feat(api): Enhance NodeDiscover with subnet auto-detection and discovery cancellation

- Updated NodeDiscover to accept an optional subnet parameter, with auto-detection of local networks if none is provided. - Removed support for IP list format in NodeDiscover request body. - Implemented discovery cancellation functionality with NodeDiscoveryCancel endpoint. - Improved error handling and response messages for better clarity. feat(cluster): Add operation tracking for cluster bootstrap process - Integrated operations manager into cluster manager for tracking bootstrap progress. - Refactored Bootstrap method to run asynchronously with detailed progress updates. - Added methods to wait for various bootstrap steps (etcd health, VIP assignment, control plane readiness, etc.). fix(discovery): Optimize node discovery process and improve maintenance mode detection - Enhanced node discovery to run in parallel with a semaphore to limit concurrent scans. - Updated probeNode to detect maintenance mode more reliably. - Added functions to expand CIDR notation into individual IP addresses and retrieve local network interfaces. refactor(node): Update node manager to handle instance-specific configurations - Modified NewManager to accept instanceName for tailored talosconfig usage. - Improved hardware detection logic to handle maintenance mode scenarios. feat(operations): Implement detailed bootstrap progress tracking - Introduced BootstrapProgress struct to track and report the status of bootstrap operations. - Updated operation management to include bootstrap-specific details. fix(tools): Improve talosctl command execution with context and error handling - Added context with timeout to talosctl commands to prevent hanging on unreachable nodes. - Enhanced error handling for version retrieval in maintenance mode.
2025-11-04 17:16:16 +00:00
parent 005dc30aa5
commit 7cd434aabf
9 changed files with 623 additions and 148 deletions
--- a/internal/api/v1/handlers.go
+++ b/internal/api/v1/handlers.go
@@ -30,6 +30,7 @@ type API struct {
 	context     *context.Manager
 	instance    *instance.Manager
 	dnsmasq     *dnsmasq.ConfigGenerator
+	opsMgr      *operations.Manager     // Operations manager
 	broadcaster *operations.Broadcaster // SSE broadcaster for operation output
 }

@@ -57,6 +58,7 @@ func NewAPI(dataDir, appsDir string) (*API, error) {
 		context:     context.NewManager(dataDir),
 		instance:    instance.NewManager(dataDir),
 		dnsmasq:     dnsmasq.NewConfigGenerator(dnsmasqConfigPath),
+		opsMgr:      operations.NewManager(dataDir),
 		broadcaster: operations.NewBroadcaster(),
 	}, nil
 }
@@ -85,6 +87,7 @@ func (api *API) RegisterRoutes(r *mux.Router) {
 	r.HandleFunc("/api/v1/instances/{name}/nodes/discover", api.NodeDiscover).Methods("POST")
 	r.HandleFunc("/api/v1/instances/{name}/nodes/detect", api.NodeDetect).Methods("POST")
 	r.HandleFunc("/api/v1/instances/{name}/discovery", api.NodeDiscoveryStatus).Methods("GET")
+	r.HandleFunc("/api/v1/instances/{name}/discovery/cancel", api.NodeDiscoveryCancel).Methods("POST")
 	r.HandleFunc("/api/v1/instances/{name}/nodes/hardware/{ip}", api.NodeHardware).Methods("GET")
 	r.HandleFunc("/api/v1/instances/{name}/nodes/fetch-templates", api.NodeFetchTemplates).Methods("POST")
 	r.HandleFunc("/api/v1/instances/{name}/nodes", api.NodeAdd).Methods("POST")
--- a/internal/api/v1/handlers_cluster.go
+++ b/internal/api/v1/handlers_cluster.go
@@ -46,15 +46,15 @@ func (api *API) ClusterGenerateConfig(w http.ResponseWriter, r *http.Request) {
 	}

 	// Create cluster config
-	config := cluster.ClusterConfig{
+	clusterConfig := cluster.ClusterConfig{
 		ClusterName: clusterName,
 		VIP:         vip,
 		Version:     version,
 	}

 	// Generate configuration
-	clusterMgr := cluster.NewManager(api.dataDir)
-	if err := clusterMgr.GenerateConfig(instanceName, &config); err != nil {
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
+	if err := clusterMgr.GenerateConfig(instanceName, &clusterConfig); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to generate config: %v", err))
 		return
 	}
@@ -90,26 +90,14 @@ func (api *API) ClusterBootstrap(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	// Start bootstrap operation
-	opsMgr := operations.NewManager(api.dataDir)
-	opID, err := opsMgr.Start(instanceName, "bootstrap", req.Node)
+	// Bootstrap with progress tracking
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
+	opID, err := clusterMgr.Bootstrap(instanceName, req.Node)
 	if err != nil {
-		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to start operation: %v", err))
+		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to start bootstrap: %v", err))
 		return
 	}

-	// Bootstrap in background
-	go func() {
-		clusterMgr := cluster.NewManager(api.dataDir)
-		_ = opsMgr.UpdateStatus(instanceName, opID, "running")
-
-		if err := clusterMgr.Bootstrap(instanceName, req.Node); err != nil {
-			_ = opsMgr.Update(instanceName, opID, "failed", err.Error(), 0)
-		} else {
-			_ = opsMgr.Update(instanceName, opID, "completed", "Bootstrap completed", 100)
-		}
-	}()
-
 	respondJSON(w, http.StatusAccepted, map[string]string{
 		"operation_id": opID,
 		"message":      "Bootstrap initiated",
@@ -138,7 +126,7 @@ func (api *API) ClusterConfigureEndpoints(w http.ResponseWriter, r *http.Request
 	}

 	// Configure endpoints
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	if err := clusterMgr.ConfigureEndpoints(instanceName, req.IncludeNodes); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to configure endpoints: %v", err))
 		return
@@ -161,7 +149,7 @@ func (api *API) ClusterGetStatus(w http.ResponseWriter, r *http.Request) {
 	}

 	// Get status
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	status, err := clusterMgr.GetStatus(instanceName)
 	if err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to get status: %v", err))
@@ -183,7 +171,7 @@ func (api *API) ClusterHealth(w http.ResponseWriter, r *http.Request) {
 	}

 	// Get health checks
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	checks, err := clusterMgr.Health(instanceName)
 	if err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to get health: %v", err))
@@ -219,7 +207,7 @@ func (api *API) ClusterGetKubeconfig(w http.ResponseWriter, r *http.Request) {
 	}

 	// Get kubeconfig
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	kubeconfig, err := clusterMgr.GetKubeconfig(instanceName)
 	if err != nil {
 		respondError(w, http.StatusNotFound, fmt.Sprintf("Kubeconfig not found: %v", err))
@@ -243,7 +231,7 @@ func (api *API) ClusterGenerateKubeconfig(w http.ResponseWriter, r *http.Request
 	}

 	// Regenerate kubeconfig from cluster
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	if err := clusterMgr.RegenerateKubeconfig(instanceName); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to generate kubeconfig: %v", err))
 		return
@@ -266,7 +254,7 @@ func (api *API) ClusterGetTalosconfig(w http.ResponseWriter, r *http.Request) {
 	}

 	// Get talosconfig
-	clusterMgr := cluster.NewManager(api.dataDir)
+	clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 	talosconfig, err := clusterMgr.GetTalosconfig(instanceName)
 	if err != nil {
 		respondError(w, http.StatusNotFound, fmt.Sprintf("Talosconfig not found: %v", err))
@@ -314,7 +302,7 @@ func (api *API) ClusterReset(w http.ResponseWriter, r *http.Request) {

 	// Reset in background
 	go func() {
-		clusterMgr := cluster.NewManager(api.dataDir)
+		clusterMgr := cluster.NewManager(api.dataDir, api.opsMgr)
 		_ = opsMgr.UpdateStatus(instanceName, opID, "running")

 		if err := clusterMgr.Reset(instanceName, req.Confirm); err != nil {
--- a/internal/api/v1/handlers_node.go
+++ b/internal/api/v1/handlers_node.go
@@ -12,6 +12,7 @@ import (
 )

 // NodeDiscover initiates node discovery
+// Accepts optional subnet parameter. If no subnet provided, auto-detects local networks.
 func (api *API) NodeDiscover(w http.ResponseWriter, r *http.Request) {
 	vars := mux.Vars(r)
 	instanceName := vars["name"]
@@ -22,10 +23,9 @@ func (api *API) NodeDiscover(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	// Parse request body - support both subnet and ip_list formats
+	// Parse request body - only subnet is supported
 	var req struct {
-		Subnet string   `json:"subnet"`
-		IPList []string `json:"ip_list"`
+		Subnet string `json:"subnet,omitempty"`
 	}

 	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
@@ -33,16 +33,38 @@ func (api *API) NodeDiscover(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	// If subnet provided, use it as a single "IP" for discovery
-	// The discovery manager will scan this subnet
+	// Build IP list
 	var ipList []string
+	var err error
+
 	if req.Subnet != "" {
-		ipList = []string{req.Subnet}
-	} else if len(req.IPList) > 0 {
-		ipList = req.IPList
+		// Expand provided CIDR notation to individual IPs
+		ipList, err = discovery.ExpandSubnet(req.Subnet)
+		if err != nil {
+			respondError(w, http.StatusBadRequest, fmt.Sprintf("Invalid subnet: %v", err))
+			return
+		}
 	} else {
-		respondError(w, http.StatusBadRequest, "subnet or ip_list is required")
-		return
+		// Auto-detect: Get local networks when no subnet provided
+		networks, err := discovery.GetLocalNetworks()
+		if err != nil {
+			respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to detect local networks: %v", err))
+			return
+		}
+
+		if len(networks) == 0 {
+			respondError(w, http.StatusNotFound, "No local networks found")
+			return
+		}
+
+		// Expand all detected networks
+		for _, network := range networks {
+			ips, err := discovery.ExpandSubnet(network)
+			if err != nil {
+				continue // Skip invalid networks
+			}
+			ipList = append(ipList, ips...)
+		}
 	}

 	// Start discovery
@@ -52,9 +74,10 @@ func (api *API) NodeDiscover(w http.ResponseWriter, r *http.Request) {
 		return
 	}

-	respondJSON(w, http.StatusAccepted, map[string]string{
-		"message": "Discovery started",
-		"status":  "running",
+	respondJSON(w, http.StatusAccepted, map[string]interface{}{
+		"message":     "Discovery started",
+		"status":      "running",
+		"ips_to_scan": len(ipList),
 	})
 }

@@ -92,7 +115,7 @@ func (api *API) NodeHardware(w http.ResponseWriter, r *http.Request) {
 	}

 	// Detect hardware
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	hwInfo, err := nodeMgr.DetectHardware(nodeIP)
 	if err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to detect hardware: %v", err))
@@ -103,6 +126,7 @@ func (api *API) NodeHardware(w http.ResponseWriter, r *http.Request) {
 }

 // NodeDetect detects hardware on a single node (POST with IP in body)
+// IP address is required.
 func (api *API) NodeDetect(w http.ResponseWriter, r *http.Request) {
 	vars := mux.Vars(r)
 	instanceName := vars["name"]
@@ -123,13 +147,14 @@ func (api *API) NodeDetect(w http.ResponseWriter, r *http.Request) {
 		return
 	}

+	// Validate IP is provided
 	if req.IP == "" {
-		respondError(w, http.StatusBadRequest, "ip is required")
+		respondError(w, http.StatusBadRequest, "IP address is required")
 		return
 	}

-	// Detect hardware
-	nodeMgr := node.NewManager(api.dataDir)
+	// Detect hardware for specific IP
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	hwInfo, err := nodeMgr.DetectHardware(req.IP)
 	if err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to detect hardware: %v", err))
@@ -158,7 +183,7 @@ func (api *API) NodeAdd(w http.ResponseWriter, r *http.Request) {
 	}

 	// Add node
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	if err := nodeMgr.Add(instanceName, &nodeData); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to add node: %v", err))
 		return
@@ -182,7 +207,7 @@ func (api *API) NodeList(w http.ResponseWriter, r *http.Request) {
 	}

 	// List nodes
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	nodes, err := nodeMgr.List(instanceName)
 	if err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to list nodes: %v", err))
@@ -207,7 +232,7 @@ func (api *API) NodeGet(w http.ResponseWriter, r *http.Request) {
 	}

 	// Get node
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	nodeData, err := nodeMgr.Get(instanceName, nodeIdentifier)
 	if err != nil {
 		respondError(w, http.StatusNotFound, fmt.Sprintf("Node not found: %v", err))
@@ -233,7 +258,7 @@ func (api *API) NodeApply(w http.ResponseWriter, r *http.Request) {
 	opts := node.ApplyOptions{}

 	// Apply node configuration
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	if err := nodeMgr.Apply(instanceName, nodeIdentifier, opts); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to apply node configuration: %v", err))
 		return
@@ -265,7 +290,7 @@ func (api *API) NodeUpdate(w http.ResponseWriter, r *http.Request) {
 	}

 	// Update node
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	if err := nodeMgr.Update(instanceName, nodeIdentifier, updates); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to update node: %v", err))
 		return
@@ -289,7 +314,7 @@ func (api *API) NodeFetchTemplates(w http.ResponseWriter, r *http.Request) {
 	}

 	// Fetch templates
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	if err := nodeMgr.FetchTemplates(instanceName); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to fetch templates: %v", err))
 		return
@@ -313,7 +338,7 @@ func (api *API) NodeDelete(w http.ResponseWriter, r *http.Request) {
 	}

 	// Delete node
-	nodeMgr := node.NewManager(api.dataDir)
+	nodeMgr := node.NewManager(api.dataDir, instanceName)
 	if err := nodeMgr.Delete(instanceName, nodeIdentifier); err != nil {
 		respondError(w, http.StatusInternalServerError, fmt.Sprintf("Failed to delete node: %v", err))
 		return
@@ -323,3 +348,26 @@ func (api *API) NodeDelete(w http.ResponseWriter, r *http.Request) {
 		"message": "Node deleted successfully",
 	})
 }
+
+// NodeDiscoveryCancel cancels an in-progress discovery operation
+func (api *API) NodeDiscoveryCancel(w http.ResponseWriter, r *http.Request) {
+	vars := mux.Vars(r)
+	instanceName := vars["name"]
+
+	// Validate instance exists
+	if err := api.instance.ValidateInstance(instanceName); err != nil {
+		respondError(w, http.StatusNotFound, fmt.Sprintf("Instance not found: %v", err))
+		return
+	}
+
+	// Cancel discovery
+	discoveryMgr := discovery.NewManager(api.dataDir, instanceName)
+	if err := discoveryMgr.CancelDiscovery(instanceName); err != nil {
+		respondError(w, http.StatusBadRequest, fmt.Sprintf("Failed to cancel discovery: %v", err))
+		return
+	}
+
+	respondJSON(w, http.StatusOK, map[string]string{
+		"message": "Discovery cancelled successfully",
+	})
+}