llm-simple-router 0.4.3 → 0.5.1

package/dist/db/log-cleaner.js

@@ -13,7 +13,8 @@ export function runLogCleanup(db) {
 /** 启动定时清理，返回 handle 用于停止 */
 export function scheduleLogCleanup(db, log) {
     let cleaning = false;
-    let timer = null;
+    let initialTimer = null;
+    let intervalTimer = null;
     const doCleanup = () => {
         if (cleaning)
             return;
@@ -23,19 +24,27 @@ export function scheduleLogCleanup(db, log) {
             if (deleted > 0)
                 log.info(`Log cleanup: deleted ${deleted} records`);
         }
+        catch (e) {
+            // DB 可能已关闭（测试清理、进程关闭等）
+            log.info(`Log cleanup skipped: ${e instanceof Error ? e.message : String(e)}`);
+        }
         finally {
             cleaning = false;
         }
     };
-    // 启动时立即执行一次
-    doCleanup();
+    // 推迟到下一个事件循环 tick，避免阻塞服务器启动
+    initialTimer = setTimeout(doCleanup, 0);
     // 定时执行
-    timer = setInterval(doCleanup, CLEANUP_INTERVAL_MS);
+    intervalTimer = setInterval(doCleanup, CLEANUP_INTERVAL_MS);
     return {
         stop: () => {
-            if (timer) {
-                clearInterval(timer);
-                timer = null;
+            if (initialTimer) {
+                clearTimeout(initialTimer);
+                initialTimer = null;
+            }
+            if (intervalTimer) {
+                clearInterval(intervalTimer);
+                intervalTimer = null;
             }
         },
     };

package/dist/db/logs.d.ts

@@ -25,6 +25,7 @@ export interface RequestLog {
     backend_model: string | null;
     metrics_complete: number;
     stream_text_content: string | null;
+    session_id: string | null;
 }
 /** 列表查询扩展字段：JOIN providers 获得 provider_name */
 export interface RequestLogListRow extends RequestLog {
@@ -49,6 +50,7 @@ export interface RequestLogInsert {
     original_request_id?: string | null;
     router_key_id?: string | null;
     original_model?: string | null;
+    session_id?: string | null;
 }
 export declare function insertRequestLog(db: Database.Database, log: RequestLogInsert): void;
 export declare function getRequestLogs(db: Database.Database, options: {
@@ -64,7 +66,7 @@ export declare function getRequestLogs(db: Database.Database, options: {
     data: RequestLogListRow[];
     total: number;
 };
-export declare function getRequestLogById(db: Database.Database, id: string): RequestLog | undefined;
+export declare function getRequestLogById(db: Database.Database, id: string): RequestLogListRow | undefined;
 type MetricsUpdate = {
     input_tokens?: number | null;
     output_tokens?: number | null;
@@ -81,6 +83,12 @@ export declare function updateLogStreamContent(db: Database.Database, logId: str
 /** 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;
+/** 删除最旧的日志，保留 keepCount 条，返回实际删除条数。分批删除避免长时间锁表 */
+export declare function deleteOldestLogs(db: Database.Database, keepCount: number): number;
+/** 获取 request_logs 总行数 */
+export declare function getLogCount(db: Database.Database): 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[];
 export interface RequestLogGroupedRow extends RequestLogListRow {

package/dist/db/logs.js

@@ -4,14 +4,14 @@ const LOG_LIST_SELECT = `rl.id, rl.api_type, rl.model, rl.provider_id, rl.status
             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.backend_model, rl.metrics_complete, rl.session_id,
             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, 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);
+      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.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);
 }
 function buildLogWhereClause(options, baseCondition) {
     let where = baseCondition;
@@ -55,7 +55,9 @@ export function getRequestLogs(db, options) {
     return { data, total };
 }
 export function getRequestLogById(db, id) {
-    return db.prepare("SELECT * FROM request_logs WHERE id = ?").get(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
+     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) {
@@ -89,7 +91,55 @@ export function backfillMetricsFromRequestMetrics(db) {
   `).run().changes;
 }
 export function deleteLogsBefore(db, beforeDate) {
-    return db.prepare("DELETE FROM request_logs WHERE created_at < ?").run(beforeDate).changes;
+    const changes = db.prepare("DELETE FROM request_logs WHERE created_at < ?").run(beforeDate).changes;
+    if (changes > 0) {
+        db.pragma("incremental_vacuum");
+    }
+    return changes;
+}
+/** 每行元数据（数字列+索引）的估算字节数 */
+const ROW_METADATA_BYTES = 500;
+/** 估算 request_logs 表占用字节数 */
+export function estimateLogTableSize(db) {
+    const row = db.prepare(`
+    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) + ?
+    ), 0) as size
+    FROM request_logs
+  `).get(ROW_METADATA_BYTES);
+    return row.size;
+}
+const DELETE_BATCH_SIZE = 1000;
+/** 删除最旧的日志，保留 keepCount 条，返回实际删除条数。分批删除避免长时间锁表 */
+export function deleteOldestLogs(db, keepCount) {
+    const total = db.prepare("SELECT count(*) as c FROM request_logs").get().c;
+    const toDelete = Math.max(0, total - keepCount);
+    if (toDelete === 0)
+        return 0;
+    let totalDeleted = 0;
+    const stmt = db.prepare(`
+    DELETE FROM request_logs
+    WHERE rowid IN (
+      SELECT rowid FROM request_logs ORDER BY created_at ASC LIMIT ?
+    )
+  `);
+    while (totalDeleted < toDelete) {
+        const batchSize = Math.min(DELETE_BATCH_SIZE, toDelete - totalDeleted);
+        const result = stmt.run(batchSize);
+        totalDeleted += result.changes;
+        if (result.changes < batchSize)
+            break;
+    }
+    if (totalDeleted > 0) {
+        db.pragma("incremental_vacuum");
+    }
+    return totalDeleted;
+}
+/** 获取 request_logs 总行数 */
+export function getLogCount(db) {
+    return db.prepare("SELECT count(*) as c FROM request_logs").get().c;
 }
 /** 查询某条日志的子请求（retry/failover 关联），上限 100 条 */
 export function getRequestLogChildren(db, parentId) {

package/dist/db/migrations/022_add_session_id_and_incremental_vacuum.sql

@@ -0,0 +1,5 @@
+ALTER TABLE request_logs ADD COLUMN session_id TEXT;
+
+-- NOTE: PRAGMA auto_vacuum only takes effect after VACUUM for existing databases.
+-- Run scripts/vacuum-migrate.js for existing deployments.
+PRAGMA auto_vacuum = 2;

package/dist/db/settings.d.ts

@@ -4,3 +4,9 @@ export declare function setSetting(db: Database.Database, key: string, value: st
 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;
+export declare function getDbMaxSizeMb(db: Database.Database): number;
+export declare function setDbMaxSizeMb(db: Database.Database, mb: number): void;
+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;

package/dist/db/settings.js

@@ -10,8 +10,32 @@ export function isInitialized(db) {
 }
 export function getLogRetentionDays(db) {
     const val = getSetting(db, "log_retention_days");
-    return val ? parseInt(val, 10) : 3;
+    const DEFAULT_LOG_RETENTION_DAYS = 3;
+    return val ? parseInt(val, 10) : DEFAULT_LOG_RETENTION_DAYS;
 }
 export function setLogRetentionDays(db, days) {
     setSetting(db, "log_retention_days", String(days));
 }
+const DEFAULT_DB_MAX_SIZE_MB = 1024;
+const DEFAULT_LOG_TABLE_MAX_SIZE_MB = 800;
+export function getDbMaxSizeMb(db) {
+    const val = getSetting(db, "db_max_size_mb");
+    return val ? parseInt(val, 10) : DEFAULT_DB_MAX_SIZE_MB;
+}
+export function setDbMaxSizeMb(db, mb) {
+    setSetting(db, "db_max_size_mb", String(mb));
+}
+export function getLogTableMaxSizeMb(db) {
+    const val = getSetting(db, "log_table_max_size_mb");
+    return val ? parseInt(val, 10) : DEFAULT_LOG_TABLE_MAX_SIZE_MB;
+}
+export function setLogTableMaxSizeMb(db, mb) {
+    setSetting(db, "log_table_max_size_mb", String(mb));
+}
+export function getConfigSyncSource(db) {
+    const val = getSetting(db, "config_sync_source");
+    return val === "gitee" ? "gitee" : "github";
+}
+export function setConfigSyncSource(db, source) {
+    setSetting(db, "config_sync_source", source);
+}

package/dist/index.d.ts

@@ -2,10 +2,12 @@
 import { FastifyInstance } from "fastify";
 import { Config } from "./config.js";
 import { UsageWindowTracker } from "./proxy/usage-window-tracker.js";
+import { CheckerOptions } from "./upgrade/checker.js";
 import Database from "better-sqlite3";
 export interface AppOptions {
     config?: Config;
     db?: Database.Database;
+    upgradeCheckerOptions?: CheckerOptions;
 }
 export declare function buildApp(options?: AppOptions): Promise<{
     app: FastifyInstance;

package/dist/index.js

@@ -33,6 +33,8 @@ 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 { scheduleDbSizeMonitor } from "./db/db-size-monitor.js";
+import { startUpgradeChecker, stopUpgradeChecker } from "./admin/upgrade.js";
 import fastifyStatic from "@fastify/static";
 export async function buildApp(options) {
     const config = options?.config ?? getBaseConfig();
@@ -106,6 +108,7 @@ export async function buildApp(options) {
         return reply.code(status).send({ error: { message: fastifyError.message } });
     });
     loadRecommendedConfig();
+    startUpgradeChecker(options?.upgradeCheckerOptions);
     // 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);
@@ -184,12 +187,17 @@ export async function buildApp(options) {
         return { status: "ok" };
     });
     const logCleanup = scheduleLogCleanup(db, app.log);
+    const dbSizeMonitor = scheduleDbSizeMonitor(db, config.DB_PATH, {
+        log: app.log,
+    });
     return {
         app,
         db,
         usageWindowTracker,
         close: async () => {
+            stopUpgradeChecker();
             logCleanup.stop();
+            dbSizeMonitor.stop();
             tracker.stopPushInterval();
             await app.close();
             db.close();

package/dist/metrics/metrics-extractor.js

@@ -92,6 +92,10 @@ export class MetricsExtractor {
         else if (type === "message_delta") {
             const msg = parsed;
             this.outputTokens = msg.usage?.output_tokens ?? null;
+            // 第三方 Anthropic 兼容 API（如 OpenRouter、智谱）可能将 input_tokens 放在 message_delta 而非 message_start
+            if (this.inputTokens === null && msg.usage?.input_tokens) {
+                this.inputTokens = msg.usage.input_tokens;
+            }
             this.stopReason = msg.delta?.stop_reason ?? null;
             this.streamEndTime = Date.now();
         }

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

@@ -34,6 +34,8 @@ export declare class RequestTracker {
     getActive(): ActiveRequest[];
     getRecent(limit?: number): ActiveRequest[];
     get(id: string): ActiveRequest | undefined;
+    /** Public alias for API endpoint use — returns full request data including clientRequest */
+    getRequestById(id: string): ActiveRequest | undefined;
     getStats(): StatsSnapshot;
     getConcurrency(): ProviderConcurrencySnapshot[];
     getRuntime(): RuntimeMetrics;

package/dist/monitor/request-tracker.js

@@ -135,6 +135,10 @@ export class RequestTracker {
     get(id) {
         return this.activeMap.get(id) ?? this.recentCompleted.find((r) => r.id === id);
     }
+    /** Public alias for API endpoint use — returns full request data including clientRequest */
+    getRequestById(id) {
+        return this.get(id);
+    }
     // --- Stats / monitoring ---
     getStats() {
         return this.statsAggregator.getStats();
@@ -195,7 +199,22 @@ export class RequestTracker {
         }
     }
     broadcast(event, data) {
-        const msg = `event: ${event}\ndata: ${JSON.stringify(data)}\n\n`;
+        // Strip clientRequest from broadcasts to reduce bandwidth;
+        // full data available on-demand via API endpoint
+        let payload = data;
+        if (event === "request_update" && Array.isArray(data)) {
+            payload = data.map((req) => {
+                const copy = { ...req };
+                delete copy.clientRequest;
+                return copy;
+            });
+        }
+        else if (event === "request_complete" && data && typeof data === "object") {
+            const copy = { ...data };
+            delete copy.clientRequest;
+            payload = copy;
+        }
+        const msg = `event: ${event}\ndata: ${JSON.stringify(payload)}\n\n`;
         const clientCount = this.clients.size;
         let sentCount = 0;
         for (const client of this.clients) {

package/dist/monitor/types.d.ts

@@ -28,6 +28,7 @@ export interface ActiveRequest {
     streamContent?: StreamContentSnapshot;
     clientIp?: string;
     sessionId?: string;
+    clientRequest?: string;
     completedAt?: number;
 }
 export interface AttemptSnapshot {

package/dist/proxy/log-helpers.d.ts

@@ -24,6 +24,7 @@ export interface RequestLogParams extends LogRetryMeta {
     upHdrs: Record<string, string>;
     routerKeyId?: string | null;
     originalModel?: string | null;
+    sessionId?: string | null;
 }
 /** 插入成功请求日志，供 openai/anthropic 插件共享 */
 export declare function insertSuccessLog(db: Database.Database, params: RequestLogParams): void;
@@ -41,6 +42,7 @@ export interface RejectedLogParams extends LogRetryMeta {
     clientHeaders: RawHeaders;
     providerId?: string | null;
     originalModel?: string | null;
+    sessionId?: string | null;
 }
 /** Log a request rejected before reaching upstream */
 export declare function insertRejectedLog(params: RejectedLogParams): void;

package/dist/proxy/log-helpers.js

@@ -1,7 +1,7 @@
 import { insertRequestLog } from "../db/index.js";
 /** 插入成功请求日志，供 openai/anthropic 插件共享 */
 export function insertSuccessLog(db, params) {
-    const { id: logId, apiType, model, provider, isStream, startTime, clientReq, upstreamReq, status, respBody, upHdrs, isRetry = false, isFailover = false, originalRequestId = null, routerKeyId = null, originalModel = null } = params;
+    const { id: logId, apiType, model, provider, isStream, startTime, clientReq, upstreamReq, status, respBody, upHdrs, isRetry = false, isFailover = false, originalRequestId = null, routerKeyId = null, originalModel = null, sessionId = null } = params;
     insertRequestLog(db, {
         id: logId, api_type: apiType, model, provider_id: provider.id,
         status_code: status, latency_ms: Date.now() - startTime,
@@ -11,11 +11,12 @@ export function insertSuccessLog(db, params) {
         upstream_response: JSON.stringify({ statusCode: status, headers: upHdrs, body: respBody }),
         is_retry: isRetry ? 1 : 0, is_failover: isFailover ? 1 : 0, original_request_id: originalRequestId,
         router_key_id: routerKeyId, original_model: originalModel,
+        session_id: sessionId,
     });
 }
 /** Log a request rejected before reaching upstream */
 export function insertRejectedLog(params) {
-    const { db, logId, apiType, model, statusCode, errorMessage, startTime, isStream, routerKeyId, originalBody, clientHeaders, providerId = null, isFailover = false, originalRequestId = null, originalModel = null } = params;
+    const { db, logId, apiType, model, statusCode, errorMessage, startTime, isStream, routerKeyId, originalBody, clientHeaders, providerId = null, isFailover = false, originalRequestId = null, originalModel = null, sessionId = null } = params;
     insertRequestLog(db, {
         id: logId,
         api_type: apiType,
@@ -31,5 +32,6 @@ export function insertRejectedLog(params) {
         original_request_id: originalRequestId,
         router_key_id: routerKeyId,
         original_model: originalModel,
+        session_id: sessionId,
     });
 }

package/dist/proxy/model-state.d.ts

@@ -4,6 +4,8 @@ export declare class ModelStateManager {
     private db;
     /** 单例注入 DB 实例，启动时调用一次 */
     init(db: Database.Database): void;
+    /** 清空所有内存缓存（导入配置后调用） */
+    clearAll(): void;
     /** 构造内存 Map 的 key：有 sessionId 时用复合键 */
     buildKey(routerKeyId: string | null, sessionId?: string): string;
     /**

package/dist/proxy/model-state.js

@@ -10,6 +10,10 @@ export class ModelStateManager {
     init(db) {
         this.db = db;
     }
+    /** 清空所有内存缓存（导入配置后调用） */
+    clearAll() {
+        this.store.clear();
+    }
     /** 构造内存 Map 的 key：有 sessionId 时用复合键 */
     buildKey(routerKeyId, sessionId) {
         if (sessionId) {

package/dist/proxy/orchestrator.d.ts

@@ -23,6 +23,8 @@ export interface OrchestratorConfig {
     trackerId?: string;
     /** Claude Code 的 session ID，从 x-claude-code-session-id 请求头获取 */
     sessionId?: string;
+    /** 客户端请求的 JSON 字符串（headers + body），用于 Monitor 实时查看 */
+    clientRequest?: string;
 }
 export interface HandleContext {
     streamTimeoutMs?: number;

package/dist/proxy/orchestrator.js

@@ -50,6 +50,7 @@ export class ProxyOrchestrator {
             attempts: [],
             clientIp: request.ip,
             sessionId: config.sessionId,
+            clientRequest: config.clientRequest,
         };
     }
     createAbortSignal(request) {

package/dist/proxy/proxy-handler.js

@@ -14,15 +14,49 @@ import { updateLogStreamContent } from "../db/index.js";
 import { callNonStream, callStream } from "./transport.js";
 import { insertRejectedLog } from "./log-helpers.js";
 const HTTP_ERROR_THRESHOLD = 400;
-const STREAM_CONTENT_MAX_RAW = 8192;
-const STREAM_CONTENT_MAX_TEXT = 4096;
+const STREAM_CONTENT_MAX_RAW = 131072;
+const STREAM_CONTENT_MAX_TEXT = 65536;
+/** 将 tracker blocks 序列化为前端 tryDirectParse 可解析的 JSON */
+function serializeBlocksForStorage(blocks, apiType) {
+    if (!blocks || blocks.length === 0)
+        return "";
+    if (apiType === "anthropic") {
+        const content = blocks.map(b => {
+            if (b.type === "thinking")
+                return { type: "thinking", thinking: b.content };
+            if (b.type === "tool_use") {
+                let input = {};
+                try {
+                    input = JSON.parse(b.content || "{}");
+                }
+                catch { /* eslint-disable-line taste/no-silent-catch -- tool_use content 非合法 JSON 时保留空对象 */ }
+                return { type: "tool_use", name: b.name ?? "", input };
+            }
+            return { type: "text", text: b.content };
+        });
+        return JSON.stringify({ content });
+    }
+    const text = blocks.filter(b => b.type === "text").map(b => b.content).join("");
+    return JSON.stringify({ choices: [{ message: { content: text } }] });
+}
+function toStreamMetrics(m) {
+    return {
+        inputTokens: m.input_tokens,
+        outputTokens: m.output_tokens,
+        cacheReadTokens: m.cache_read_tokens,
+        ttftMs: m.ttft_ms,
+        tokensPerSecond: m.tokens_per_second,
+        stopReason: m.stop_reason,
+        isComplete: m.is_complete === 1,
+    };
+}
 export async function handleProxyRequest(request, reply, apiType, upstreamPath, errors, deps, options) {
     request.raw.socket.on("error", (err) => request.log.debug({ err }, "client socket error"));
     const clientModel = request.body.model || "unknown";
     const sessionId = request.headers["x-claude-code-session-id"];
     const { effectiveModel, originalModel, interceptResponse } = applyEnhancement(deps.db, request, clientModel, sessionId);
     if (interceptResponse)
-        return handleIntercept(deps.db, apiType, request, reply, interceptResponse, clientModel);
+        return handleIntercept(deps.db, apiType, request, reply, interceptResponse, clientModel, sessionId);
     const group = getMappingGroup(deps.db, effectiveModel);
     const isFailover = group?.strategy === "failover";
     const excludeTargets = [];
@@ -47,7 +81,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                     db: deps.db, logId, apiType, model: effectiveModel, statusCode: e.statusCode,
                     errorMessage: `All failover targets exhausted (${excludeTargets.length} attempted)`,
                     startTime, isStream, routerKeyId, originalBody, clientHeaders: cliHdrs, originalModel,
-                    isFailover: true, originalRequestId: rootLogId,
+                    isFailover: true, originalRequestId: rootLogId, sessionId,
                 });
                 return reply.status(e.statusCode).send(e.body);
             }
@@ -56,7 +90,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                 db: deps.db, logId, apiType, model: effectiveModel, statusCode: e.statusCode,
                 errorMessage: `No mapping found for model '${effectiveModel}'`, startTime, isStream,
                 routerKeyId, originalBody, clientHeaders: cliHdrs, originalModel,
-                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
             });
             return reply.status(e.statusCode).send(e.body);
         }
@@ -71,7 +105,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                             db: deps.db, logId, apiType, model: effectiveModel, statusCode: e.statusCode,
                             errorMessage: `Model '${resolved.backend_model}' not allowed`, startTime, isStream, routerKeyId,
                             originalBody, clientHeaders: cliHdrs, providerId: resolved.provider_id, originalModel,
-                            isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                            isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
                         });
                         return reply.status(e.statusCode).send(e.body);
                     }
@@ -88,7 +122,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                 db: deps.db, logId, apiType, model: effectiveModel, statusCode: e.statusCode,
                 errorMessage: `Provider '${resolved.provider_id}' unavailable`, startTime, isStream, routerKeyId,
                 originalBody, clientHeaders: cliHdrs, providerId: resolved.provider_id, originalModel,
-                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
             });
             return reply.status(e.statusCode).send(e.body);
         }
@@ -98,7 +132,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                 db: deps.db, logId, apiType, model: effectiveModel, statusCode: e.statusCode,
                 errorMessage: `API type mismatch: expected '${apiType}'`, startTime, isStream, routerKeyId,
                 originalBody, clientHeaders: cliHdrs, providerId: resolved.provider_id, originalModel,
-                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
             });
             return reply.status(e.statusCode).send(e.body);
         }
@@ -119,14 +153,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
             if (isStream) {
                 const metricsTransform = new SSEMetricsTransform(apiType, startTime, {
                     onMetrics: (m) => {
-                        deps.tracker?.update(logId, {
-                            streamMetrics: {
-                                inputTokens: m.input_tokens, outputTokens: m.output_tokens,
-                                cacheReadTokens: m.cache_read_tokens,
-                                ttftMs: m.ttft_ms, tokensPerSecond: m.tokens_per_second,
-                                stopReason: m.stop_reason, isComplete: m.is_complete === 1,
-                            },
-                        });
+                        deps.tracker?.update(logId, { streamMetrics: toStreamMetrics(m) });
                     },
                     onChunk: (rawLine) => {
                         deps.tracker?.appendStreamChunk(logId, rawLine, apiType, STREAM_CONTENT_MAX_RAW, STREAM_CONTENT_MAX_TEXT);
@@ -135,21 +162,19 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                 const checkEarlyError = deps.matcher
                     ? (data) => deps.matcher.test(UPSTREAM_SUCCESS, data)
                     : undefined;
-                return callStream(provider, apiKey, body, cliHdrs, reply, deps.streamTimeoutMs, upstreamPath, buildUpstreamHeaders, metricsTransform, checkEarlyError);
+                const streamResult = await callStream(provider, apiKey, body, cliHdrs, reply, deps.streamTimeoutMs, upstreamPath, buildUpstreamHeaders, metricsTransform, checkEarlyError);
+                const m = (streamResult.kind === "stream_success" || streamResult.kind === "stream_abort")
+                    ? streamResult.metrics : undefined;
+                if (m)
+                    deps.tracker?.update(logId, { streamMetrics: toStreamMetrics(m) });
+                return streamResult;
             }
             const result = await callNonStream(provider, apiKey, body, cliHdrs, upstreamPath, buildUpstreamHeaders);
             // 非流式请求：从响应体提取指标并更新 tracker
             if (result.kind === "success") {
                 const mr = MetricsExtractor.fromNonStreamResponse(apiType, result.body);
                 if (mr) {
-                    deps.tracker?.update(logId, {
-                        streamMetrics: {
-                            inputTokens: mr.input_tokens, outputTokens: mr.output_tokens,
-                            cacheReadTokens: mr.cache_read_tokens,
-                            ttftMs: mr.ttft_ms, tokensPerSecond: mr.tokens_per_second,
-                            stopReason: mr.stop_reason, isComplete: mr.is_complete === 1,
-                        },
-                    });
+                    deps.tracker?.update(logId, { streamMetrics: toStreamMetrics(mr) });
                 }
             }
             // 非流式响应注入模型信息标签（模型映射场景）
@@ -168,19 +193,25 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
             return result;
         };
         try {
-            const resilienceResult = await deps.orchestrator.handle(request, reply, apiType, { resolved, provider, clientModel: effectiveModel, isStream, trackerId: logId, sessionId }, { retryMaxAttempts: deps.retryMaxAttempts, retryBaseDelayMs: deps.retryBaseDelayMs, isFailover, ruleMatcher: deps.matcher, transportFn });
+            const resilienceResult = await deps.orchestrator.handle(request, reply, apiType, { resolved, provider, clientModel: effectiveModel, isStream, trackerId: logId, sessionId, clientRequest: clientReq }, { retryMaxAttempts: deps.retryMaxAttempts, retryBaseDelayMs: deps.retryBaseDelayMs, isFailover, ruleMatcher: deps.matcher, transportFn });
             const lastLogId = logResilienceResult(deps.db, {
                 apiType, model: effectiveModel, providerId: provider.id, isStream,
-                clientReq, upstreamReqBase, logId, routerKeyId, originalModel,
+                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);
-            // 流式请求：将 tracker 中累积的文本内容持久化到日志
+            // 流式请求：将 tracker 中累积的内容持久化到日志
+            // blocks 含非 text 类型时（thinking/tool_use）必须序列化为 JSON 以保留结构
+            // 注意：tracker 在原 logId 下累积内容，lastLogId 可能因 resilience 重试而指向不同记录
             if (isStream && deps.tracker) {
-                const req = deps.tracker.get(lastLogId);
-                const text = req?.streamContent?.textContent;
-                if (text)
-                    updateLogStreamContent(deps.db, lastLogId, text);
+                const sc = deps.tracker.get(logId)?.streamContent;
+                const blocks = sc?.blocks;
+                const hasStructured = blocks && blocks.length > 0 && blocks.some(b => b.type !== "text");
+                const content = hasStructured
+                    ? serializeBlocksForStorage(blocks, apiType)
+                    : (sc?.textContent || "");
+                if (content)
+                    updateLogStreamContent(deps.db, lastLogId, content);
             }
             // Failover: 单 provider 内重试已耗尽但仍失败，尝试下一个 target
             if (isFailover && !reply.raw.headersSent) {
@@ -216,7 +247,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                     errorMessage: `Concurrency queue full for provider '${provider.id}'`,
                     startTime, isStream, routerKeyId, originalBody, clientHeaders: cliHdrs,
                     providerId: provider.id, originalModel,
-                    isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                    isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
                 });
                 return reply.status(err.statusCode).send(err.body);
             }
@@ -227,7 +258,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                     errorMessage: `Concurrency wait timeout for provider '${provider.id}' (${e.timeoutMs}ms)`,
                     startTime, isStream, routerKeyId, originalBody, clientHeaders: cliHdrs,
                     providerId: provider.id, originalModel,
-                    isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null,
+                    isFailover: isFailoverIteration, originalRequestId: isFailoverIteration ? rootLogId : null, sessionId,
                 });
                 return reply.status(err.statusCode).send(err.body);
             }
@@ -240,6 +271,7 @@ export async function handleProxyRequest(request, reply, apiType, upstreamPath,
                 client_request: clientReq, upstream_request: upstreamReqBase,
                 is_failover: isFailoverIteration ? 1 : 0, original_request_id: isFailoverIteration ? rootLogId : null,
                 router_key_id: routerKeyId, original_model: originalModel,
+                session_id: sessionId,
             });
             const err = errors.upstreamConnectionFailed();
             return reply.status(err.statusCode).send(err.body);

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

@@ -11,7 +11,7 @@ export declare function handleIntercept(db: Database.Database, apiType: "openai"
     statusCode: number;
     body: unknown;
     meta?: unknown;
-}, clientModel: string): import("fastify").FastifyReply;
+}, clientModel: string, sessionId?: string): import("fastify").FastifyReply;
 export declare function logResilienceResult(db: Database.Database, params: {
     apiType: "openai" | "anthropic";
     model: string;
@@ -22,6 +22,7 @@ export declare function logResilienceResult(db: Database.Database, params: {
     logId: string;
     routerKeyId: string | null;
     originalModel: string | null;
+    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;