net-controller: receive one-shot details and deliver HTTP polling commands

cmd/unbounded-net-controller/cluster_status.go

@@ -18,6 +18,7 @@ import (
 	"k8s.io/klog/v2"
 
 	"github.com/Azure/unbounded/internal/net/controller"
+	statusv1alpha1 "github.com/Azure/unbounded/internal/net/status/v1alpha1"
 	"github.com/Azure/unbounded/internal/version"
 )
 
@@ -236,6 +237,13 @@ func fetchClusterStatus(ctx context.Context, health *healthState, pullEnabled bo
 	}
 
 	cachedStatuses := health.statusCache.GetAll()
+	status.NodeOverviews = make(map[string]*statusv1alpha1.NodeStatusOverview)
+
+	for name, cached := range cachedStatuses {
+		if cached.Overview != nil {
+			status.NodeOverviews[name] = cached.Overview
+		}
+	}
 
 	type pullNode struct{ nodeName, nodeIP string }
 
@@ -371,6 +379,7 @@ func fetchClusterStatus(ctx context.Context, health *healthState, pullEnabled bo
 			} else if result.status != nil {
 				result.status.StatusSource = "pull"
 				cachedResults[result.nodeName] = *result.status
+				delete(status.NodeOverviews, result.nodeName)
 			}
 		}
 	}
@@ -447,6 +456,9 @@ func fetchClusterStatus(ctx context.Context, health *healthState, pullEnabled bo
 			if pubKey := node.Annotations[controller.WireGuardPubKeyAnnotation]; pubKey != "" {
 				if nodeStatus.NodeInfo.WireGuard == nil {
 					nodeStatus.NodeInfo.WireGuard = &WireGuardStatusInfo{}
+				} else {
+					wireguard := *nodeStatus.NodeInfo.WireGuard
+					nodeStatus.NodeInfo.WireGuard = &wireguard
 				}
 
 				nodeStatus.NodeInfo.WireGuard.PublicKey = pubKey
@@ -710,7 +722,6 @@ func fetchClusterStatus(ctx context.Context, health *healthState, pullEnabled bo
 	}
 
 	sort.Slice(status.Peerings, func(i, j int) bool { return status.Peerings[i].Name < status.Peerings[j].Name })
-	status.ConnectivityMatrix = buildConnectivityMatrix(status.Nodes, status.GatewayPools)
 	status.Problems = collectClusterProblems(status)
 
 	return status
@@ -809,6 +820,18 @@ func collectClusterProblems(status *ClusterStatusResponse) []StatusProblem {
 			appendProblem("node", nodeName, summary)
 		}
 
+		if overview := status.NodeOverviews[node.NodeInfo.Name]; overview != nil {
+			if overview.RouteMismatch {
+				appendProblem("node", nodeName, "Route next-hop mismatches (expected vs present)")
+			}
+
+			if unhealthy := overview.PeerCount - overview.HealthyPeers; unhealthy > 0 {
+				appendProblem("node", nodeName, fmt.Sprintf("%d peers are not healthy", unhealthy))
+			}
+
+			continue
+		}
+
 		if mismatchCount := routeMismatchCount(node); mismatchCount > 0 {
 			appendProblem("node", nodeName, fmt.Sprintf("%d route next-hop mismatches (expected vs present)", mismatchCount))
 		}
@@ -1038,163 +1061,3 @@ func latestNodeUpdateTime(node *corev1.Node) time.Time {
 
 	return latest
 }
-
-// buildConnectivityMatrix builds health check connectivity matrices from node peer data.
-func buildConnectivityMatrix(nodes []*NodeStatusResponse, gatewayPools []GatewayPoolStatus) map[string]*SiteMatrix {
-	siteNodes := make(map[string]map[string]bool)
-	nodePeers := make(map[string][]WireGuardPeerStatus)
-	nodeByName := make(map[string]*NodeStatusResponse)
-
-	for _, n := range nodes {
-		name := n.NodeInfo.Name
-
-		site := n.NodeInfo.SiteName
-		if name == "" || site == "" {
-			continue
-		}
-
-		nodeByName[name] = n
-
-		if siteNodes[site] == nil {
-			siteNodes[site] = make(map[string]bool)
-		}
-
-		siteNodes[site][name] = true
-
-		// Keep the immutable snapshot's slice; filter when reading rather
-		// than copying every peer, including for scopes above the size limit.
-		nodePeers[name] = n.Peers
-
-		for _, p := range n.Peers {
-			if p.PeerType == "gateway" && p.Name != "" && p.SiteName == site {
-				siteNodes[site][p.Name] = true
-			}
-		}
-	}
-
-	if len(siteNodes) == 0 {
-		siteNodes = make(map[string]map[string]bool)
-	}
-
-	result := make(map[string]*SiteMatrix)
-	selfMatrixStatusFromCNI := func(node *NodeStatusResponse) string {
-		if node.NodeInfo.WireGuard != nil && strings.TrimSpace(node.NodeInfo.WireGuard.Interface) != "" {
-			return "up"
-		}
-
-		return ""
-	}
-
-	buildScopeMatrix := func(nodeSet map[string]bool) *SiteMatrix {
-		if len(nodeSet) == 0 || len(nodeSet) > 100 {
-			return nil
-		}
-
-		nodeNames := make([]string, 0, len(nodeSet))
-		for name := range nodeSet {
-			nodeNames = append(nodeNames, name)
-		}
-
-		sort.Strings(nodeNames)
-
-		results := make(map[string]map[string]string)
-		for _, srcNode := range nodeNames {
-			results[srcNode] = make(map[string]string)
-			if node, ok := nodeByName[srcNode]; ok {
-				results[srcNode][srcNode] = selfMatrixStatusFromCNI(node)
-			}
-
-			for _, peer := range nodePeers[srcNode] {
-				if !isConnectivityMatrixPeer(peer) {
-					continue
-				}
-
-				tgtNode := peer.Name
-				if tgtNode == "" || tgtNode == srcNode || !nodeSet[tgtNode] {
-					continue
-				}
-
-				cellStatus := ""
-				if peer.HealthCheck != nil {
-					cellStatus = peer.HealthCheck.Status
-				} else if peer.PeerType == "gateway" && !peer.Tunnel.LastHandshake.IsZero() {
-					cellStatus = "up"
-				}
-
-				results[srcNode][tgtNode] = cellStatus
-			}
-		}
-
-		return &SiteMatrix{Nodes: nodeNames, Results: results}
-	}
-
-	for site, nodeSet := range siteNodes {
-		scopeMatrix := buildScopeMatrix(nodeSet)
-		if scopeMatrix != nil {
-			result[site] = scopeMatrix
-		}
-	}
-
-	for _, pool := range gatewayPools {
-		poolName := strings.TrimSpace(pool.Name)
-		if poolName == "" {
-			continue
-		}
-
-		poolNodeSet := make(map[string]bool)
-
-		for _, gatewayName := range pool.Gateways {
-			name := strings.TrimSpace(gatewayName)
-			if name == "" {
-				continue
-			}
-
-			poolNodeSet[name] = true
-			for _, peer := range nodePeers[name] {
-				if !isConnectivityMatrixPeer(peer) {
-					continue
-				}
-
-				peerName := strings.TrimSpace(peer.Name)
-				if peerName == "" {
-					continue
-				}
-
-				if _, ok := nodeByName[peerName]; ok {
-					poolNodeSet[peerName] = true
-				}
-			}
-
-			for srcNodeName, peers := range nodePeers {
-				for _, peer := range peers {
-					if !isConnectivityMatrixPeer(peer) {
-						continue
-					}
-
-					if strings.TrimSpace(peer.Name) == name {
-						if _, ok := nodeByName[srcNodeName]; ok {
-							poolNodeSet[srcNodeName] = true
-						}
-
-						break
-					}
-				}
-			}
-		}
-
-		scopeMatrix := buildScopeMatrix(poolNodeSet)
-		if scopeMatrix != nil {
-			result["pool:"+poolName] = scopeMatrix
-		}
-	}
-
-	if len(result) == 0 {
-		return nil
-	}
-
-	return result
-}
-
-func isConnectivityMatrixPeer(peer WireGuardPeerStatus) bool {
-	return peer.PeerType == "site" || peer.PeerType == "gateway"
-}

cmd/unbounded-net-controller/cluster_status_cache.go

@@ -5,6 +5,7 @@ package main
 
 import (
 	"context"
+	"maps"
 	"reflect"
 	"slices"
 	"sync"
@@ -15,6 +16,8 @@ import (
 
 	unboundednetv1alpha1 "github.com/Azure/unbounded/api/net/v1alpha1"
 	"github.com/Azure/unbounded/internal/net/controller"
+	statuspkg "github.com/Azure/unbounded/internal/net/status"
+	statusv1alpha1 "github.com/Azure/unbounded/internal/net/status/v1alpha1"
 )
 
 // ClusterStatusCache maintains a pre-built ClusterStatusResponse in memory,
@@ -115,6 +118,15 @@ func (c *ClusterStatusCache) Rebuild(ctx context.Context) {
 // PatchNode updates a single node's cached status in-place without a full
 // rebuild.
 func (c *ClusterStatusCache) PatchNode(nodeName string, nodeStatus NodeStatusResponse) {
+	c.patchNode(nodeName, nodeStatus, nil)
+}
+
+// PatchOverview updates metadata and observed facts without collecting details.
+func (c *ClusterStatusCache) PatchOverview(nodeName string, overview statusv1alpha1.NodeStatusOverview) {
+	c.patchNode(nodeName, statuspkg.OverviewMetadata(overview), &overview)
+}
+
+func (c *ClusterStatusCache) patchNode(nodeName string, nodeStatus NodeStatusResponse, overview *statusv1alpha1.NodeStatusOverview) {
 	now := time.Now()
 	nodeStatus.NodeInfo.ExternalIPs = c.resolveNodeExternalIPs(nodeName, now)
 
@@ -125,6 +137,16 @@ func (c *ClusterStatusCache) PatchNode(nodeName string, nodeStatus NodeStatusRes
 		return
 	}
 
+	if overview == nil {
+		delete(c.status.NodeOverviews, nodeName)
+	} else {
+		if c.status.NodeOverviews == nil {
+			c.status.NodeOverviews = make(map[string]*statusv1alpha1.NodeStatusOverview)
+		}
+
+		c.status.NodeOverviews[nodeName] = overview
+	}
+
 	if i, ok := c.nodeIndex[nodeName]; ok && i < len(c.status.Nodes) {
 		// Preserve controller-enriched fields across node-agent status updates.
 		existing := c.status.Nodes[i]
@@ -225,13 +247,21 @@ func (c *ClusterStatusCache) MarkFullRebuildNeeded() {
 	}
 }
 
-// Get returns the current pre-built status (read-locked, fast).
+// Get snapshots mutable containers; nested node data remains immutable and shared.
 // Returns nil if the status has not been built yet.
 func (c *ClusterStatusCache) Get() *ClusterStatusResponse {
 	c.mu.RLock()
 	defer c.mu.RUnlock()
 
-	return c.status
+	if c.status == nil {
+		return nil
+	}
+
+	snapshot := *c.status
+	snapshot.Nodes = slices.Clone(c.status.Nodes)
+	snapshot.NodeOverviews = maps.Clone(c.status.NodeOverviews)
+
+	return &snapshot
 }
 
 // GetSeq returns the current sequence number.