llm-simple-router 0.6.7 → 0.7.1

package/dist/core/container.d.ts

@@ -0,0 +1,31 @@
+/** 服务键常量 — 避免魔法字符串，编译期类型保护 */
+export declare const SERVICE_KEYS: {
+    readonly db: "db";
+    readonly matcher: "matcher";
+    readonly semaphoreManager: "semaphoreManager";
+    readonly tracker: "tracker";
+    readonly usageWindowTracker: "usageWindowTracker";
+    readonly sessionTracker: "sessionTracker";
+    readonly adaptiveController: "adaptiveController";
+    readonly logFileWriter: "logFileWriter";
+};
+export type ServiceKey = (typeof SERVICE_KEYS)[keyof typeof SERVICE_KEYS];
+/**
+ * 轻量服务容器 — 懒加载单例工厂注册表。
+ *
+ * 用法：
+ *   const c = new ServiceContainer();
+ *   c.register(SERVICE_KEYS.db, () => db);
+ *   c.register(SERVICE_KEYS.tracker, (c) => new RequestTracker(c.resolve(SERVICE_KEYS.db)));
+ *   const tracker = c.resolve<RequestTracker>(SERVICE_KEYS.tracker);
+ *
+ * 注册的工厂最多执行一次（惰性求值 + 缓存）。
+ */
+export declare class ServiceContainer {
+    private readonly factories;
+    private readonly cache;
+    /** 注册服务工厂。重复注册同一 key 会覆盖（但已缓存的实例不会被清除）。 */
+    register<T>(key: string, factory: (c: ServiceContainer) => T): void;
+    /** 获取服务实例。首次调用时执行工厂并缓存。 */
+    resolve<T>(key: string): T;
+}

package/dist/core/container.js

@@ -0,0 +1,41 @@
+/** 服务键常量 — 避免魔法字符串，编译期类型保护 */
+export const SERVICE_KEYS = {
+    db: "db",
+    matcher: "matcher",
+    semaphoreManager: "semaphoreManager",
+    tracker: "tracker",
+    usageWindowTracker: "usageWindowTracker",
+    sessionTracker: "sessionTracker",
+    adaptiveController: "adaptiveController",
+    logFileWriter: "logFileWriter",
+};
+/**
+ * 轻量服务容器 — 懒加载单例工厂注册表。
+ *
+ * 用法：
+ *   const c = new ServiceContainer();
+ *   c.register(SERVICE_KEYS.db, () => db);
+ *   c.register(SERVICE_KEYS.tracker, (c) => new RequestTracker(c.resolve(SERVICE_KEYS.db)));
+ *   const tracker = c.resolve<RequestTracker>(SERVICE_KEYS.tracker);
+ *
+ * 注册的工厂最多执行一次（惰性求值 + 缓存）。
+ */
+export class ServiceContainer {
+    factories = new Map();
+    cache = new Map();
+    /** 注册服务工厂。重复注册同一 key 会覆盖（但已缓存的实例不会被清除）。 */
+    register(key, factory) {
+        this.factories.set(key, factory);
+    }
+    /** 获取服务实例。首次调用时执行工厂并缓存。 */
+    resolve(key) {
+        if (this.cache.has(key))
+            return this.cache.get(key);
+        const factory = this.factories.get(key);
+        if (!factory)
+            throw new Error(`Service not registered: "${key}"`);
+        const instance = factory(this);
+        this.cache.set(key, instance);
+        return instance;
+    }
+}

package/dist/core/errors.d.ts

@@ -0,0 +1,26 @@
+import type { TransportResult, ResilienceAttempt } from "./types.js";
+/**
+ * Provider 并发队列已满时抛出。
+ */
+export declare class SemaphoreQueueFullError extends Error {
+    readonly providerId: string;
+    constructor(providerId: string);
+}
+/**
+ * Provider 并发等待超时时抛出。
+ */
+export declare class SemaphoreTimeoutError extends Error {
+    readonly providerId: string;
+    readonly timeoutMs: number;
+    constructor(providerId: string, timeoutMs: number);
+}
+/**
+ * 跨 provider failover 时由 ResilienceLayer 抛出，
+ * orchestrator 捕获后释放当前信号量并获取新 provider 的信号量。
+ */
+export declare class ProviderSwitchNeeded extends Error {
+    readonly targetProviderId: string;
+    readonly attempts?: ResilienceAttempt[] | undefined;
+    readonly lastResult?: TransportResult | undefined;
+    constructor(targetProviderId: string, attempts?: ResilienceAttempt[] | undefined, lastResult?: TransportResult | undefined);
+}

package/dist/core/errors.js

@@ -0,0 +1,42 @@
+// src/core/errors.ts
+// 被多目录共享的错误类型（从 proxy/semaphore.ts 和 proxy/types.ts 移出）
+/**
+ * Provider 并发队列已满时抛出。
+ */
+export class SemaphoreQueueFullError extends Error {
+    providerId;
+    constructor(providerId) {
+        super(`Provider '${providerId}' concurrency queue is full`);
+        this.providerId = providerId;
+        this.name = "SemaphoreQueueFullError";
+    }
+}
+/**
+ * Provider 并发等待超时时抛出。
+ */
+export class SemaphoreTimeoutError extends Error {
+    providerId;
+    timeoutMs;
+    constructor(providerId, timeoutMs) {
+        super(`Provider '${providerId}' concurrency wait timeout (${timeoutMs}ms)`);
+        this.providerId = providerId;
+        this.timeoutMs = timeoutMs;
+        this.name = "SemaphoreTimeoutError";
+    }
+}
+/**
+ * 跨 provider failover 时由 ResilienceLayer 抛出，
+ * orchestrator 捕获后释放当前信号量并获取新 provider 的信号量。
+ */
+export class ProviderSwitchNeeded extends Error {
+    targetProviderId;
+    attempts;
+    lastResult;
+    constructor(targetProviderId, attempts, lastResult) {
+        super(`Provider switch needed: ${targetProviderId}`);
+        this.targetProviderId = targetProviderId;
+        this.attempts = attempts;
+        this.lastResult = lastResult;
+        this.name = "ProviderSwitchNeeded";
+    }
+}

package/dist/core/registry.d.ts

@@ -0,0 +1,43 @@
+import type { ConcurrencyConfig } from "./types.js";
+export type { ConcurrencyConfig };
+export interface EnhancementConfig {
+    claude_code_enabled: boolean;
+    tool_call_loop_enabled: boolean;
+    stream_loop_enabled: boolean;
+}
+/** Provider 自适应/手动并发配置（DB 字段） */
+export interface ProviderConcurrencyParams {
+    adaptive_enabled: number;
+    max_concurrency: number;
+    queue_timeout_ms: number;
+    max_queue_size: number;
+}
+export interface StateRegistry {
+    /** 刷新重试规则缓存（RetryRuleMatcher.load） */
+    refreshRetryRules(): void;
+    /** 更新 provider 并发配置（ProviderSemaphoreManager.updateConfig） */
+    updateProviderConcurrency(providerId: string, config: ConcurrencyConfig): void;
+    /** 移除 provider 的信号量（ProviderSemaphoreManager.remove） */
+    removeProvider(providerId: string): void;
+    /** 移除所有信号量配置（ProviderSemaphoreManager.removeAll） */
+    removeAllProviders(): void;
+    /** 获取 provider 并发状态（ProviderSemaphoreManager.getStatus） */
+    getProviderStatus(providerId: string): {
+        active: number;
+        queued: number;
+    };
+    /** 清空所有会话模型状态（modelState.clearAll） */
+    clearModelState(): void;
+    /** 删除指定会话模型状态（modelState.delete） */
+    deleteModelState(keyId: string, sessionId: string): void;
+    /** 读取 proxy enhancement 配置 */
+    getEnhancementConfig(): EnhancementConfig;
+    /** 同步 provider 的自适应并发配置（AdaptiveConcurrencyController.syncProvider） */
+    syncAdaptiveProvider(providerId: string, params: ProviderConcurrencyParams): void;
+    /** 移除 provider 的自适应并发状态（AdaptiveConcurrencyController.remove） */
+    removeAdaptiveProvider(providerId: string): void;
+    /** 获取 provider 的自适应并发状态 */
+    getAdaptiveStatus(providerId: string): import("../proxy/adaptive-controller.js").AdaptiveState | undefined;
+    /** 从 DB 重新读取所有 provider 配置，重建信号量/adaptive/tracker 缓存（导入配置后调用） */
+    reinitializeProviders(): void;
+}

package/dist/core/registry.js

@@ -0,0 +1,3 @@
+// src/core/registry.ts
+// Admin 层通过此接口触发 proxy 层状态刷新，消除 admin→proxy 直接依赖
+export {};

package/dist/core/types.d.ts

@@ -0,0 +1,105 @@
+export interface Target {
+    backend_model: string;
+    provider_id: string;
+    overflow_provider_id?: string;
+    overflow_model?: string;
+}
+export interface ResolveContext {
+    now: Date;
+    excludeTargets?: Target[];
+}
+/** Provider 级并发控制配置（唯一来源，替代 semaphore.ts 和 registry.ts 中的重复定义） */
+export interface ConcurrencyConfig {
+    maxConcurrency: number;
+    queueTimeoutMs: number;
+    maxQueueSize: number;
+}
+export interface ConcurrencyOverride {
+    max_concurrency?: number;
+    queue_timeout_ms?: number;
+    max_queue_size?: number;
+}
+export interface ResolveResult {
+    target: Target;
+    concurrency_override?: ConcurrencyOverride;
+    /** 活跃规则（schedule 或 base）中的 target 总数，用于 failover 判断 */
+    targetCount: number;
+}
+export interface MetricsResult {
+    input_tokens: number | null;
+    output_tokens: number | null;
+    cache_creation_tokens: number | null;
+    cache_read_tokens: number | null;
+    ttft_ms: number | null;
+    /** T6 - T0: proxy end-to-end streaming duration */
+    total_duration_ms: number | null;
+    /** @deprecated Use total_tps instead */
+    tokens_per_second: number | null;
+    stop_reason: string | null;
+    is_complete: number;
+    input_tokens_estimated?: number;
+    thinking_tokens: number | null;
+    /** T3 - T0: request start to last thinking delta */
+    thinking_duration_ms: number | null;
+    thinking_tps: number | null;
+    /** T6 - T3 (thinking) or T6 - T0 (non-thinking) */
+    non_thinking_duration_ms: number | null;
+    non_thinking_tps: number | null;
+    total_tps: number | null;
+    text_tokens: number | null;
+    tool_use_tokens: number | null;
+}
+export type RawHeaders = Record<string, string | string[] | undefined>;
+export type TransportResult = {
+    kind: "success";
+    statusCode: number;
+    body: string;
+    headers: Record<string, string>;
+    sentHeaders: Record<string, string>;
+    sentBody: string;
+} | {
+    kind: "stream_success";
+    statusCode: number;
+    metrics?: MetricsResult;
+    upstreamResponseHeaders?: Record<string, string>;
+    sentHeaders: Record<string, string>;
+} | {
+    kind: "stream_error";
+    statusCode: number;
+    body: string;
+    headers: Record<string, string>;
+    sentHeaders: Record<string, string>;
+    headersSent?: boolean;
+} | {
+    kind: "stream_abort";
+    statusCode: number;
+    metrics?: MetricsResult;
+    upstreamResponseHeaders?: Record<string, string>;
+    sentHeaders: Record<string, string>;
+} | {
+    kind: "error";
+    statusCode: number;
+    body: string;
+    headers: Record<string, string>;
+    sentHeaders: Record<string, string>;
+    sentBody: string;
+} | {
+    kind: "throw";
+    error: Error;
+    headersSent?: boolean;
+};
+/** 单次 resilience 尝试的记录 */
+export interface ResilienceAttempt {
+    target: Target;
+    attemptIndex: number;
+    statusCode: number | null;
+    error: string | null;
+    latencyMs: number;
+    responseBody: string | null;
+    /** 上游响应 headers（throw 和 stream_success/stream_abort 时为 null） */
+    responseHeaders: Record<string, string> | null;
+    /** TransportResult.kind，用于区分 stream_error 等特殊类型 */
+    resultKind: TransportResult["kind"];
+}
+/** 流式传输阶段状态 */
+export type StreamState = "BUFFERING" | "STREAMING" | "COMPLETED" | "EARLY_ERROR" | "ABORTED";

package/dist/core/types.js

@@ -0,0 +1,3 @@
+// src/core/types.ts
+// 被多个目录（proxy, db, monitor, admin）共享的类型定义
+export {};

package/dist/db/index.d.ts

@@ -6,7 +6,7 @@ export { getMappingGroup, getMappingGroupById, getAllMappingGroups, createMappin
 export type { MappingGroup, ProviderModelEntry } from "./mappings.js";
 export { getActiveRetryRules, getAllRetryRules, getRetryRuleById, createRetryRule, updateRetryRule, deleteRetryRule, } from "./retry-rules.js";
 export type { RetryRule } from "./retry-rules.js";
-export { insertRequestLog, getRequestLogs, getRequestLogById, deleteLogsBefore, getRequestLogChildren, getRequestLogsGrouped, updateLogMetrics, updateLogStreamContent, updateLogClientStatus, backfillMetricsFromRequestMetrics, estimateLogTableSize, deleteOldestLogs, getLogCount, } from "./logs.js";
+export { insertRequestLog, getRequestLogs, getRequestLogById, deleteLogsBefore, getRequestLogChildren, getRequestLogsGrouped, updateLogStreamContent, updateLogClientStatus, estimateLogTableSize, deleteOldestLogs, getLogCount, updateLogPipelineSnapshot, } from "./logs.js";
 export type { RequestLog, RequestLogGroupedRow, RequestLogListRow } from "./logs.js";
 export { getRouterKeyByHash, getAllRouterKeys, getRouterKeyById, createRouterKey, updateRouterKey, deleteRouterKey, getAvailableModels, } from "./router-keys.js";
 export type { RouterKey } from "./router-keys.js";

package/dist/db/index.js

@@ -85,7 +85,7 @@ export function initDatabase(dbPath) {
 export { getActiveProviders, getAllProviders, getProviderById, getActiveProviderByName, getActiveProvidersWithModels, createProvider, updateProvider, deleteProvider, PROVIDER_CONCURRENCY_DEFAULTS, } from "./providers.js";
 export { getMappingGroup, getMappingGroupById, getAllMappingGroups, createMappingGroup, updateMappingGroup, deleteMappingGroup, getActiveProviderModels, resolveByProviderModel, } from "./mappings.js";
 export { getActiveRetryRules, getAllRetryRules, getRetryRuleById, createRetryRule, updateRetryRule, deleteRetryRule, } from "./retry-rules.js";
-export { insertRequestLog, getRequestLogs, getRequestLogById, deleteLogsBefore, getRequestLogChildren, getRequestLogsGrouped, updateLogMetrics, updateLogStreamContent, updateLogClientStatus, backfillMetricsFromRequestMetrics, estimateLogTableSize, deleteOldestLogs, getLogCount, } from "./logs.js";
+export { insertRequestLog, getRequestLogs, getRequestLogById, deleteLogsBefore, getRequestLogChildren, getRequestLogsGrouped, updateLogStreamContent, updateLogClientStatus, estimateLogTableSize, deleteOldestLogs, getLogCount, updateLogPipelineSnapshot, } from "./logs.js";
 export { getRouterKeyByHash, getAllRouterKeys, getRouterKeyById, createRouterKey, updateRouterKey, deleteRouterKey, getAvailableModels, } from "./router-keys.js";
 export { getMetricsSummary, getMetricsTimeseries, insertMetrics } from "./metrics.js";
 export { getStats } from "./stats.js";

package/dist/db/logs.d.ts

@@ -1,4 +1,6 @@
 import Database from "better-sqlite3";
+import type { LogFileWriter } from "../storage/log-file-writer.js";
+import { type RetryMatcher } from "../proxy/log-detail-policy.js";
 export interface RequestLog {
     id: string;
     api_type: string;
@@ -17,14 +19,6 @@ export interface RequestLog {
     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;
     session_id: string | null;
 }
@@ -53,8 +47,14 @@ export interface RequestLogInsert {
     original_model?: string | null;
     session_id?: string | null;
     client_status_code?: number | null;
+    pipeline_snapshot?: string | null;
 }
-export declare function insertRequestLog(db: Database.Database, log: RequestLogInsert): void;
+export interface LogWriteContext {
+    matcher?: RetryMatcher | null;
+    logFileWriter?: LogFileWriter | null;
+    responseBody?: string | null;
+}
+export declare function insertRequestLog(db: Database.Database, log: RequestLogInsert, writeContext?: LogWriteContext): void;
 export declare function getRequestLogs(db: Database.Database, options: {
     page: number;
     limit: number;
@@ -69,24 +69,10 @@ export declare function getRequestLogs(db: Database.Database, options: {
     total: number;
 };
 export declare function getRequestLogById(db: Database.Database, id: string): RequestLogListRow | 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;
-    input_tokens_estimated?: 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;
 /** 当 router 返回给客户端的 status code 与上游不同时，记录实际发送的 status */
 export declare function updateLogClientStatus(db: Database.Database, logId: string, clientStatusCode: number): 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;
 /** 估算 request_logs 表占用字节数 */
 export declare function estimateLogTableSize(db: Database.Database): number;
@@ -113,4 +99,5 @@ export declare function getRequestLogsGrouped(db: Database.Database, options: {
     data: RequestLogGroupedRow[];
     total: number;
 };
-export {};
+/** 后续 pipeline 阶段完成后，回写 snapshot 到已有日志 */
+export declare function updateLogPipelineSnapshot(db: Database.Database, logId: string, snapshot: string): void;

package/dist/db/logs.js

@@ -1,17 +1,35 @@
+import { shouldPreserveDetail } from "../proxy/log-detail-policy.js";
 // --- request_logs ---
-/** 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.client_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,
-            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, rl.session_id, rl.input_tokens_estimated,
+            rl.session_id, rl.pipeline_snapshot,
+            rm.input_tokens, rm.output_tokens, rm.cache_read_tokens, rm.ttft_ms,
+            rm.tokens_per_second, rm.stop_reason, rm.backend_model, rm.is_complete AS metrics_complete,
+            rm.input_tokens_estimated,
             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) {
+const LOG_LIST_JOIN = `LEFT JOIN providers p ON p.id = rl.provider_id LEFT JOIN request_metrics rm ON rm.request_log_id = rl.id`;
+export function insertRequestLog(db, log, writeContext) {
+    // 文件写入：始终写入全文
+    if (writeContext?.logFileWriter) {
+        writeContext.logFileWriter.write({
+            id: log.id,
+            created_at: log.created_at,
+            api_type: log.api_type,
+            status_code: log.status_code,
+            client_request: log.client_request ?? null,
+            upstream_request: log.upstream_request ?? null,
+            upstream_response: log.upstream_response ?? null,
+            stream_text_content: null,
+            pipeline_snapshot: log.pipeline_snapshot ?? null,
+        });
+    }
+    // 详情保留判定
+    const preserveDetail = shouldPreserveDetail(log.status_code, writeContext?.responseBody ?? null, writeContext?.matcher ?? null, !!writeContext?.logFileWriter);
     db.prepare(`INSERT INTO request_logs (id, api_type, model, provider_id, status_code, client_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, session_id)
-     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(log.id, log.api_type, log.model, log.provider_id, log.status_code, log.client_status_code ?? null, 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, log.session_id ?? null);
+      is_retry, is_failover, original_request_id, router_key_id, original_model, session_id, pipeline_snapshot)
+     VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)`).run(log.id, log.api_type, log.model, log.provider_id, log.status_code, log.client_status_code ?? null, log.latency_ms, log.is_stream, log.error_message, log.created_at, preserveDetail ? (log.client_request ?? null) : null, preserveDetail ? (log.upstream_request ?? null) : null, preserveDetail ? (log.upstream_response ?? null) : null, log.is_retry ?? 0, log.is_failover ?? 0, log.original_request_id ?? null, log.router_key_id ?? null, log.original_model ?? null, log.session_id ?? null, log.pipeline_snapshot ?? null);
 }
 function buildLogWhereClause(options, baseCondition) {
     let where = baseCondition;
@@ -63,19 +81,15 @@ export function getRequestLogs(db, options) {
     return { data, total };
 }
 export function getRequestLogById(db, id) {
-    return db.prepare(`SELECT rl.*, COALESCE(p.name, rl.provider_id) AS provider_name
-     FROM request_logs rl LEFT JOIN providers p ON p.id = rl.provider_id
+    return db.prepare(`SELECT rl.*, rm.input_tokens, rm.output_tokens, rm.cache_read_tokens, rm.ttft_ms,
+            rm.tokens_per_second, rm.stop_reason, rm.backend_model, rm.is_complete AS metrics_complete,
+            rm.input_tokens_estimated,
+            COALESCE(p.name, rl.provider_id) AS provider_name
+     FROM request_logs rl
+     LEFT JOIN providers p ON p.id = rl.provider_id
+     LEFT JOIN request_metrics rm ON rm.request_log_id = rl.id
      WHERE rl.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 = ?, input_tokens_estimated = ?
-     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, m.input_tokens_estimated ?? 0, 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);
@@ -84,25 +98,6 @@ export function updateLogStreamContent(db, logId, textContent) {
 export function updateLogClientStatus(db, logId, clientStatusCode) {
     db.prepare("UPDATE request_logs SET client_status_code = ? WHERE id = ?").run(clientStatusCode, 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,
-      input_tokens_estimated = rm.input_tokens_estimated
-    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) {
     const changes = db.prepare("DELETE FROM request_logs WHERE created_at < ?").run(beforeDate).changes;
     if (changes > 0) {
@@ -118,7 +113,7 @@ export function estimateLogTableSize(db) {
     SELECT COALESCE(SUM(
       COALESCE(length(client_request), 0) + COALESCE(length(upstream_request), 0) +
       COALESCE(length(upstream_response), 0) + COALESCE(length(stream_text_content), 0) +
-      COALESCE(length(error_message), 0) + ?
+      COALESCE(length(error_message), 0) + COALESCE(length(pipeline_snapshot), 0) + ?
     ), 0) as size
     FROM request_logs
   `).get(ROW_METADATA_BYTES);
@@ -177,3 +172,7 @@ export function getRequestLogsGrouped(db, options) {
         .all(...params, options.limit, offset);
     return { data, total };
 }
+/** 后续 pipeline 阶段完成后，回写 snapshot 到已有日志 */
+export function updateLogPipelineSnapshot(db, logId, snapshot) {
+    db.prepare("UPDATE request_logs SET pipeline_snapshot = ? WHERE id = ?").run(snapshot, logId);
+}

package/dist/db/metrics.js

@@ -1,5 +1,5 @@
 import { randomUUID } from "crypto";
-import { MS_PER_SECOND } from "../constants.js";
+import { MS_PER_SECOND } from "../core/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, router_key_id, status_code,

package/dist/db/migrations/033_add_pipeline_snapshot.sql

@@ -0,0 +1 @@
+ALTER TABLE request_logs ADD COLUMN pipeline_snapshot TEXT;

package/dist/db/migrations/034_drop_redundant_log_columns.sql

@@ -0,0 +1,13 @@
+-- 034_drop_redundant_log_columns.sql
+-- request_logs 与 request_metrics 双写冗余清理：
+-- metrics 字段统一由 request_metrics 承载，日志列表查询改用 LEFT JOIN。
+
+ALTER TABLE request_logs DROP COLUMN input_tokens;
+ALTER TABLE request_logs DROP COLUMN output_tokens;
+ALTER TABLE request_logs DROP COLUMN cache_read_tokens;
+ALTER TABLE request_logs DROP COLUMN ttft_ms;
+ALTER TABLE request_logs DROP COLUMN tokens_per_second;
+ALTER TABLE request_logs DROP COLUMN stop_reason;
+ALTER TABLE request_logs DROP COLUMN backend_model;
+ALTER TABLE request_logs DROP COLUMN metrics_complete;
+ALTER TABLE request_logs DROP COLUMN input_tokens_estimated;

package/dist/db/settings.d.ts

@@ -10,3 +10,5 @@ export declare function getLogTableMaxSizeMb(db: Database.Database): number;
 export declare function setLogTableMaxSizeMb(db: Database.Database, mb: number): void;
 export declare function getConfigSyncSource(db: Database.Database): "github" | "gitee";
 export declare function setConfigSyncSource(db: Database.Database, source: "github" | "gitee"): void;
+export declare function getDetailLogEnabled(db: Database.Database): boolean;
+export declare function getLogFileRetentionDays(db: Database.Database): number;

package/dist/db/settings.js

@@ -39,3 +39,12 @@ export function getConfigSyncSource(db) {
 export function setConfigSyncSource(db, source) {
     setSetting(db, "config_sync_source", source);
 }
+export function getDetailLogEnabled(db) {
+    const row = db.prepare("SELECT value FROM settings WHERE key = ?").get("detail_log_enabled");
+    return row ? row.value !== "0" : true;
+}
+const DEFAULT_LOG_FILE_RETENTION_DAYS = 3;
+export function getLogFileRetentionDays(db) {
+    const row = db.prepare("SELECT value FROM settings WHERE key = ?").get("log_file_retention_days");
+    return row ? parseInt(row.value, 10) : DEFAULT_LOG_FILE_RETENTION_DAYS;
+}

package/dist/index.d.ts

@@ -1,7 +1,10 @@
 #!/usr/bin/env node
 import { FastifyInstance } from "fastify";
-import { Config } from "./config.js";
-import { UsageWindowTracker } from "./proxy/usage-window-tracker.js";
+import { Config } from "./config/index.js";
+import { ProviderSemaphoreManager } from "./proxy/orchestration/semaphore.js";
+import { AdaptiveConcurrencyController } from "./proxy/adaptive-controller.js";
+import { RequestTracker } from "./monitor/request-tracker.js";
+import { UsageWindowTracker } from "./proxy/routing/usage-window-tracker.js";
 import { CheckerOptions } from "./upgrade/checker.js";
 import Database from "better-sqlite3";
 export interface AppOptions {
@@ -9,6 +12,11 @@ export interface AppOptions {
     db?: Database.Database;
     upgradeCheckerOptions?: CheckerOptions;
 }
+/**
+ * 共享初始化逻辑 — 启动时和导入配置后都需要调用。
+ * 从 DB 读取所有 provider，初始化信号量/自适应并发/tracker 缓存。
+ */
+export declare function initializeProviderState(db: Database.Database, semaphoreManager: ProviderSemaphoreManager, adaptiveController: AdaptiveConcurrencyController, tracker: RequestTracker): void;
 export declare function buildApp(options?: AppOptions): Promise<{
     app: FastifyInstance;
     db: Database.Database;