feat(usage): 改进请求追踪与缓存预热

- 区分请求重试与故障转移：新增 has_retry 标识亲和缓存重试
- 修正候选 TTFB 计算：记录候选自身的首字节时间而非全局时间
- Streaming 状态同步 rate_multiplier，支持按 API 格式配置
- 新增缓存预热服务：启动时预热仪表盘统计、热力图、每日统计
- 优化缓存 TTL 配置：仪表盘统计 2 分钟、热力图和每日统计 10 分钟
- 自动刷新间隔从 10 秒调整为 5 秒

src/api/handlers/base/base_handler.py

@@ -432,6 +432,7 @@ class BaseMessageHandler:
         endpoint_id = ctx.endpoint_id
         key_id = ctx.key_id
         first_byte_time_ms = ctx.first_byte_time_ms
+        api_format = ctx.api_format
 
         # 如果 provider 为空，记录警告（不应该发生，但用于调试）
         if not provider:
@@ -455,6 +456,7 @@ class BaseMessageHandler:
                         provider_endpoint_id=endpoint_id,
                         provider_api_key_id=key_id,
                         first_byte_time_ms=first_byte_time_ms,
+                        api_format=api_format,
                     )
                 finally:
                     db.close()

src/api/handlers/base/cli_handler_base.py

@@ -1432,6 +1432,16 @@ class CliMessageHandlerBase(BaseMessageHandler):
                 if ctx.attempt_id:
                     from src.services.request.candidate import RequestCandidateService
 
+                    # 计算候选自身的 TTFB
+                    candidate_first_byte_time_ms: Optional[int] = None
+                    if ctx.first_byte_time_ms is not None:
+                        candidate_first_byte_time_ms = RequestCandidateService.calculate_candidate_ttfb(
+                            db=bg_db,
+                            candidate_id=ctx.attempt_id,
+                            request_start_time=self.start_time,
+                            global_first_byte_time_ms=ctx.first_byte_time_ms,
+                        )
+
                     # 根据状态码决定是成功还是失败
                     # 499 = 客户端断开连接，应标记为失败
                     # 503 = 服务不可用（如流中断），应标记为失败
@@ -1443,8 +1453,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
                             "chunk_count": ctx.chunk_count,
                             "data_count": ctx.data_count,
                         }
-                        if ctx.first_byte_time_ms is not None:
-                            extra_data["first_byte_time_ms"] = ctx.first_byte_time_ms
+                        if candidate_first_byte_time_ms is not None:
+                            extra_data["first_byte_time_ms"] = candidate_first_byte_time_ms
                         RequestCandidateService.mark_candidate_failed(
                             db=bg_db,
                             candidate_id=ctx.attempt_id,
@@ -1462,8 +1472,8 @@ class CliMessageHandlerBase(BaseMessageHandler):
                             "chunk_count": ctx.chunk_count,
                             "data_count": ctx.data_count,
                         }
-                        if ctx.first_byte_time_ms is not None:
-                            extra_data["first_byte_time_ms"] = ctx.first_byte_time_ms
+                        if candidate_first_byte_time_ms is not None:
+                            extra_data["first_byte_time_ms"] = candidate_first_byte_time_ms
                         RequestCandidateService.mark_candidate_success(
                             db=bg_db,
                             candidate_id=ctx.attempt_id,

src/api/handlers/base/stream_telemetry.py

@@ -132,7 +132,7 @@ class StreamTelemetryRecorder:
                     )
 
                 # 更新候选记录状态
-                await self._update_candidate_status(bg_db, ctx, response_time_ms)
+                await self._update_candidate_status(bg_db, ctx, response_time_ms, start_time)
 
             finally:
                 if bg_db:
@@ -234,6 +234,7 @@ class StreamTelemetryRecorder:
         db: Session,
         ctx: StreamContext,
         response_time_ms: int,
+        request_start_time: float,
     ) -> None:
         """更新候选记录状态"""
         if not ctx.attempt_id:
@@ -246,7 +247,14 @@ class StreamTelemetryRecorder:
             "data_count": ctx.data_count,
         }
         if ctx.first_byte_time_ms is not None:
-            extra_data["first_byte_time_ms"] = ctx.first_byte_time_ms
+            # 计算候选自身的 TTFB
+            first_byte_time_ms = RequestCandidateService.calculate_candidate_ttfb(
+                db=db,
+                candidate_id=ctx.attempt_id,
+                request_start_time=request_start_time,
+                global_first_byte_time_ms=ctx.first_byte_time_ms,
+            )
+            extra_data["first_byte_time_ms"] = first_byte_time_ms
 
         if ctx.is_success():
             RequestCandidateService.mark_candidate_success(