_per_second REAL;
+ALTER TABLE request_logs ADD COLUMN stop_reason TEXT;
+ALTER TABLE request_logs ADD COLUMN backend_model TEXT;
+ALTER TABLE request_logs ADD COLUMN metrics_complete INTEGER NOT NULL DEFAULT 0;
+
+-- 流式请求的累积文本内容（从 tracker.appendStreamChunk 中提取的纯文本）
+ALTER TABLE request_logs ADD COLUMN stream_text_content TEXT;
+
+-- 回填历史数据：把已有的 request_metrics 写入 request_logs 新列
+UPDATE request_logs
+SET
+  input_tokens = rm.input_tokens,
+  output_tokens = rm.output_tokens,
+  cache_read_tokens = rm.cache_read_tokens,
+  ttft_ms = rm.ttft_ms,
+  tokens_per_second = rm.tokens_per_second,
+  stop_reason = rm.stop_reason,
+  backend_model = rm.backend_model,
+  metrics_complete = rm.is_complete
+FROM request_metrics rm
+WHERE rm.request_log_id = request_logs.id;

package/dist/db/retry-rules.d.ts

@@ -25,8 +25,3 @@ export declare function createRetryRule(db: Database.Database, rule: {
 }): string;
 export declare function updateRetryRule(db: Database.Database, id: string, fields: Partial<Pick<RetryRule, "name" | "status_code" | "body_pattern" | "is_active" | "retry_strategy" | "retry_delay_ms" | "max_retries" | "max_delay_ms">>): void;
 export declare function deleteRetryRule(db: Database.Database, id: string): void;
-/**
- * 启动时按名称查重插入默认重试规则。
- * 已存在的规则不会被重复插入或覆盖。
- */
-export declare function seedDefaultRules(db: Database.Database): void;

package/dist/db/retry-rules.js

@@ -27,39 +27,3 @@ export function updateRetryRule(db, id, fields) {
 export function deleteRetryRule(db, id) {
     deleteById(db, "retry_rules", id);
 }
-// ---------- Default seed rules ----------
-const DEFAULT_RETRY_FIELDS = {
-    is_active: 1,
-    retry_strategy: "exponential",
-    retry_delay_ms: DEFAULT_RETRY_DELAY_MS,
-    max_retries: DEFAULT_MAX_RETRIES,
-    max_delay_ms: DEFAULT_MAX_DELAY_MS,
-};
-const DEFAULT_RULES = [
-    { name: "429 Too Many Requests", status_code: 429, body_pattern: ".*", ...DEFAULT_RETRY_FIELDS },
-    { name: "503 Service Unavailable", status_code: 503, body_pattern: ".*", ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI 网络错误 (code 1234)', status_code: 400, body_pattern: '"type"\\s*:\\s*"error".*"code"\\s*:\\s*"1234"', ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI 临时不可用', status_code: 400, body_pattern: '"type"\\s*:\\s*"error".*请稍后重试', ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI 操作失败 (code 500)', status_code: 400, body_pattern: '"type"\\s*:\\s*"error".*"code"\\s*:\\s*"500"', ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI 速率限制 (HTTP 200, code 1302)', status_code: 200, body_pattern: '"error".*"code"\\s*:\\s*"1302"', ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI SSE 错误 (HTTP 200, code 500)', status_code: 200, body_pattern: '"error".*"code"\\s*:\\s*"500"', ...DEFAULT_RETRY_FIELDS },
-    { name: 'ZAI SSE 错误 (HTTP 200, code 1234)', status_code: 200, body_pattern: '"error".*"code"\\s*:\\s*"1234"', ...DEFAULT_RETRY_FIELDS },
-];
-/**
- * 启动时按名称查重插入默认重试规则。
- * 已存在的规则不会被重复插入或覆盖。
- */
-export function seedDefaultRules(db) {
-    const names = DEFAULT_RULES.map(r => r.name);
-    const placeholders = names.map(() => '?').join(',');
-    const existing = db.prepare(`SELECT name FROM retry_rules WHERE name IN (${placeholders})`).all(...names);
-    const existingSet = new Set(existing.map(r => r.name));
-    const stmt = db.prepare(`INSERT INTO retry_rules (id, name, status_code, body_pattern, is_active, created_at, retry_strategy, retry_delay_ms, max_retries, max_delay_ms)
-     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`);
-    const now = new Date().toISOString();
-    for (const rule of DEFAULT_RULES) {
-        if (!existingSet.has(rule.name)) {
-            stmt.run(randomUUID(), rule.name, rule.status_code, rule.body_pattern, rule.is_active, now, rule.retry_strategy, rule.retry_delay_ms, rule.max_retries, rule.max_delay_ms);
-        }
-    }
-}

package/dist/db/settings.d.ts

@@ -2,3 +2,5 @@ import Database from "better-sqlite3";
 export declare function getSetting(db: Database.Database, key: string): string | null;
 export declare function setSetting(db: Database.Database, key: string, value: string): void;
 export declare function isInitialized(db: Database.Database): boolean;
+export declare function getLogRetentionDays(db: Database.Database): number;
+export declare function setLogRetentionDays(db: Database.Database, days: number): void;

package/dist/db/settings.js

@@ -8,3 +8,10 @@ export function setSetting(db, key, value) {
 export function isInitialized(db) {
     return getSetting(db, "initialized") === "true";
 }
+export function getLogRetentionDays(db) {
+    const val = getSetting(db, "log_retention_days");
+    return val ? parseInt(val, 10) : 3;
+}
+export function setLogRetentionDays(db, days) {
+    setSetting(db, "log_retention_days", String(days));
+}

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

@@ -0,0 +1,19 @@
+import Database from "better-sqlite3";
+export interface UsageWindow {
+    id: string;
+    router_key_id: string | null;
+    start_time: string;
+    end_time: string;
+    created_at: string;
+}
+export interface WindowUsage {
+    request_count: number;
+    total_input_tokens: number;
+    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[];
+/** 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;

package/dist/db/usage-windows.js

@@ -0,0 +1,37 @@
+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);
+    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] : [];
+    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) {
+    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);
+    }
+    return db.prepare("SELECT * FROM usage_windows WHERE start_time < ? AND end_time > ? ORDER BY start_time ASC").all(end, start);
+}
+/** 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 = `
+    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);
+}

package/dist/index.d.ts

@@ -1,6 +1,7 @@
 #!/usr/bin/env node
 import { FastifyInstance } from "fastify";
 import { Config } from "./config.js";
+import { UsageWindowTracker } from "./proxy/usage-window-tracker.js";
 import Database from "better-sqlite3";
 export interface AppOptions {
     config?: Config;
@@ -9,6 +10,7 @@ export interface AppOptions {
 export declare function buildApp(options?: AppOptions): Promise<{
     app: FastifyInstance;
     db: Database.Database;
+    usageWindowTracker: UsageWindowTracker;
     close: () => Promise<void>;
 }>;
 export declare function main(): Promise<void>;

package/dist/index.js

@@ -21,7 +21,8 @@ function getProxyApiType(url) {
 const __filename = fileURLToPath(import.meta.url);
 const __dirname = path.dirname(__filename);
 import { getConfig } from "./config.js";
-import { initDatabase, seedDefaultRules, getAllProviders } from "./db/index.js";
+import { initDatabase, getAllProviders, backfillMetricsFromRequestMetrics } from "./db/index.js";
+import { loadRecommendedConfig } from "./config/recommended.js";
 import { authMiddleware } from "./middleware/auth.js";
 import { openaiProxy } from "./proxy/openai.js";
 import { anthropicProxy } from "./proxy/anthropic.js";
@@ -30,16 +31,20 @@ import { RetryRuleMatcher } from "./proxy/retry-rules.js";
 import { ProviderSemaphoreManager } from "./proxy/semaphore.js";
 import { RequestTracker } from "./monitor/request-tracker.js";
 import { modelState } from "./proxy/model-state.js";
+import { UsageWindowTracker } from "./proxy/usage-window-tracker.js";
+import { scheduleLogCleanup } from "./db/log-cleaner.js";
 import fastifyStatic from "@fastify/static";
 export async function buildApp(options) {
     const config = options?.config ?? getBaseConfig();
     // 允许外部传入已初始化的 DB（测试用），否则自行创建
     let db;
+    let shouldBackfill = false;
     if (options?.db) {
         db = options.db;
     }
     else {
         db = initDatabase(config.DB_PATH);
+        shouldBackfill = true;
     }
     const isDev = process.env.NODE_ENV !== "production";
     const app = Fastify({
@@ -100,8 +105,14 @@ export async function buildApp(options) {
         }
         return reply.code(status).send({ error: { message: fastifyError.message } });
     });
-    // 首次启动时插入默认重试规则（表为空时）
-    seedDefaultRules(db);
+    loadRecommendedConfig();
+    // 启动时回填：补齐回退老版本期间缺失的 metrics 冗余列
+    if (shouldBackfill) {
+        const backfilled = backfillMetricsFromRequestMetrics(db);
+        if (backfilled > 0) {
+            app.log.info({ backfilled }, "Backfilled metrics from request_metrics");
+        }
+    }
     // 注入 DB 到 modelState 单例，启用会话级持久化
     modelState.init(db);
     const matcher = new RetryRuleMatcher();
@@ -109,6 +120,9 @@ export async function buildApp(options) {
     const semaphoreManager = new ProviderSemaphoreManager();
     const tracker = new RequestTracker({ semaphoreManager, logger: app.log });
     tracker.startPushInterval();
+    // 5h 用量窗口追踪器，启动时自动补齐缺失窗口
+    const usageWindowTracker = new UsageWindowTracker(db);
+    usageWindowTracker.reconcileOnStartup();
     // 从 DB 读取已有 provider 的并发配置，初始化信号量管理器和 tracker
     const allProviders = getAllProviders(db);
     for (const p of allProviders) {
@@ -169,10 +183,13 @@ export async function buildApp(options) {
     app.get("/health", async () => {
         return { status: "ok" };
     });
+    const logCleanup = scheduleLogCleanup(db, app.log);
     return {
         app,
         db,
+        usageWindowTracker,
         close: async () => {
+            logCleanup.stop();
             tracker.stopPushInterval();
             await app.close();
             db.close();

package/dist/monitor/types.d.ts

@@ -27,6 +27,7 @@ export interface ActiveRequest {
     streamMetrics?: StreamMetricsSnapshot;
     streamContent?: StreamContentSnapshot;
     clientIp?: string;
+    sessionId?: string;
     completedAt?: number;
 }
 export interface AttemptSnapshot {
@@ -38,7 +39,9 @@ export interface AttemptSnapshot {
 export interface StreamMetricsSnapshot {
     inputTokens: number | null;
     outputTokens: number | null;
+    cacheReadTokens: number | null;
     ttftMs: number | null;
+    tokensPerSecond: number | null;
     stopReason: string | null;
     isComplete: boolean;
 }