fix(agent): make discovery enrollment compatible and idempotent (#53)

* fix(agent): make discovery enrollment compatible and idempotent

* fix(agent): confirm enrollment persistence and document token aliases

docs/changelog/index.md

@@ -19,6 +19,8 @@ sidebar: false
 
 ### 🛠 修复
 
+- 修复 Agent 自动注册响应凭证字段不一致导致节点不断重复创建的问题；新增持久化注册重试凭证与退避，响应丢失、并发重试或配置写入失败后仍复用原节点，且不会覆盖管理员修改或恢复已更换的节点凭证。支持目录刷盘的平台会确认原子替换后的目录持久化，失败时阻止使用未确认的注册凭证；接口文档同步列出两种兼容凭证字段。
+
 - SQLite 文件数据库启动时自动启用 WAL 模式，改善读写并发；数据库导出改用一致性快照，确保包含 WAL 中已提交的数据。
 
 - 修复 IP 组按规则自动抓取时会收录内网、回环及链路本地等地址的问题；规则测试同步过滤这些地址，历史抓取结果会在下次同步时清理。

docs/design/agent-design.md

@@ -37,10 +37,13 @@ Agent 主要由以下核心子模块组成，共同配合完成其完整的生
 Agent 在生命周期中主要通过 **基于 Token 的自动注册** 和 **心跳/WebSocket 双通道** 与控制面通信。
 
 ### 1. 自动注册流程
-若 Agent 启动时本地 `agent.json` 的 `access_token` 为空，但配置了 `discovery_token`，将触发自动注册流程：
+若 Agent 启动时本地 `agent.json` 的 `agent_token` 为空，但配置了 `discovery_token`，将触发自动注册流程：
-1. Agent 向控制面 `/api/v1/agent/nodes/register` 发送注册请求，携带本地硬件摘要、IP 及主机名。
+1. Agent 在本地状态文件原子保存随机 256 位 `registration_key`，再向控制面 `/api/v1/agent/nodes/register` 发送注册请求，携带该重试凭证、本地硬件摘要、IP 及主机名。支持目录刷盘的平台会在文件刷盘及重命名后刷新所在目录；目录刷盘失败会返回错误，已有凭证也必须在重试时确认目录刷盘成功后才能发送。Windows 保留文件刷盘与替换，但目录持久化仅为尽力而为；上述行为针对所在目录已存在的文件更新，不保证所有文件系统或突然断电场景下的完整持久性。该凭证仅用于注册，不进入心跳、WebSocket 或观测数据，必须与节点 Token 一样保密。
-2. Server 校验 `discovery_token` 有效后，在数据库生成唯一的 `NodeID` 与专属 `AccessToken`（即 `agent_token`）并返回。
+2. Server 校验当前 `discovery_token` 有效后，从重试凭证使用不同的哈希域分别生成 `NodeID` 与专属 Token。相同凭证的重复或并发请求只保留一个节点，不按 IP、主机名或公开 `node_id` 合并，也不会覆盖管理员修改的名称、升级标志等数据。响应以 `agent_token` 为准，同时保留旧的 `access_token` 字段；新版 Agent 能识别两种字段，值冲突时拒绝响应。
-3. Agent 将获取的专用 Token 写入本地配置文件，擦除一次性 `discovery_token`，后续所有的通信均基于专属 `AccessToken` 进行鉴权认证。
+3. Agent 原子保存服务端分配的节点 ID 和配置 Token，全部写入成功后才切换运行时鉴权并清空配置中的 `discovery_token`，后续通信均使用节点专属 Token。响应丢失或配置写入失败时，重启仍沿用原重试凭证；状态文件中的重试凭证与节点 ID 不会被旧的后台任务快照覆盖。
+4. 注册失败按 3、6、12、24、48、60 秒退避，之后每次至少间隔 60 秒重试（实际请求在下一个心跳定时器周期触发），不再随每次心跳立即注册。成功后重置退避。
+
+若节点 Token 已被清空或更换，旧重试凭证不能取回新 Token 或恢复旧 Token，必须使用管理员分配的节点凭证。物理删除节点不构成永久封禁：持有当前有效 discovery Token 的客户端仍能重新注册。旧版 Agent 不携带 `registration_key`，修复后的响应可使其正常完成注册，但其响应丢失后的重试仍不具备幂等保证。请保留每个实例独立的持久化状态与配置，不要把已初始化的状态目录复制给其他实例。
 
 ### 2. 双通道心跳与同步机制
 * **HTTP 轮询通道（兜底与探测）**：Agent 默认按设定的 `heartbeat_interval` 间隔发送 POST 心跳包。上报指标的同时获取当前激活版本的摘要信息（Version & Checksum）。

docs/docs.go

@@ -4443,12 +4443,12 @@ const docTemplate = `{
                 "summary": "注册或发现 Agent 节点",
                 "parameters": [
                     {
-                        "description": "节点上报数据",
+                        "description": "节点上报数据；registration_key 为首次注册重试凭证",
                         "name": "body",
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/agent.NodePayload"
+                            "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest"
                         }
                     }
                 ],
@@ -4464,7 +4464,7 @@ const docTemplate = `{
                                     "type": "object",
                                     "properties": {
                                         "data": {
-                                            "$ref": "#/definitions/agent.RegistrationResponse"
+                                            "$ref": "#/definitions/agent.RegistrationResponseDocs"
                                         }
                                     }
                                 }
@@ -14989,12 +14989,15 @@ const docTemplate = `{
                 }
             }
         },
-        "agent.RegistrationResponse": {
+        "agent.RegistrationResponseDocs": {
             "type": "object",
             "properties": {
                 "access_token": {
                     "type": "string"
                 },
+                "agent_token": {
+                    "type": "string"
+                },
                 "name": {
                     "type": "string"
                 },
@@ -16293,6 +16296,77 @@ const docTemplate = `{
                 }
             }
         },
+        "github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest": {
+            "type": "object",
+            "properties": {
+                "access_logs": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeAccessLog"
+                    }
+                },
+                "buffered": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.BufferedObservabilityRecord"
+                    }
+                },
+                "current_version": {
+                    "type": "string"
+                },
+                "edge_health": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeEdgeHealth"
+                },
+                "ext_version": {
+                    "type": "string"
+                },
+                "health_events": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeHealthEvent"
+                    }
+                },
+                "host_metrics": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeMetricSnapshot"
+                },
+                "ip": {
+                    "type": "string"
+                },
+                "last_error": {
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "node_id": {
+                    "type": "string"
+                },
+                "openresty_message": {
+                    "type": "string"
+                },
+                "openresty_status": {
+                    "type": "string"
+                },
+                "profile": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeSystemProfile"
+                },
+                "registration_key": {
+                    "type": "string"
+                },
+                "schema_version": {
+                    "type": "integer"
+                },
+                "version": {
+                    "type": "string"
+                },
+                "waf_ip_group_checksums": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
         "github_com_Rain-kl_Wavelet_pkg_protocol.WAFIPGroup": {
             "type": "object",
             "properties": {

docs/swagger.json

@@ -4436,12 +4436,12 @@
                 "summary": "注册或发现 Agent 节点",
                 "parameters": [
                     {
-                        "description": "节点上报数据",
+                        "description": "节点上报数据；registration_key 为首次注册重试凭证",
                         "name": "body",
                         "in": "body",
                         "required": true,
                         "schema": {
-                            "$ref": "#/definitions/agent.NodePayload"
+                            "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest"
                         }
                     }
                 ],
@@ -4457,7 +4457,7 @@
                                     "type": "object",
                                     "properties": {
                                         "data": {
-                                            "$ref": "#/definitions/agent.RegistrationResponse"
+                                            "$ref": "#/definitions/agent.RegistrationResponseDocs"
                                         }
                                     }
                                 }
@@ -14982,12 +14982,15 @@
                 }
             }
         },
-        "agent.RegistrationResponse": {
+        "agent.RegistrationResponseDocs": {
             "type": "object",
             "properties": {
                 "access_token": {
                     "type": "string"
                 },
+                "agent_token": {
+                    "type": "string"
+                },
                 "name": {
                     "type": "string"
                 },
@@ -16286,6 +16289,77 @@
                 }
             }
         },
+        "github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest": {
+            "type": "object",
+            "properties": {
+                "access_logs": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeAccessLog"
+                    }
+                },
+                "buffered": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.BufferedObservabilityRecord"
+                    }
+                },
+                "current_version": {
+                    "type": "string"
+                },
+                "edge_health": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeEdgeHealth"
+                },
+                "ext_version": {
+                    "type": "string"
+                },
+                "health_events": {
+                    "type": "array",
+                    "items": {
+                        "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeHealthEvent"
+                    }
+                },
+                "host_metrics": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeMetricSnapshot"
+                },
+                "ip": {
+                    "type": "string"
+                },
+                "last_error": {
+                    "type": "string"
+                },
+                "name": {
+                    "type": "string"
+                },
+                "node_id": {
+                    "type": "string"
+                },
+                "openresty_message": {
+                    "type": "string"
+                },
+                "openresty_status": {
+                    "type": "string"
+                },
+                "profile": {
+                    "$ref": "#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeSystemProfile"
+                },
+                "registration_key": {
+                    "type": "string"
+                },
+                "schema_version": {
+                    "type": "integer"
+                },
+                "version": {
+                    "type": "string"
+                },
+                "waf_ip_group_checksums": {
+                    "type": "object",
+                    "additionalProperties": {
+                        "type": "string"
+                    }
+                }
+            }
+        },
         "github_com_Rain-kl_Wavelet_pkg_protocol.WAFIPGroup": {
             "type": "object",
             "properties": {

docs/swagger.yaml

@@ -99,10 +99,12 @@ definitions:
           type: string
         type: object
     type: object
-  agent.RegistrationResponse:
+  agent.RegistrationResponseDocs:
     properties:
       access_token:
         type: string
+      agent_token:
+        type: string
       name:
         type: string
       node_id:
@@ -955,6 +957,53 @@ definitions:
       total_size:
         type: integer
     type: object
+  github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest:
+    properties:
+      access_logs:
+        items:
+          $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeAccessLog'
+        type: array
+      buffered:
+        items:
+          $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.BufferedObservabilityRecord'
+        type: array
+      current_version:
+        type: string
+      edge_health:
+        $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeEdgeHealth'
+      ext_version:
+        type: string
+      health_events:
+        items:
+          $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeHealthEvent'
+        type: array
+      host_metrics:
+        $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeMetricSnapshot'
+      ip:
+        type: string
+      last_error:
+        type: string
+      name:
+        type: string
+      node_id:
+        type: string
+      openresty_message:
+        type: string
+      openresty_status:
+        type: string
+      profile:
+        $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.NodeSystemProfile'
+      registration_key:
+        type: string
+      schema_version:
+        type: integer
+      version:
+        type: string
+      waf_ip_group_checksums:
+        additionalProperties:
+          type: string
+        type: object
+    type: object
   github_com_Rain-kl_Wavelet_pkg_protocol.WAFIPGroup:
     properties:
       checksum:
@@ -7020,12 +7069,12 @@ paths:
       - application/json
       description: 使用节点 access token 重新注册，或使用全局 discovery token 发现新节点；请求头需携带 X-Agent-Token
       parameters:
-      - description: 节点上报数据
+      - description: 节点上报数据；registration_key 为首次注册重试凭证
         in: body
         name: body
         required: true
         schema:
-          $ref: '#/definitions/agent.NodePayload'
+          $ref: '#/definitions/github_com_Rain-kl_Wavelet_pkg_protocol.RegisterNodeRequest'
       produces:
       - application/json
       responses:
@@ -7036,7 +7085,7 @@ paths:
             - $ref: '#/definitions/response.Any'
             - properties:
                 data:
-                  $ref: '#/definitions/agent.RegistrationResponse'
+                  $ref: '#/definitions/agent.RegistrationResponseDocs'
               type: object
         "400":
           description: 参数错误

internal/apps/agent/agent/discovery_integration_test.go

@@ -0,0 +1,223 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/config"
+	agentheartbeat "github.com/Rain-kl/Wavelet/internal/apps/agent/heartbeat"
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/httpclient"
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/protocol"
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/state"
+	serveragent "github.com/Rain-kl/Wavelet/internal/apps/openflare/agent"
+	db "github.com/Rain-kl/Wavelet/internal/infra/persistence"
+	"github.com/Rain-kl/Wavelet/internal/model"
+	"github.com/Rain-kl/Wavelet/internal/repository"
+	"github.com/Rain-kl/Wavelet/internal/shared/response"
+	"github.com/Rain-kl/Wavelet/internal/testhelper"
+	"github.com/gin-gonic/gin"
+	"github.com/glebarez/sqlite"
+	"github.com/pressly/goose/v3"
+	"gorm.io/gorm"
+)
+
+func newDiscoveryTestRunner(t *testing.T, configPath string) (*Runner, string) {
+	t.Helper()
+	cfg, err := config.Load(configPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	store := state.NewStore(cfg.StatePath)
+	nodeID, err := store.EnsureNodeID()
+	if err != nil {
+		t.Fatal(err)
+	}
+	client := httpclient.New(cfg.ServerURL, cfg.InitialAuthToken(), time.Second)
+	runner := withHeartbeatCycle(&Runner{Config: cfg, StateStore: store,
+		HeartbeatService: agentheartbeat.New(client), SyncService: &fakeSyncService{}}, nil)
+	return runner, nodeID
+}
+
+func setupDiscoveryIntegrationDB(t *testing.T) *gorm.DB {
+	t.Helper()
+	ctx := context.Background()
+	conn, err := gorm.Open(sqlite.Open(filepath.Join(t.TempDir(), "discovery.db")), &gorm.Config{DisableForeignKeyConstraintWhenMigrating: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	sqlDB, err := conn.DB()
+	if err != nil {
+		t.Fatal(err)
+	}
+	previous := db.DB(ctx)
+	t.Cleanup(func() {
+		if invalidateErr := repository.InvalidateAllSystemConfigCaches(ctx); invalidateErr != nil {
+			t.Error(invalidateErr)
+		}
+		db.SetDB(previous)
+		if closeErr := sqlDB.Close(); closeErr != nil {
+			t.Error(closeErr)
+		}
+	})
+	provider, err := goose.NewProvider(goose.DialectSQLite3, sqlDB, os.DirFS("../../../infra/persistence/migrator/goose/sqlite"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err = provider.Up(ctx); err != nil {
+		t.Fatal(err)
+	}
+	db.SetDB(conn)
+	return conn
+}
+
+func TestDiscoveryHTTPRecoveryAndRestart(t *testing.T) {
+	ctx := context.Background()
+	conn := setupDiscoveryIntegrationDB(t)
+	var err error
+	if err = repository.SaveOrUpdateSystemConfig(ctx, model.ConfigKeyAgentDiscoveryToken, "discovery-test"); err != nil {
+		t.Fatal(err)
+	}
+	if err = repository.SaveOrUpdateSystemConfig(ctx, model.ConfigKeyGeoIPProvider, "disabled"); err != nil {
+		t.Fatal(err)
+	}
+	var registrations, heartbeats atomic.Int32
+	var dropResponse atomic.Bool
+	dropResponse.Store(true)
+	router := testhelper.NewTestGinEngine()
+	router.POST("/api/v1/agent/nodes/register", serveragent.RegisterAuth(), func(c *gin.Context) {
+		registrations.Add(1)
+		serveragent.RegisterHandler(c)
+	})
+	router.POST("/api/v1/agent/nodes/heartbeat", serveragent.Auth(), func(c *gin.Context) {
+		var payload map[string]json.RawMessage
+		if decodeErr := c.ShouldBindJSON(&payload); decodeErr != nil {
+			t.Error(decodeErr)
+		}
+		if _, exists := payload["registration_key"]; exists {
+			t.Error("heartbeat exposed registration_key")
+		}
+		heartbeats.Add(1)
+		c.JSON(http.StatusOK, response.OKNil())
+	})
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/api/v1/agent/nodes/register" && dropResponse.Swap(false) {
+			recorder := httptest.NewRecorder()
+			router.ServeHTTP(recorder, r)
+			if recorder.Code != http.StatusOK {
+				t.Errorf("First registration status = %d, want 200", recorder.Code)
+			}
+			w.WriteHeader(http.StatusBadGateway) // Enrollment committed, but its response was lost.
+			return
+		}
+		router.ServeHTTP(w, r)
+	}))
+	t.Cleanup(server.Close)
+	configPath := filepath.Join(t.TempDir(), "agent.json")
+	data, err := json.Marshal(map[string]string{"server_url": server.URL, "discovery_token": "discovery-test", "node_ip": "10.0.0.8", "node_name": "edge"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err = os.WriteFile(configPath, data, 0o600); err != nil {
+		t.Fatal(err)
+	}
+
+	first, nodeID := newDiscoveryTestRunner(t, configPath)
+	if err = first.tryRegister(ctx, &nodeID); err == nil {
+		t.Fatal("tryRegister(lost response) succeeded, want transport failure")
+	}
+	key, err := first.StateStore.EnsureRegistrationKey()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if first.hasAccessToken() {
+		t.Fatal("tryRegister(lost response) activated token")
+	}
+
+	// Restart with the same private key, then fail the second durable identity write.
+	second, secondID := newDiscoveryTestRunner(t, configPath)
+	backup := configPath + ".original"
+	if err = os.Rename(configPath, backup); err != nil {
+		t.Fatal(err)
+	}
+	if err = os.Mkdir(configPath, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err = second.tryRegister(ctx, &secondID); err == nil {
+		t.Fatal("tryRegister(config save failure) succeeded, want write error")
+	}
+	if second.hasAccessToken() || second.Config.DiscoveryToken == "" {
+		t.Fatal("Config write failure activated partial in-memory credentials")
+	}
+	if err = os.Remove(configPath); err != nil {
+		t.Fatal(err)
+	}
+	if err = os.Rename(backup, configPath); err != nil {
+		t.Fatal(err)
+	}
+
+	third, thirdID := newDiscoveryTestRunner(t, configPath)
+	if err = third.tryRegister(ctx, &thirdID); err != nil {
+		t.Fatalf("tryRegister(after restart) error = %v", err)
+	}
+	if !third.hasAccessToken() || third.Config.DiscoveryToken != "" {
+		t.Fatal("Successful enrollment did not activate node credentials")
+	}
+	persistedKey, err := third.StateStore.EnsureRegistrationKey()
+	if err != nil || persistedKey != key {
+		t.Errorf("Retry key changed across partial-write restarts: %v", err)
+	}
+
+	restarted, restartedID := newDiscoveryTestRunner(t, configPath)
+	restarted.runStartupAuth(ctx, &restartedID)
+	if restartedID != thirdID {
+		t.Errorf("Restarted node ID = %q, want %q", restartedID, thirdID)
+	}
+	if registrations.Load() != 3 || heartbeats.Load() != 2 {
+		t.Errorf("Enrollment/restart calls = %d registrations, %d heartbeats; want 3, 2", registrations.Load(), heartbeats.Load())
+	}
+	var count int64
+	if err = conn.Model(&model.OpenFlareNode{}).Count(&count).Error; err != nil {
+		t.Fatal(err)
+	}
+	if count != 1 {
+		t.Errorf("Enrollment through lost response and partial write created %d nodes, want 1", count)
+	}
+
+	// Knowing the private retry key alone is insufficient after the global discovery token rotates.
+	if err = repository.SaveOrUpdateSystemConfig(ctx, model.ConfigKeyAgentDiscoveryToken, "rotated-discovery"); err != nil {
+		t.Fatal(err)
+	}
+	request := protocol.RegisterNodeRequest{NodePayload: protocol.NodePayload{IP: "10.0.0.8", Version: "test"}, RegistrationKey: key}
+	for _, bearer := range []string{"", "invalid", "discovery-test"} {
+		client := httpclient.New(server.URL, bearer, time.Second)
+		result, authErr := client.RegisterNode(ctx, request)
+		if authErr == nil || result != nil {
+			t.Errorf("RegisterNode(unauthorized bearer) = %+v, %v; want rejection", result, authErr)
+		}
+	}
+	encoded, err := json.Marshal(request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	unauthorized := httptest.NewRecorder()
+	router.ServeHTTP(unauthorized, httptest.NewRequest(http.MethodPost, "/api/v1/agent/nodes/register", strings.NewReader(string(encoded))))
+	if unauthorized.Code != http.StatusUnauthorized || strings.Contains(unauthorized.Body.String(), third.Config.AccessToken) {
+		t.Errorf("Unauthorized retry response status = %d; want 401 without credential", unauthorized.Code)
+	}
+	client := httpclient.New(server.URL, "rotated-discovery", time.Second)
+	result, err := client.RegisterNode(ctx, request)
+	if err != nil || result.AccessToken != third.Config.AccessToken {
+		t.Errorf("RegisterNode(current discovery bearer) error = %v, want same identity", err)
+	}
+}

internal/apps/agent/agent/discovery_persistence_test.go

@@ -0,0 +1,85 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"testing"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/config"
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/protocol"
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/state"
+)
+
+func TestDiscoveryDoesNotRegisterBeforeRetrySecretPersists(t *testing.T) {
+	blockedParent := filepath.Join(t.TempDir(), "file")
+	if err := os.WriteFile(blockedParent, []byte("not a directory"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	service := &fakeHeartbeatService{}
+	runner := &Runner{Config: &config.Config{DiscoveryToken: "discovery"},
+		StateStore: state.NewStore(filepath.Join(blockedParent, "state.json")), HeartbeatService: service}
+	nodeID := "local"
+	if err := runner.tryRegister(context.Background(), &nodeID); err == nil {
+		t.Fatal("tryRegister(unwritable state) succeeded, want persistence error")
+	}
+	if service.registerCalls != 0 {
+		t.Errorf("tryRegister(unwritable retry key) sent %d requests, want 0", service.registerCalls)
+	}
+}
+
+func TestDiscoveryStateWriteFailureDoesNotActivateCredentials(t *testing.T) {
+	configPath := filepath.Join(t.TempDir(), "agent.json")
+	if err := os.WriteFile(configPath, []byte(`{"server_url":"http://127.0.0.1:3000","discovery_token":"discovery","node_ip":"10.0.0.8"}`), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	runner, nodeID := newDiscoveryTestRunner(t, configPath)
+	statePath := runner.Config.StatePath
+	backup := statePath + ".original"
+	service := &fakeHeartbeatService{registerResp: &protocol.RegisterNodeResponse{NodeID: "server-assigned", AccessToken: "issued"}}
+	var originalKey string
+	service.onRegister = func() {
+		snapshot, err := runner.StateStore.Load()
+		if err != nil {
+			t.Fatal(err)
+		}
+		originalKey = snapshot.RegistrationKey
+		if err = os.Rename(statePath, backup); err != nil {
+			t.Fatal(err)
+		}
+		if err = os.Mkdir(statePath, 0o700); err != nil {
+			t.Fatal(err)
+		}
+	}
+	runner.HeartbeatService = service
+	runner.HeartbeatCycle.Heartbeat = service
+	if err := runner.tryRegister(context.Background(), &nodeID); err == nil {
+		t.Fatal("tryRegister(state write failure) succeeded, want error")
+	}
+	if runner.hasAccessToken() || service.lastToken != "" {
+		t.Fatal("Failed state persistence activated credentials")
+	}
+	if err := os.Remove(statePath); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.Rename(backup, statePath); err != nil {
+		t.Fatal(err)
+	}
+	restarted, restartedID := newDiscoveryTestRunner(t, configPath)
+	service.onRegister = nil
+	restarted.HeartbeatService = service
+	restarted.HeartbeatCycle.Heartbeat = service
+	if err := restarted.tryRegister(context.Background(), &restartedID); err != nil {
+		t.Fatal(err)
+	}
+	key, err := restarted.StateStore.EnsureRegistrationKey()
+	if err != nil || key != originalKey {
+		t.Errorf("Key after state write failure/restart changed: %v", err)
+	}
+	if restartedID != "server-assigned" || service.lastToken != "issued" {
+		t.Errorf("Recovered identity/token = %q/%q, want server-assigned/issued", restartedID, service.lastToken)
+	}
+}

internal/apps/agent/agent/discovery_retry.go

@@ -0,0 +1,30 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import "time"
+
+const (
+	discoveryRetryInitialDelay = 3 * time.Second
+	discoveryRetryMaxDelay     = time.Minute
+	discoveryRetryMultiplier   = 2
+)
+
+type discoveryRetry struct {
+	delay time.Duration
+	next  time.Time
+}
+
+func (r *discoveryRetry) ready(now time.Time) bool {
+	return !now.Before(r.next)
+}
+
+func (r *discoveryRetry) failed(now time.Time) {
+	if r.delay == 0 {
+		r.delay = discoveryRetryInitialDelay
+	} else {
+		r.delay = min(r.delay*discoveryRetryMultiplier, discoveryRetryMaxDelay)
+	}
+	r.next = now.Add(r.delay)
+}

internal/apps/agent/agent/discovery_retry_test.go

@@ -0,0 +1,30 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import (
+	"testing"
+	"time"
+)
+
+func TestDiscoveryRetryBackoffIsIndependentOfHeartbeat(t *testing.T) {
+	now := time.Unix(100, 0)
+	var retry discoveryRetry
+	for _, delay := range []time.Duration{3, 6, 12, 24, 48, 60, 60} {
+		if !retry.ready(now) {
+			t.Fatal("Discovery retry not ready at scheduled time")
+		}
+		retry.failed(now)
+		wantNext := now.Add(delay * time.Second)
+		if retry.next != wantNext {
+			t.Errorf("Discovery retry next = %v, want %v", retry.next, wantNext)
+		}
+		for simulatedHeartbeat := now; simulatedHeartbeat.Before(wantNext); simulatedHeartbeat = simulatedHeartbeat.Add(time.Millisecond) {
+			if retry.ready(simulatedHeartbeat) {
+				t.Fatal("Discovery retry ready before independent enrollment backoff expires")
+			}
+		}
+		now = wantNext
+	}
+}

internal/apps/agent/agent/runner.go

@@ -23,7 +23,7 @@ import (
 
 // HeartbeatService handles node registration and periodic heartbeat reporting.
 type HeartbeatService interface {
-	Register(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error)
+	Register(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error)
 	Heartbeat(ctx context.Context, payload protocol.NodePayload) (*protocol.HeartbeatResult, error)
 	SetToken(token string)
 }
@@ -65,6 +65,7 @@ type Runner struct {
 	configSyncRetry         *configSyncRetry
 	restartOpenrestyNow     bool
 	websocketUpgradeEnabled bool
+	discoveryRetry          discoveryRetry
 }
 
 // Run starts the agent's main loop, performing heartbeats and upgrading to WebSocket when available.
@@ -120,6 +121,9 @@ func (r *Runner) Run(ctx context.Context) error {
 	tryStartWebSocket()
 
 	for {
+		if err := ctx.Err(); err != nil {
+			return err
+		}
 		select {
 		case <-ctx.Done():
 			slog.Info("agent runner shutting down", "error", ctx.Err())
@@ -162,6 +166,9 @@ func (r *Runner) handleWSDisconnect(ctx context.Context, nodeID string) {
 
 func (r *Runner) handleHeartbeatTick(ctx context.Context, nodeID *string, heartbeatTicker *time.Ticker, tryStartWebSocket func()) {
 	if !r.hasAccessToken() {
+		if !r.discoveryRetry.ready(time.Now()) {
+			return
+		}
 		if err := r.tryRegister(ctx, nodeID); err != nil {
 			slog.Error("agent discovery register failed", "error", err)
 		}
@@ -465,36 +472,46 @@ func (r *Runner) tryRestartOpenresty(ctx context.Context) {
 	r.recordOpenrestyHealthy()
 }
 
-func (r *Runner) tryRegister(ctx context.Context, nodeID *string) error {
+func (r *Runner) tryRegister(ctx context.Context, nodeID *string) (resultErr error) {
+	defer func() {
+		if resultErr != nil {
+			r.discoveryRetry.failed(time.Now())
+		}
+	}()
 	if strings.TrimSpace(r.Config.DiscoveryToken) == "" {
-		return errors.New("agent_token 为空且未配置 discovery_token")
+		return errors.New("agent token is empty and no discovery token is configured")
+	}
+	registrationKey, err := r.StateStore.EnsureRegistrationKey()
+	if err != nil {
+		return err
 	}
 	slog.Info("agent discovery registration started")
-	response, err := r.HeartbeatService.Register(ctx, r.HeartbeatCycle.NodePayload(ctx, *nodeID))
+	response, err := r.HeartbeatService.Register(ctx, protocol.RegisterNodeRequest{
+		NodePayload:     r.HeartbeatCycle.NodePayload(ctx, *nodeID),
+		RegistrationKey: registrationKey,
+	})
 	if err != nil {
 		return err
 	}
 	if response == nil || strings.TrimSpace(response.AccessToken) == "" || strings.TrimSpace(response.NodeID) == "" {
-		return errors.New("discovery register response 缺少 node_id 或 agent_token")
+		return errors.New("discovery registration response is missing node_id or agent_token")
 	}
-	snapshot, err := r.StateStore.Load()
+	if err = r.StateStore.SetRegisteredNodeID(response.NodeID); err != nil {
-	if err != nil {
 		return err
 	}
-	snapshot.NodeID = response.NodeID
+	updatedConfig := *r.Config
-	if err = r.StateStore.Save(snapshot); err != nil {
+	updatedConfig.AccessToken = response.AccessToken
-		return err
+	updatedConfig.DiscoveryToken = ""
-	}
+	if err = updatedConfig.Save(); err != nil {
-	r.Config.AccessToken = response.AccessToken
-	r.Config.DiscoveryToken = ""
-	if err = r.Config.Save(); err != nil {
 		return err
 	}
+	*r.Config = updatedConfig
 	r.HeartbeatService.SetToken(response.AccessToken)
 	if r.WebSocketService != nil {
 		r.WebSocketService.SetToken(response.AccessToken)
 	}
 	*nodeID = response.NodeID
+	r.discoveryRetry = discoveryRetry{}
 	slog.Info("agent discovery registration succeeded", "node_id", response.NodeID)
 	r.refreshOpenrestyHealth(ctx)
 	if _, err = r.HeartbeatCycle.Perform(ctx, *nodeID, true, r); err != nil {

internal/apps/agent/agent/runner_test.go

@@ -42,13 +42,17 @@ type fakeHeartbeatService struct {
 	heartbeatResults  []*protocol.HeartbeatResult
 	heartbeatPayloads []protocol.NodePayload
 	onHeartbeat       func(int)
+	onRegister        func()
 	lastToken         string
 }
 
-func (f *fakeHeartbeatService) Register(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error) {
+func (f *fakeHeartbeatService) Register(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error) {
 	f.mu.Lock()
 	defer f.mu.Unlock()
 	f.registerCalls++
+	if f.onRegister != nil {
+		f.onRegister()
+	}
 	return f.registerResp, f.registerErr
 }
 

internal/apps/agent/config/config.go

@@ -15,6 +15,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/localfile"
 	"github.com/Rain-kl/Wavelet/internal/apps/edge/nodeip"
 	"github.com/Rain-kl/Wavelet/pkg/util"
 )
@@ -406,7 +407,7 @@ func (cfg *Config) Save() error {
 	if err != nil {
 		return err
 	}
-	return os.WriteFile(cfg.configPath, data, configFilePerm)
+	return localfile.WriteAtomic(cfg.configPath, data, configFilePerm)
 }
 
 func detectHostname() string {

internal/apps/agent/heartbeat/service.go

@@ -11,14 +11,14 @@ import (
 
 // RemoteClient is the interface that abstracts the remote API calls performed by Service.
 type RemoteClient interface {
-	RegisterNode(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error)
+	RegisterNode(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error)
 	Heartbeat(ctx context.Context, payload protocol.NodePayload) (*protocol.HeartbeatResult, error)
 	SetToken(token string)
 }
 
 // API abstracts registration and heartbeat operations used by Cycle.
 type API interface {
-	Register(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error)
+	Register(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error)
 	Heartbeat(ctx context.Context, payload protocol.NodePayload) (*protocol.HeartbeatResult, error)
 	SetToken(token string)
 }
@@ -34,7 +34,7 @@ func New(client RemoteClient) *Service {
 }
 
 // Register sends a node registration request to the server.
-func (s *Service) Register(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error) {
+func (s *Service) Register(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error) {
 	return s.client.RegisterNode(ctx, payload)
 }
 

internal/apps/agent/httpclient/client.go

@@ -32,7 +32,7 @@ func New(baseURL string, token string, timeout time.Duration) *Client {
 }
 
 // RegisterNode registers the agent node with the control plane server.
-func (c *Client) RegisterNode(ctx context.Context, payload protocol.NodePayload) (*protocol.RegisterNodeResponse, error) {
+func (c *Client) RegisterNode(ctx context.Context, payload protocol.RegisterNodeRequest) (*protocol.RegisterNodeResponse, error) {
 	resp := protocol.APIResponse[protocol.RegisterNodeResponse]{}
 	if err := c.base.PostJSON(ctx, "/api/v1/agent/nodes/register", payload, &resp); err != nil {
 		return nil, err

internal/apps/agent/localfile/atomic.go

@@ -0,0 +1,66 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+// Package localfile persists agent identity and configuration without truncating existing files.
+package localfile
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+)
+
+// WriteAtomic syncs the complete contents, replaces path, then syncs its containing directory.
+// An error from the final sync means the replacement is visible but its durability is unconfirmed.
+func WriteAtomic(path string, data []byte, perm os.FileMode) error {
+	return writeAtomic(path, data, perm, SyncDirectory)
+}
+
+func writeAtomic(path string, data []byte, perm os.FileMode, syncDir func(string) error) error {
+	file, err := os.CreateTemp(filepath.Dir(path), "."+filepath.Base(path)+".tmp-*")
+	if err != nil {
+		return err
+	}
+	defer func() {
+		_ = file.Close()           // Best effort after an earlier write/close error.
+		_ = os.Remove(file.Name()) // Best effort; rename normally removes the temporary path.
+	}()
+	if err = file.Chmod(perm); err != nil {
+		return err
+	}
+	if _, err = file.Write(data); err != nil {
+		return err
+	}
+	if err = file.Sync(); err != nil {
+		return err
+	}
+	if err = file.Close(); err != nil {
+		return err
+	}
+	if err = os.Rename(file.Name(), path); err != nil {
+		return err
+	}
+	return syncDir(filepath.Dir(path))
+}
+
+// SyncDirectory confirms a directory entry update on platforms with directory flushing.
+// Windows provides best-effort replacement without a portable directory flush guarantee.
+func SyncDirectory(path string) (resultErr error) {
+	directory, err := os.Open(path) //nolint:gosec // callers provide the directory of a managed local state/config file
+	if err != nil {
+		return err
+	}
+	defer func() {
+		if closeErr := directory.Close(); resultErr == nil {
+			resultErr = closeErr
+		}
+	}()
+	info, err := directory.Stat()
+	if err != nil {
+		return err
+	}
+	if !info.IsDir() {
+		return errors.New("cannot sync a non-directory path")
+	}
+	return syncDirectory(directory)
+}

internal/apps/agent/localfile/atomic_test.go

@@ -0,0 +1,88 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package localfile
+
+import (
+	"errors"
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+func TestWriteAtomicReplacesCompleteFileAndCleansFailure(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "identity.json")
+	if err := WriteAtomic(path, []byte("old identity"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := WriteAtomic(path, []byte("new identity"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	data, err := os.ReadFile(path)
+	if err != nil || string(data) != "new identity" {
+		t.Errorf("WriteAtomic replacement = %q, %v; want complete new identity", data, err)
+	}
+	blocked := filepath.Join(dir, "directory")
+	if err = os.Mkdir(blocked, 0o700); err != nil {
+		t.Fatal(err)
+	}
+	if err = WriteAtomic(blocked, []byte("cannot replace directory"), 0o600); err == nil {
+		t.Fatal("WriteAtomic(directory) succeeded, want rename failure")
+	}
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) != 2 {
+		t.Errorf("WriteAtomic(after failure) directory entries = %d, want 2 with no temporary file", len(entries))
+	}
+}
+
+func TestWriteAtomicReportsDirectorySyncFailureAfterReplacement(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "identity.json")
+	if err := os.WriteFile(path, []byte("old identity"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	syncFailure := errors.New("directory I/O failure")
+	syncCalls := 0
+	err := writeAtomic(path, []byte("new identity"), 0o600, func(gotDir string) error {
+		syncCalls++
+		if gotDir != dir {
+			t.Errorf("writeAtomic sync path = %q, want %q", gotDir, dir)
+		}
+		data, readErr := os.ReadFile(path)
+		if readErr != nil || string(data) != "new identity" {
+			t.Errorf("Directory sync ran before replacement: contents = %q, error %v", data, readErr)
+		}
+		return syncFailure
+	})
+	if !errors.Is(err, syncFailure) || syncCalls != 1 {
+		t.Errorf("writeAtomic(directory sync failure) = %v, %d sync calls; want I/O failure, 1", err, syncCalls)
+	}
+	entries, err := os.ReadDir(dir)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(entries) != 1 {
+		t.Errorf("writeAtomic(failed directory sync) left %d entries, want only destination", len(entries))
+	}
+}
+
+func TestSyncDirectoryRejectsMissingAndNonDirectoryPaths(t *testing.T) {
+	dir := t.TempDir()
+	if err := SyncDirectory(dir); err != nil {
+		t.Fatalf("SyncDirectory(existing directory) = %v, want nil", err)
+	}
+	if err := SyncDirectory(filepath.Join(dir, "missing")); !errors.Is(err, os.ErrNotExist) {
+		t.Errorf("SyncDirectory(missing directory) = %v, want not-exist error", err)
+	}
+	path := filepath.Join(dir, "file")
+	if err := os.WriteFile(path, []byte("contents"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	if err := SyncDirectory(path); err == nil {
+		t.Error("SyncDirectory(regular file) = nil, want non-directory error")
+	}
+}

internal/apps/agent/localfile/sync_directory_unix.go

@@ -0,0 +1,12 @@
+//go:build !windows
+
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package localfile
+
+import "os"
+
+func syncDirectory(directory *os.File) error {
+	return directory.Sync()
+}

internal/apps/agent/localfile/sync_directory_windows.go

@@ -0,0 +1,12 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package localfile
+
+import "os"
+
+// Windows does not support a portable directory fsync through os.File.Sync.
+// The file is flushed before replacement; persisting the directory entry is best effort.
+func syncDirectory(_ *os.File) error {
+	return nil
+}

internal/apps/agent/protocol/alias.go

@@ -51,6 +51,9 @@ type NodeHealthEvent = pkgprotocol.NodeHealthEvent
 // RegisterNodeResponse is an alias for pkgprotocol.RegisterNodeResponse.
 type RegisterNodeResponse = pkgprotocol.RegisterNodeResponse
 
+// RegisterNodeRequest is an enrollment-only request, including its retry credential.
+type RegisterNodeRequest = pkgprotocol.RegisterNodeRequest
+
 // ApplyLogPayload is an alias for pkgprotocol.ApplyLogPayload.
 type ApplyLogPayload = pkgprotocol.ApplyLogPayload
 

internal/apps/agent/state/registration_test.go

@@ -0,0 +1,84 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package state
+
+import (
+	"encoding/hex"
+	"errors"
+	"path/filepath"
+	"testing"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/localfile"
+)
+
+func TestRegistrationKeySurvivesStaleSnapshotAndRestart(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "state.json")
+	store := NewStore(path)
+	stale, err := store.Load()
+	if err != nil {
+		t.Fatal(err)
+	}
+	key, err := store.EnsureRegistrationKey()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if err = store.SetRegisteredNodeID("node-server-assigned"); err != nil {
+		t.Fatal(err)
+	}
+	decoded, err := hex.DecodeString(key)
+	if err != nil || len(decoded) != 32 {
+		t.Fatalf("EnsureRegistrationKey() length = %d, error %v; want 32, nil", len(decoded), err)
+	}
+	stale.LastError = "background sync failed"
+	if err = store.Save(stale); err != nil {
+		t.Fatal(err)
+	}
+	restartedKey, err := NewStore(path).EnsureRegistrationKey()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if restartedKey != key {
+		t.Error("EnsureRegistrationKey(after stale save and restart) changed enrollment identity")
+	}
+	snapshot, err := NewStore(path).Load()
+	if err != nil {
+		t.Fatal(err)
+	}
+	if snapshot.NodeID != "node-server-assigned" {
+		t.Errorf("Load(after stale save).NodeID = %q, want server-assigned ID", snapshot.NodeID)
+	}
+}
+
+func TestExistingRegistrationKeyRequiresDirectorySyncBeforeUse(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "state.json")
+	store := NewStore(path)
+	key, err := store.EnsureRegistrationKey()
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Model a prior successful rename whose final directory flush was not confirmed.
+	syncFailure := errors.New("directory I/O failure")
+	syncCalls := 0
+	store.syncDirectory = func(dir string) error {
+		syncCalls++
+		if dir != filepath.Dir(path) {
+			t.Errorf("EnsureRegistrationKey sync directory = %q, want state parent", dir)
+		}
+		return syncFailure
+	}
+	for range 2 {
+		got, retryErr := store.EnsureRegistrationKey()
+		if got != "" || !errors.Is(retryErr, syncFailure) {
+			t.Errorf("EnsureRegistrationKey(unconfirmed directory) returned usable key or lost error: %v", retryErr)
+		}
+	}
+	if syncCalls != 2 {
+		t.Errorf("EnsureRegistrationKey(repeated failures) sync calls = %d, want 2", syncCalls)
+	}
+	store.syncDirectory = localfile.SyncDirectory
+	confirmedKey, err := store.EnsureRegistrationKey()
+	if err != nil || confirmedKey != key {
+		t.Errorf("EnsureRegistrationKey(after successful sync) changed identity or failed: %v", err)
+	}
+}

internal/apps/agent/state/state.go

@@ -10,12 +10,15 @@ import (
 	"os"
 	"path/filepath"
 	"sync"
+
+	"github.com/Rain-kl/Wavelet/internal/apps/agent/localfile"
 )
 
 const (
-	stateDirPerm      = 0o750
+	stateDirPerm         = 0o750
-	stateFilePerm     = 0o600
+	stateFilePerm        = 0o600
-	nodeIDRandomBytes = 8
+	nodeIDRandomBytes    = 8
+	registrationKeyBytes = 32
 )
 
 // PagesDeployment records a Pages project tracked by the agent and the last
@@ -31,6 +34,7 @@ type PagesDeployment struct {
 // Snapshot represents the state of the agent at a given point in time.
 type Snapshot struct {
 	NodeID                 string            `json:"node_id"`
+	RegistrationKey        string            `json:"registration_key,omitempty"`
 	CurrentVersion         string            `json:"current_version"`
 	CurrentChecksum        string            `json:"current_checksum"`
 	PagesDeployments       []PagesDeployment `json:"pages_deployments"`
@@ -47,15 +51,42 @@ type Snapshot struct {
 	AccessLogOffset        int64             `json:"access_log_offset"`
 }
 
+// EnsureRegistrationKey persists the private retry credential before enrollment can reach the server.
+func (s *Store) EnsureRegistrationKey() (string, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	snapshot, err := s.loadUnlocked()
+	if err != nil {
+		return "", err
+	}
+	if snapshot.RegistrationKey != "" {
+		// A previous rename may have succeeded even though its directory sync failed.
+		if err = s.syncDirectory(filepath.Dir(s.path)); err != nil {
+			return "", err
+		}
+		return snapshot.RegistrationKey, nil
+	}
+	key := make([]byte, registrationKeyBytes)
+	if _, err = rand.Read(key); err != nil {
+		return "", err
+	}
+	snapshot.RegistrationKey = hex.EncodeToString(key)
+	if err = s.saveUnlocked(snapshot); err != nil {
+		return "", err
+	}
+	return snapshot.RegistrationKey, nil
+}
+
 // Store manages the storage and retrieval of the agent state snapshot.
 type Store struct {
-	path string
+	path          string
-	mu   sync.Mutex
+	mu            sync.Mutex
+	syncDirectory func(string) error
 }
 
 // NewStore creates a new Store instance at the given path.
 func NewStore(path string) *Store {
-	return &Store{path: filepath.Clean(path)}
+	return &Store{path: filepath.Clean(path), syncDirectory: localfile.SyncDirectory}
 }
 
 // Load loads the snapshot from the store.
@@ -91,6 +122,29 @@ func (s *Store) EnsureNodeID() (string, error) {
 func (s *Store) Save(snapshot *Snapshot) error {
 	s.mu.Lock()
 	defer s.mu.Unlock()
+	current, err := s.loadUnlocked()
+	if err != nil {
+		return err
+	}
+	updated := *snapshot
+
+	// Other workers can hold a snapshot loaded before enrollment first persisted its key.
+	if current.RegistrationKey != "" {
+		updated.RegistrationKey = current.RegistrationKey
+		updated.NodeID = current.NodeID
+	}
+	return s.saveUnlocked(&updated)
+}
+
+// SetRegisteredNodeID commits the server-assigned identity without overwriting concurrent runtime state.
+func (s *Store) SetRegisteredNodeID(nodeID string) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	snapshot, err := s.loadUnlocked()
+	if err != nil {
+		return err
+	}
+	snapshot.NodeID = nodeID
 	return s.saveUnlocked(snapshot)
 }
 
@@ -120,7 +174,7 @@ func (s *Store) saveUnlocked(snapshot *Snapshot) error {
 	if err != nil {
 		return err
 	}
-	return os.WriteFile(s.path, data, stateFilePerm)
+	return localfile.WriteAtomic(s.path, data, stateFilePerm)
 }
 
 func newNodeID() (string, error) {

internal/apps/openflare/agent/errs.go

@@ -4,17 +4,19 @@
 package agent
 
 const (
-	errMissingAgentToken      = "缺少 Agent Token"         //nolint:gosec // false positive: this is an error message, not hardcoded credentials
+	errMissingAgentToken             = "缺少 Agent Token"         //nolint:gosec // false positive: this is an error message, not hardcoded credentials
-	errInvalidAgentToken      = "无权进行此操作，Agent Token 无效" //nolint:gosec // false positive: this is an error message, not hardcoded credentials
+	errInvalidAgentToken             = "无权进行此操作，Agent Token 无效" //nolint:gosec // false positive: this is an error message, not hardcoded credentials
-	errInvalidDiscoveryToken  = "无权进行此操作，注册 Token 无效"    //nolint:gosec // false positive: this is an error message, not hardcoded credentials
+	errInvalidDiscoveryToken         = "无权进行此操作，注册 Token 无效"    //nolint:gosec // false positive: this is an error message, not hardcoded credentials
-	errNodeMissingFromContext = "Node object missing from context"
+	errNodeMissingFromContext        = "Node object missing from context"
-	errNoActiveConfig         = "当前没有激活版本"
+	errNoActiveConfig                = "当前没有激活版本"
-	errNodeNotFound           = "节点不存在"
+	errNodeNotFound                  = "节点不存在"
-	errNodeIDRequired         = "node_id 不能为空"
+	errNodeIDRequired                = "node_id 不能为空"
-	errVersionRequired        = "version 不能为空"
+	errVersionRequired               = "version 不能为空"
-	errInvalidApplyResult     = "result 仅支持 success、warning 或 failed"
+	errInvalidApplyResult            = "result 仅支持 success、warning 或 failed"
-	errIPRequired             = "ip 不能为空"
+	errIPRequired                    = "ip 不能为空"
-	errIPInvalid              = "ip 格式无效"
+	errIPInvalid                     = "ip 格式无效"
-	errAgentVersionRequired   = "version 不能为空"
+	errAgentVersionRequired          = "version 不能为空"
-	errNodeIDConflict         = "节点标识生成冲突，请重试"
+	errNodeIDConflict                = "节点标识生成冲突，请重试"
+	errRegistrationKeyInvalid        = "invalid discovery registration key"
+	errRegistrationCredentialChanged = "discovery enrollment credential has changed; use the assigned agent token"
 )

internal/apps/openflare/agent/logics.go

@@ -5,6 +5,7 @@ package agent
 
 import (
 	"context"
+	"crypto/subtle"
 	"errors"
 	"strings"
 	"time"
@@ -41,18 +42,14 @@ func RegisterWithAccessToken(ctx context.Context, authNode *model.OpenFlareNode,
 }
 
 // RegisterWithDiscovery registers a new node using the global discovery token.
-func RegisterWithDiscovery(ctx context.Context, payload NodePayload) (*RegistrationResponse, error) {
+func RegisterWithDiscovery(ctx context.Context, payload NodePayload, registrationKey string) (*RegistrationResponse, error) {
 	_ = ofgeoip.EnsureRuntimeProvider(ctx)
 	payload = normalizeNodePayload(payload)
 	if err := validateNodePayload(payload); err != nil {
 		return nil, err
 	}
 
-	nodeID, err := newServerNodeID()
+	nodeID, accessToken, err := newDiscoveryIdentity(registrationKey)
-	if err != nil {
-		return nil, err
-	}
-	accessToken, err := newRandomToken()
 	if err != nil {
 		return nil, err
 	}
@@ -73,7 +70,15 @@ func RegisterWithDiscovery(ctx context.Context, payload NodePayload) (*Registrat
 	}
 	applyNodeRuntime(ctx, record, payload, false)
 
-	if err = repository.CreateOpenFlareNode(ctx, record); err != nil {
+	if registrationKey != "" {
+		record, err = repository.CreateOrGetOpenFlareDiscoveryNode(ctx, record)
+		if err == nil && (record.NodeType != "edge_node" || subtle.ConstantTimeCompare([]byte(record.AccessToken), []byte(accessToken)) != 1) {
+			return nil, errors.New(errRegistrationCredentialChanged)
+		}
+	} else {
+		err = repository.CreateOpenFlareNode(ctx, record)
+	}
+	if err != nil {
 		if isUniqueConstraintError(err) {
 			return nil, errors.New(errNodeIDConflict)
 		}
@@ -87,6 +92,18 @@ func RegisterWithDiscovery(ctx context.Context, payload NodePayload) (*Registrat
 	}, nil
 }
 
+func newDiscoveryIdentity(registrationKey string) (string, string, error) {
+	if registrationKey != "" {
+		return discoveryIdentity(registrationKey)
+	}
+	nodeID, err := newServerNodeID()
+	if err != nil {
+		return "", "", err
+	}
+	token, err := newRandomToken()
+	return nodeID, token, err
+}
+
 // HeartbeatNode updates runtime state and returns agent settings.
 func HeartbeatNode(ctx context.Context, authNode *model.OpenFlareNode, payload NodePayload) (*HeartbeatResponse, error) {
 	_ = ofgeoip.EnsureRuntimeProvider(ctx)

internal/apps/openflare/agent/registration.go

@@ -0,0 +1,25 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import (
+	"crypto/sha256"
+	"encoding/base64"
+	"encoding/hex"
+	"errors"
+)
+
+const registrationKeyBytes = 32
+
+// discoveryIdentity derives separate public identity and private credential from a random retry secret.
+// Public node IDs cannot be used as registration keys to recover credentials.
+func discoveryIdentity(key string) (string, string, error) {
+	decoded, err := hex.DecodeString(key)
+	if err != nil || len(decoded) != registrationKeyBytes {
+		return "", "", errors.New(errRegistrationKeyInvalid)
+	}
+	nodeHash := sha256.Sum256(append([]byte("openflare:discovery:node:v1:"), decoded...))
+	tokenHash := sha256.Sum256(append([]byte("openflare:discovery:token:v1:"), decoded...))
+	return "node-" + base64.RawURLEncoding.EncodeToString(nodeHash[:]), hex.EncodeToString(tokenHash[:]), nil
+}

internal/apps/openflare/agent/registration_test.go

@@ -0,0 +1,168 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package agent
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	agentclient "github.com/Rain-kl/Wavelet/internal/apps/agent/httpclient"
+	db "github.com/Rain-kl/Wavelet/internal/infra/persistence"
+	"github.com/Rain-kl/Wavelet/internal/model"
+	"github.com/Rain-kl/Wavelet/internal/repository"
+	"github.com/Rain-kl/Wavelet/internal/shared/response"
+	"github.com/Rain-kl/Wavelet/pkg/protocol"
+)
+
+func TestRegistrationResponseServerClientContract(t *testing.T) {
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		if err := json.NewEncoder(w).Encode(response.OK(RegistrationResponse{NodeID: "node-contract", AccessToken: "issued", Name: "edge"})); err != nil {
+			t.Error(err)
+		}
+	}))
+	t.Cleanup(server.Close)
+	client := agentclient.New(server.URL, "discovery", time.Second)
+	got, err := client.RegisterNode(context.Background(), protocol.RegisterNodeRequest{})
+	if err != nil {
+		t.Fatalf("RegisterNode(server response) error = %v", err)
+	}
+	if got.NodeID != "node-contract" || got.AccessToken != "issued" || got.Name != "edge" {
+		t.Errorf("RegisterNode(server response) = %+v, want node-contract/issued/edge", got)
+	}
+	data, err := json.Marshal(RegistrationResponse{AccessToken: "issued"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	var legacy struct {
+		AccessToken string `json:"access_token"`
+	}
+	if err = json.Unmarshal(data, &legacy); err != nil || legacy.AccessToken != "issued" {
+		t.Errorf("Decode legacy server response = %q, %v; want issued, nil", legacy.AccessToken, err)
+	}
+	var documented RegistrationResponseDocs
+	if err = json.Unmarshal(data, &documented); err != nil {
+		t.Fatal(err)
+	}
+	if documented.AgentToken != "issued" || documented.AccessToken != documented.AgentToken {
+		t.Error("RegistrationResponseDocs does not describe both emitted token aliases")
+	}
+}
+
+func TestDiscoveryRetriesKeepOneIdentityAndPreserveChanges(t *testing.T) {
+	t.Cleanup(setupAgentAuthTestDB(t))
+	ctx := context.Background()
+	key := strings.Repeat("a1", 32)
+	payload := NodePayload{IP: "10.0.0.8", Version: "test", Name: "edge"}
+	first, err := RegisterWithDiscovery(ctx, payload, key)
+	if err != nil {
+		t.Fatal(err)
+	}
+	record, err := repository.GetOpenFlareNodeByNodeID(ctx, first.NodeID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	record.Name = "administrator name"
+	record.UpdateRequested = true
+	if err = repository.SaveOpenFlareNode(ctx, record); err != nil {
+		t.Fatal(err)
+	}
+	second, err := RegisterWithDiscovery(ctx, payload, key)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if first.NodeID != second.NodeID || first.AccessToken != second.AccessToken {
+		t.Errorf("RegisterWithDiscovery(retry) identity = %+v, want original identity", second)
+	}
+	record, err = repository.GetOpenFlareNodeByNodeID(ctx, first.NodeID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if record.Name != "administrator name" || !record.UpdateRequested {
+		t.Errorf("RegisterWithDiscovery(retry) overwrote administrator metadata: %+v", record)
+	}
+	other, err := RegisterWithDiscovery(ctx, payload, strings.Repeat("b2", 32))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if other.NodeID == first.NodeID || other.AccessToken == first.AccessToken {
+		t.Error("RegisterWithDiscovery(distinct key, same host/IP) reused identity")
+	}
+	var count int64
+	if err = db.DB(ctx).Model(&model.OpenFlareNode{}).Count(&count).Error; err != nil {
+		t.Fatal(err)
+	}
+	if count != 2 {
+		t.Errorf("RegisterWithDiscovery(two keys, retry) node count = %d, want 2", count)
+	}
+}
+
+func TestDiscoveryConcurrentRetriesAndRevokedCredential(t *testing.T) {
+	t.Cleanup(setupAgentAuthTestDB(t))
+	ctx := context.Background()
+	sqlDB, err := db.DB(ctx).DB()
+	if err != nil {
+		t.Fatal(err)
+	}
+	sqlDB.SetMaxOpenConns(1)
+	key := strings.Repeat("c3", 32)
+	payload := NodePayload{IP: "10.0.0.8", Version: "test"}
+	const attempts = 12
+	results := make(chan *RegistrationResponse, attempts)
+	errors := make(chan error, attempts)
+	var wg sync.WaitGroup
+	for range attempts {
+		wg.Go(func() {
+			result, registrationErr := RegisterWithDiscovery(ctx, payload, key)
+			results <- result
+			errors <- registrationErr
+		})
+	}
+	wg.Wait()
+	close(results)
+	close(errors)
+	for registrationErr := range errors {
+		if registrationErr != nil {
+			t.Fatalf("Concurrent RegisterWithDiscovery error = %v", registrationErr)
+		}
+	}
+	var first *RegistrationResponse
+	for result := range results {
+		if first == nil {
+			first = result
+		}
+		if result.NodeID != first.NodeID || result.AccessToken != first.AccessToken {
+			t.Error("Concurrent enrollment returned different identities")
+		}
+	}
+	var count int64
+	if err = db.DB(ctx).Model(&model.OpenFlareNode{}).Count(&count).Error; err != nil {
+		t.Fatal(err)
+	}
+	if count != 1 {
+		t.Errorf("Concurrent enrollment node count = %d, want 1", count)
+	}
+	record, err := repository.GetOpenFlareNodeByNodeID(ctx, first.NodeID)
+	if err != nil {
+		t.Fatal(err)
+	}
+	record.AccessToken = "rotated-by-administrator"
+	if err = repository.SaveOpenFlareNode(ctx, record); err != nil {
+		t.Fatal(err)
+	}
+	if result, retryErr := RegisterWithDiscovery(ctx, payload, key); retryErr == nil || result != nil {
+		t.Errorf("RegisterWithDiscovery(rotated credential) = %+v, %v; want rejection", result, retryErr)
+	}
+	for _, invalid := range []string{"short", strings.Repeat("g", 64), first.NodeID} {
+		if result, retryErr := RegisterWithDiscovery(ctx, payload, invalid); retryErr == nil || result != nil {
+			t.Errorf("RegisterWithDiscovery(invalid/public key) = %+v, %v; want rejection", result, retryErr)
+		}
+	}
+}

internal/apps/openflare/agent/routers.go

@@ -22,16 +22,17 @@ import (
 // @Accept json
 // @Produce json
 // @Security AgentTokenAuth
-// @Param body body agent.NodePayload true "节点上报数据"
+// @Param body body protocol.RegisterNodeRequest true "节点上报数据；registration_key 为首次注册重试凭证"
-// @Success 200 {object} response.Any{data=agent.RegistrationResponse} "注册成功"
+// @Success 200 {object} response.Any{data=agent.RegistrationResponseDocs} "注册成功"
 // @Failure 400 {object} response.Any "参数错误"
 // @Failure 401 {object} response.Any "Token 无效"
 // @Router /api/v1/agent/nodes/register [post]
 func RegisterHandler(c *gin.Context) {
-	var payload NodePayload
+	var request protocol.RegisterNodeRequest
-	if !apiutil.BindJSON(c, &payload) {
+	if !apiutil.BindJSON(c, &request) {
 		return
 	}
+	payload := request.NodePayload
 	payload.IP = resolveReportedNodeIP(payload.IP, c.Request.RemoteAddr)
 
 	var (
@@ -41,7 +42,7 @@ func RegisterHandler(c *gin.Context) {
 	if authNode, ok := NodeFromContext(c); ok {
 		result, err = RegisterWithAccessToken(c.Request.Context(), authNode, payload)
 	} else {
-		result, err = RegisterWithDiscovery(c.Request.Context(), payload)
+		result, err = RegisterWithDiscovery(c.Request.Context(), payload, request.RegistrationKey)
 	}
 	if apiutil.AbortBadRequestOnError(c, err) {
 		return

internal/apps/openflare/agent/types.go

@@ -7,6 +7,7 @@ import (
 	"time"
 
 	"github.com/Rain-kl/Wavelet/internal/model"
+	"github.com/Rain-kl/Wavelet/pkg/protocol"
 )
 
 const (
@@ -17,9 +18,13 @@ const (
 )
 
 // RegistrationResponse is returned after agent registration.
-// Server uses access_token; the agent client expects agent_token via RegisterNodeResponse.
+type RegistrationResponse = protocol.RegisterNodeResponse
-type RegistrationResponse struct {
+
+// RegistrationResponseDocs describes both token aliases emitted by the custom JSON marshaler.
+// It is used only by Swagger; the runtime contract remains protocol.RegisterNodeResponse.
+type RegistrationResponseDocs struct {
 	NodeID      string `json:"node_id"`
+	AgentToken  string `json:"agent_token"`
 	AccessToken string `json:"access_token"`
 	Name        string `json:"name"`
 }

internal/repository/openflare_node_registration.go

@@ -0,0 +1,28 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package repository
+
+import (
+	"context"
+	"errors"
+
+	db "github.com/Rain-kl/Wavelet/internal/infra/persistence"
+	"github.com/Rain-kl/Wavelet/internal/model"
+	"gorm.io/gorm/clause"
+)
+
+// CreateOrGetOpenFlareDiscoveryNode inserts an enrollment once, without changing an existing identity.
+func CreateOrGetOpenFlareDiscoveryNode(ctx context.Context, node *model.OpenFlareNode) (*model.OpenFlareNode, error) {
+	conn := db.DB(ctx)
+	if conn == nil {
+		return nil, errors.New(errDatabaseNotInitialized)
+	}
+	if err := conn.Clauses(clause.OnConflict{
+		Columns:   []clause.Column{{Name: "node_id"}},
+		DoNothing: true,
+	}).Create(node).Error; err != nil {
+		return nil, err
+	}
+	return GetOpenFlareNodeByNodeID(ctx, node.NodeID)
+}

pkg/protocol/agent.go

@@ -4,7 +4,11 @@
 // Package protocol defines the communication protocol between OpenFlare server, agent, and relay components.
 package protocol
 
-import "encoding/json"
+import (
+	"encoding/json"
+	"errors"
+	"strings"
+)
 
 // APIResponse is a generic API response wrapper.
 type APIResponse[T any] struct {
@@ -166,6 +170,12 @@ type NodeHealthEvent struct {
 	Metadata        map[string]string `json:"metadata,omitempty"`
 }
 
+// RegisterNodeRequest carries enrollment data separately from heartbeat telemetry.
+type RegisterNodeRequest struct {
+	NodePayload
+	RegistrationKey string `json:"registration_key,omitempty"`
+}
+
 // RegisterNodeResponse is the node registration response.
 type RegisterNodeResponse struct {
 	NodeID      string `json:"node_id"`
@@ -173,6 +183,38 @@ type RegisterNodeResponse struct {
 	Name        string `json:"name"`
 }
 
+// MarshalJSON retains access_token for consumers of older server responses.
+func (r RegisterNodeResponse) MarshalJSON() ([]byte, error) {
+	type response RegisterNodeResponse
+	return json.Marshal(struct {
+		response
+		LegacyAccessToken string `json:"access_token"`
+	}{response(r), r.AccessToken})
+}
+
+// UnmarshalJSON accepts either spelling used by released servers.
+func (r *RegisterNodeResponse) UnmarshalJSON(data []byte) error {
+	type response RegisterNodeResponse
+	var decoded struct {
+		response
+		LegacyAccessToken string `json:"access_token"`
+	}
+	if err := json.Unmarshal(data, &decoded); err != nil {
+		return err
+	}
+	canonical := strings.TrimSpace(decoded.AccessToken)
+	legacy := strings.TrimSpace(decoded.LegacyAccessToken)
+	if canonical != "" && legacy != "" && canonical != legacy {
+		return errors.New("registration response contains conflicting agent tokens")
+	}
+	*r = RegisterNodeResponse(decoded.response)
+	r.AccessToken = canonical
+	if canonical == "" {
+		r.AccessToken = legacy
+	}
+	return nil
+}
+
 // ActiveConfigResponse is the active configuration response.
 type ActiveConfigResponse struct {
 	Version          string        `json:"version"`

pkg/protocol/agent_registration_test.go

@@ -0,0 +1,54 @@
+// Copyright 2026 Arctel.net
+// SPDX-License-Identifier: Apache-2.0
+
+package protocol
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+func TestRegisterNodeResponseTokenAliases(t *testing.T) {
+	for _, tt := range []struct {
+		name    string
+		body    string
+		want    string
+		wantErr bool
+	}{
+		{name: "canonical", body: `{"node_id":"node-1","agent_token":"issued"}`, want: "issued"},
+		{name: "legacy", body: `{"node_id":"node-1","access_token":"issued"}`, want: "issued"},
+		{name: "matching aliases", body: `{"agent_token":"issued","access_token":"issued"}`, want: "issued"},
+		{name: "conflicting aliases", body: `{"agent_token":"a","access_token":"b"}`, wantErr: true},
+		{name: "invalid JSON", body: `{`, wantErr: true},
+	} {
+		t.Run(tt.name, func(t *testing.T) {
+			var got RegisterNodeResponse
+			err := json.Unmarshal([]byte(tt.body), &got)
+			if (err != nil) != tt.wantErr {
+				t.Fatalf("Unmarshal(%s) error = %v, want error %t", tt.name, err, tt.wantErr)
+			}
+			if got.AccessToken != tt.want {
+				t.Errorf("Unmarshal(%s).AccessToken = %q, want %q", tt.name, got.AccessToken, tt.want)
+			}
+		})
+	}
+}
+
+func TestRegistrationRequestKeepsRetryCredentialOutOfTelemetry(t *testing.T) {
+	request := RegisterNodeRequest{NodePayload: NodePayload{NodeID: "node-1"}, RegistrationKey: "private"}
+	registration, err := json.Marshal(request)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(registration), `"registration_key":"private"`) {
+		t.Errorf("Marshal(registration) = %s, want retry credential", registration)
+	}
+	telemetry, err := json.Marshal(request.NodePayload)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if strings.Contains(string(telemetry), "registration_key") || strings.Contains(string(telemetry), "private") {
+		t.Errorf("Marshal(heartbeat) = %s, want no enrollment credential", telemetry)
+	}
+}