merge: fix/agent-clear-last-error-on-sync-success

Merge agent sticky LastError clear fix into main.
2026-09-28 05:46:36 +08:00 · 2026-08-06 13:48:14 +08:00
parent 1f5bebd18a 1a4a03a20d
commit 9d93dc0b9f
4 changed files with 168 additions and 0 deletions
@@ -26,6 +26,10 @@ sidebar: false

 - 新增 Cloudflare DNS 指向管理：可复用现有 Cloudflare DNS 账号或配置独立 Token，按分组将 ZoneDomain 的单条 A 记录异步同步到边缘节点 IPv4，并支持成员橙云、同步状态与节点 IP 变更联动。

+### 修复
+
+- 修复 Agent 在配置已对齐但磁盘校验和不一致时，Pages 等对账成功后仍保留 `LastError` 的问题，避免偶发网络失败被健康事件长期显示为「活动中」且无法自动恢复。
+
 ### 改进

 - 删除、撤销与未保存离开等确认操作统一改用页面内 AlertDialog，不再使用浏览器原生 `confirm` 弹窗，交互风格与系统其余对话框保持一致。
@@ -166,6 +166,8 @@ func (s *Service) applyIfNeeded(ctx context.Context, mode string, startup bool,
 			return err
 		}
 		slog.Debug("skipping apply because state already records target version/checksum", "version", config.Version, "checksum", config.Checksum)
+		// Successful reconcile must clear sticky LastError so health events can resolve.
+		snapshot.LastError = ""
 		return s.stateStore.Save(snapshot)
 	}
 	return s.applyRenderedConfig(ctx, mode, snapshot, currentChecksum, config)
@@ -1804,3 +1804,163 @@ func TestSyncOnceDownloadsPagesDeploymentWithTopLevelFolder(t *testing.T) {
 		t.Fatalf("unexpected Pages assets/app.js content: %s", string(jsData))
 	}
 }
+
+func TestSyncOnceClearsStickyLastErrorWhenChecksumMatches(t *testing.T) {
+	const sticky = "pages project 1: fetch latest hash: context deadline exceeded"
+	client := &fakeClient{
+		config: protocol.ActiveConfigResponse{
+			Version:          "20260309-201",
+			Checksum:         "checksum-201",
+			SourceConfigJSON: testSourceConfigJSON("auto", 90),
+			CreatedAt:        time.Now().Format(time.RFC3339),
+		},
+	}
+	stateStore := state.NewStore(filepath.Join(t.TempDir(), "state.json"))
+	nodeID, err := stateStore.EnsureNodeID()
+	if err != nil {
+		t.Fatalf("EnsureNodeID failed: %v", err)
+	}
+	if err = stateStore.Save(&state.Snapshot{
+		NodeID:           nodeID,
+		CurrentVersion:   client.config.Version,
+		CurrentChecksum:  client.config.Checksum,
+		PagesDeployments: []state.PagesDeployment{},
+		LastError:        sticky,
+	}); err != nil {
+		t.Fatalf("failed to seed state: %v", err)
+	}
+
+	manager := &fakeManager{currentChecksum: client.config.Checksum}
+	service := New(client, manager, stateStore)
+	if err = service.SyncOnce(context.Background(), &protocol.ActiveConfigMeta{
+		Version:  client.config.Version,
+		Checksum: client.config.Checksum,
+	}); err != nil {
+		t.Fatalf("SyncOnce failed: %v", err)
+	}
+
+	snapshot, err := stateStore.Load()
+	if err != nil {
+		t.Fatalf("failed to load state: %v", err)
+	}
+	if snapshot.LastError != "" {
+		t.Fatalf("expected sticky LastError to clear after successful up-to-date sync, got %q", snapshot.LastError)
+	}
+}
+
+func TestSyncOnceClearsStickyLastErrorWhenStateMatchesButDiskChecksumDiffers(t *testing.T) {
+	// Reproduces the production sticky-alert bug: nginx on-disk checksum no longer
+	// equals the active config checksum, but agent state already records the target.
+	// Pages reconcile succeeds, yet older code saved state without clearing LastError.
+	const sticky = "pages project 1: fetch latest hash: connection reset by peer"
+	const version = "20260309-202"
+	const checksum = "checksum-202"
+
+	packageBytes := testPagesPackage(t, map[string]string{"index.html": "ok"})
+	hash := testBytesChecksum(packageBytes)
+	projectID := uint(1)
+	client := &fakeClient{
+		config: protocol.ActiveConfigResponse{
+			Version:          version,
+			Checksum:         checksum,
+			SourceConfigJSON: testPagesSourceConfigJSON(projectID, projectID, hash),
+			CreatedAt:        time.Now().Format(time.RFC3339),
+		},
+		pagesPackages: map[uint][]byte{projectID: packageBytes},
+		pagesHashes:   map[uint]string{projectID: hash},
+	}
+	stateStore := state.NewStore(filepath.Join(t.TempDir(), "state.json"))
+	nodeID, err := stateStore.EnsureNodeID()
+	if err != nil {
+		t.Fatalf("EnsureNodeID failed: %v", err)
+	}
+	if err = stateStore.Save(&state.Snapshot{
+		NodeID:          nodeID,
+		CurrentVersion:  version,
+		CurrentChecksum: checksum,
+		LastError:       sticky,
+		PagesDeployments: []state.PagesDeployment{{
+			ProjectID:    projectID,
+			DeploymentID: projectID,
+			Hash:         hash,
+		}},
+	}); err != nil {
+		t.Fatalf("failed to seed state: %v", err)
+	}
+
+	// Disk checksum differs from target → mismatched path; state already matches target.
+	manager := &fakeManager{currentChecksum: "on-disk-differs-from-target"}
+	service := New(client, manager, stateStore)
+	service.SetPagesDir(t.TempDir())
+
+	if err = service.SyncOnce(context.Background(), &protocol.ActiveConfigMeta{
+		Version:  version,
+		Checksum: checksum,
+	}); err != nil {
+		t.Fatalf("SyncOnce failed: %v", err)
+	}
+
+	snapshot, err := stateStore.Load()
+	if err != nil {
+		t.Fatalf("failed to load state: %v", err)
+	}
+	if snapshot.LastError != "" {
+		t.Fatalf("expected sticky LastError to clear when pages reconcile succeeds, got %q", snapshot.LastError)
+	}
+}
+
+func TestSyncOnceClearsStickyLastErrorAfterStaleHeartbeatFetchesActiveConfig(t *testing.T) {
+	// Heartbeat meta is stale vs active config; disk checksum mismatches heartbeat.
+	// After fetch, snapshot already matches active config → applyIfNeeded state-match path.
+	const sticky = "previous transient pages hash timeout"
+	const activeVersion = "20260309-203"
+	const activeChecksum = "checksum-203"
+
+	client := &fakeClient{
+		config: protocol.ActiveConfigResponse{
+			Version:          activeVersion,
+			Checksum:         activeChecksum,
+			SourceConfigJSON: testSourceConfigJSON("auto", 91),
+			CreatedAt:        time.Now().Format(time.RFC3339),
+		},
+	}
+	stateStore := state.NewStore(filepath.Join(t.TempDir(), "state.json"))
+	nodeID, err := stateStore.EnsureNodeID()
+	if err != nil {
+		t.Fatalf("EnsureNodeID failed: %v", err)
+	}
+	if err = stateStore.Save(&state.Snapshot{
+		NodeID:           nodeID,
+		CurrentVersion:   activeVersion,
+		CurrentChecksum:  activeChecksum,
+		PagesDeployments: []state.PagesDeployment{},
+		LastError:        sticky,
+	}); err != nil {
+		t.Fatalf("failed to seed state: %v", err)
+	}
+
+	// Disk differs from stale heartbeat target so SyncOnce does not take the matching path.
+	manager := &fakeManager{currentChecksum: "disk-checksum-differs"}
+	service := New(client, manager, stateStore)
+
+	if err = service.SyncOnce(context.Background(), &protocol.ActiveConfigMeta{
+		Version:  "20260309-202",
+		Checksum: "checksum-stale-heartbeat",
+	}); err != nil {
+		t.Fatalf("SyncOnce failed: %v", err)
+	}
+
+	snapshot, err := stateStore.Load()
+	if err != nil {
+		t.Fatalf("failed to load state: %v", err)
+	}
+	if snapshot.LastError != "" {
+		t.Fatalf("expected sticky LastError to clear on applyIfNeeded state-match success, got %q", snapshot.LastError)
+	}
+	if client.fetchCalls != 1 {
+		t.Fatalf("expected one config fetch for stale heartbeat target, got %d", client.fetchCalls)
+	}
+	if len(manager.applyMainContents) != 0 {
+		t.Fatal("expected state-match path to skip OpenResty apply")
+	}
+}
@@ -122,6 +122,8 @@ func (s *Service) syncMismatchedChecksum(ctx context.Context, mode string, start
 		if !reconciled {
 			slog.Debug("skipping config fetch because state already records target version/checksum", "version", target.Version, "checksum", target.Checksum)
 		}
+		// Successful reconcile must clear sticky LastError so health events can resolve.
+		snapshot.LastError = ""
 		return s.stateStore.Save(snapshot)
 	}