fix(prod): 后端生产级 A 类硬伤全清（7 项：授权/崩溃点/全表扫/限流/鉴权/限额）

部署前生产级审计（可靠性/数据层/安全三路）后，清掉 7 处代码级硬伤： A1 后台定时器 goroutine 无 panic recover → 单个 DB panic 崩整个 gateway。加 safeGo/ safeCall，包住订阅/掉单补偿/微信推送/探针 goroutine，单轮 tick 再兜一层。 A2 提示词控制面(建/激活/停用，热广播全服务)只 RequireAuth → 任意登录用户改全局提示词。三写端点+列表挂 RequireAdmin。 A3 HITL 审批端点无角色门 → viewer 可放行烧钱执行。加 RequireTenantRole(member)。 A4 审计/护栏列表 limit 无校验，limit=-1 让 gorm 取消 LIMIT 全表扫。加 clampLimit/ clampOffset，AdminTasks/AdminSpaces 补上界。 A5 限流 Redis 一挂就完全放行(fail-open)。加进程内固定窗口兜底(fail-safe) + 登录/注册按 IP 专用严限流(10/min)。 A6 公开 by-id 端点(stream/exec/report导出/kb导入流)无鉴权无租户过滤。加 AuthFromHeaderOrQuery(从 ?token= 取 JWT) + task/report 按 owner 归属校验；桌面端 5 处 EventSource/下载 URL 经 tokenQuery 附 JWT。 A7 文件上传无大小上限(整文件进内存 OOM 面) → 50MB 闸(KB_MAX_UPLOAD_BYTES)+ LimitReader； http.Server 加 ReadHeaderTimeout/ReadTimeout/MaxHeaderBytes(不设 WriteTimeout 保 SSE)。带单测：clampLimit/safeCall/procLimiter/AuthFromHeaderOrQuery/TaskOwner。 build+vet+全量 test 绿；desktop tsc 绿。B(迁移工具/实时探针/出网韧性/登录锁定/leader选举) 与 C(TLS/PG HA/K8s/备份自动化/可观测)分期后做，参照 production_readiness.md。 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-07-21 15:04:47 +08:00
parent ce7cca657e
commit ac38d5e663
20 changed files with 443 additions and 100 deletions
@@ -75,58 +75,68 @@ func (h *Handler) AdminStatus(c *gin.Context) {
 	// 1) mcp-go health → milvus / neo4j 基建灯
 	go func() {
 		defer wg.Done()
-		ctx, cancel := context.WithTimeout(parent, probeTimeout)
-		defer cancel()
-		if res, err := h.bus.CallTool(ctx, contract.ToolSubjectGo("health"),
-			&contract.ToolCall{Tool: "health"}); err == nil && res != nil && res.OK {
-			var sub map[string]bool
-			if json.Unmarshal([]byte(res.Content), &sub) == nil {
-				milvus, neo4j = sub["milvus"], sub["neo4j"]
-				ftDisk = sub["fulltext_disk"]
+		safeCall("status-probe-mcpgo-health", func() {
+			ctx, cancel := context.WithTimeout(parent, probeTimeout)
+			defer cancel()
+			if res, err := h.bus.CallTool(ctx, contract.ToolSubjectGo("health"),
+				&contract.ToolCall{Tool: "health"}); err == nil && res != nil && res.OK {
+				var sub map[string]bool
+				if json.Unmarshal([]byte(res.Content), &sub) == nil {
+					milvus, neo4j = sub["milvus"], sub["neo4j"]
+					ftDisk = sub["fulltext_disk"]
+				}
 			}
-		}
+		})
 	}()

 	// 2) mcp-go list_tools → 在线判定 + 工具清单
 	go func() {
 		defer wg.Done()
-		goUp, goTools, goLatency = h.probeTools(parent, contract.ToolSubjectGo("list_tools"))
+		safeCall("status-probe-mcpgo-tools", func() {
+			goUp, goTools, goLatency = h.probeTools(parent, contract.ToolSubjectGo("list_tools"))
+		})
 	}()

 	// 3) mcp-py list_tools → 在线判定 + 工具清单
 	go func() {
 		defer wg.Done()
-		pyUp, pyTools, pyLatency = h.probeTools(parent, contract.ToolSubjectPy("list_tools"))
+		safeCall("status-probe-mcppy-tools", func() {
+			pyUp, pyTools, pyLatency = h.probeTools(parent, contract.ToolSubjectPy("list_tools"))
+		})
 	}()

 	// 4) dispatcher 心跳 → 在线判定 + 模型/运行时长
 	go func() {
 		defer wg.Done()
-		ctx, cancel := context.WithTimeout(parent, probeTimeout)
-		defer cancel()
-		start := time.Now()
-		if data, err := h.bus.Ping(ctx, contract.SubjectHealthDispatcher); err == nil {
-			dispUp = true
-			dispLatency = int(time.Since(start).Milliseconds())
-			var st struct {
-				Model   string `json:"model"`
-				Ready   bool   `json:"ready"`
-				UptimeS int    `json:"uptime_s"`
+		safeCall("status-probe-dispatcher", func() {
+			ctx, cancel := context.WithTimeout(parent, probeTimeout)
+			defer cancel()
+			start := time.Now()
+			if data, err := h.bus.Ping(ctx, contract.SubjectHealthDispatcher); err == nil {
+				dispUp = true
+				dispLatency = int(time.Since(start).Milliseconds())
+				var st struct {
+					Model   string `json:"model"`
+					Ready   bool   `json:"ready"`
+					UptimeS int    `json:"uptime_s"`
+				}
+				if json.Unmarshal(data, &st) == nil {
+					dispDetail = dispatcherDetail(st.Model, st.Ready, st.UptimeS)
+				}
 			}
-			if json.Unmarshal(data, &st) == nil {
-				dispDetail = dispatcherDetail(st.Model, st.Ready, st.UptimeS)
-			}
-		}
+		})
 	}()

 	// 5) 基建活性探针：PG / Redis / MinIO 实时 ping（非仅启动标志，能反映中途掉线）。
 	go func() {
 		defer wg.Done()
-		ctx, cancel := context.WithTimeout(parent, probeTimeout)
-		defer cancel()
-		pgUp = h.db.Ping(ctx)
-		redisUp = h.cache.Ping(ctx)
-		minioUp = h.blob != nil && h.blob.Ping(ctx)
+		safeCall("status-probe-infra", func() {
+			ctx, cancel := context.WithTimeout(parent, probeTimeout)
+			defer cancel()
+			pgUp = h.db.Ping(ctx)
+			redisUp = h.cache.Ping(ctx)
+			minioUp = h.blob != nil && h.blob.Ping(ctx)
+		})
 	}()

 	wg.Wait()