|
| 1 | +package server |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "encoding/json" |
| 6 | + "errors" |
| 7 | + "fmt" |
| 8 | + "net/http" |
| 9 | + "os" |
| 10 | + "time" |
| 11 | + |
| 12 | + v1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 13 | + "k8s.io/client-go/kubernetes" |
| 14 | + "k8s.io/klog/v2" |
| 15 | + |
| 16 | + "github.com/openkruise/agents/api/v1alpha1" |
| 17 | + "github.com/openkruise/agents/pkg/peers" |
| 18 | + "github.com/openkruise/agents/pkg/proxy" |
| 19 | + "github.com/openkruise/agents/pkg/sandbox-gateway/registry" |
| 20 | + "github.com/openkruise/agents/pkg/sandbox-manager/consts" |
| 21 | + "github.com/openkruise/agents/pkg/utils" |
| 22 | +) |
| 23 | + |
| 24 | +// Environment variable names for peer discovery |
| 25 | +const ( |
| 26 | + EnvNamespace = "PEER_NAMESPACE" |
| 27 | + EnvLabelSelector = "PEER_LABEL_SELECTOR" |
| 28 | +) |
| 29 | + |
| 30 | +const ( |
| 31 | + // MemberlistBindPort is the default port for memberlist gossip |
| 32 | + MemberlistBindPort = 7946 |
| 33 | +) |
| 34 | + |
| 35 | +// Server handles peer-to-peer communication for route synchronization |
| 36 | +type Server struct { |
| 37 | + httpServer *http.Server |
| 38 | + peerManager *peers.MemberlistPeers |
| 39 | + port int |
| 40 | + client kubernetes.Interface |
| 41 | +} |
| 42 | + |
| 43 | +// NewServer creates a new peer server |
| 44 | +func NewServer(client kubernetes.Interface, port int) *Server { |
| 45 | + if port == 0 { |
| 46 | + port = proxy.SystemPort |
| 47 | + } |
| 48 | + return &Server{ |
| 49 | + port: port, |
| 50 | + client: client, |
| 51 | + } |
| 52 | +} |
| 53 | + |
| 54 | +// Start starts the HTTP server for handling refresh requests from peers |
| 55 | +func (s *Server) Start(ctx context.Context) error { |
| 56 | + log := klog.FromContext(ctx) |
| 57 | + |
| 58 | + mux := http.NewServeMux() |
| 59 | + mux.HandleFunc(proxy.RefreshAPI, s.handleRefresh) |
| 60 | + |
| 61 | + s.httpServer = &http.Server{ |
| 62 | + Addr: fmt.Sprintf(":%d", s.port), |
| 63 | + Handler: mux, |
| 64 | + ReadHeaderTimeout: 5 * time.Second, |
| 65 | + } |
| 66 | + |
| 67 | + // Get node name from environment variables |
| 68 | + nodeName := os.Getenv("HOSTNAME") |
| 69 | + if nodeName == "" { |
| 70 | + nodeName = os.Getenv("POD_NAME") |
| 71 | + } |
| 72 | + if nodeName == "" { |
| 73 | + return fmt.Errorf("HOSTNAME or POD_NAME environment variable must be set") |
| 74 | + } |
| 75 | + |
| 76 | + // Get local IP |
| 77 | + localIP := utils.GetFirstNonLoopbackIP() |
| 78 | + if localIP == "" { |
| 79 | + return fmt.Errorf("failed to determine local IP") |
| 80 | + } |
| 81 | + |
| 82 | + // Get namespace and label selector from environment variables |
| 83 | + namespace := os.Getenv(EnvNamespace) |
| 84 | + labelSelector := os.Getenv(EnvLabelSelector) |
| 85 | + |
| 86 | + // Discover existing peers from Kubernetes API |
| 87 | + var existingPeers []string |
| 88 | + if s.client != nil && namespace != "" && labelSelector != "" { |
| 89 | + log.Info("discovering existing peers for memberlist join", "namespace", namespace, "selector", labelSelector) |
| 90 | + peerList, err := s.client.CoreV1().Pods(namespace).List(ctx, v1.ListOptions{ |
| 91 | + LabelSelector: labelSelector, |
| 92 | + }) |
| 93 | + if err != nil { |
| 94 | + log.Error(err, "failed to list peer pods, continuing without existing peers") |
| 95 | + } else { |
| 96 | + for _, peer := range peerList.Items { |
| 97 | + ip := peer.Status.PodIP |
| 98 | + if ip == "" || ip == localIP { |
| 99 | + continue |
| 100 | + } |
| 101 | + existingPeers = append(existingPeers, fmt.Sprintf("%s:%d", ip, MemberlistBindPort)) |
| 102 | + } |
| 103 | + log.Info("found existing peers for memberlist join", "count", len(existingPeers)) |
| 104 | + } |
| 105 | + } else { |
| 106 | + log.Info("skipping peer discovery: client not available or namespace/labelSelector not set") |
| 107 | + } |
| 108 | + |
| 109 | + s.peerManager = peers.NewMemberlistPeers(nodeName) |
| 110 | + |
| 111 | + if err := s.peerManager.Start(ctx, localIP, MemberlistBindPort, existingPeers); err != nil { |
| 112 | + return err |
| 113 | + } |
| 114 | + |
| 115 | + go func() { |
| 116 | + klog.InfoS("Starting sandbox-gateway peer server", "address", s.httpServer.Addr) |
| 117 | + if err := s.httpServer.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) { |
| 118 | + klog.ErrorS(err, "Peer server failed to start") |
| 119 | + } |
| 120 | + }() |
| 121 | + |
| 122 | + return nil |
| 123 | +} |
| 124 | + |
| 125 | +func (s *Server) Stop(ctx context.Context) error { |
| 126 | + var errs []error |
| 127 | + if s.httpServer != nil { |
| 128 | + if err := s.httpServer.Shutdown(ctx); err != nil { |
| 129 | + errs = append(errs, err) |
| 130 | + } |
| 131 | + } |
| 132 | + if s.peerManager != nil { |
| 133 | + if err := s.peerManager.Stop(); err != nil { |
| 134 | + errs = append(errs, err) |
| 135 | + } |
| 136 | + } |
| 137 | + if len(errs) > 0 { |
| 138 | + return errors.Join(errs...) |
| 139 | + } |
| 140 | + return nil |
| 141 | +} |
| 142 | + |
| 143 | +// handleRefresh handles the /refresh endpoint for route synchronization |
| 144 | +func (s *Server) handleRefresh(w http.ResponseWriter, r *http.Request) { |
| 145 | + if r.Method != http.MethodPost { |
| 146 | + http.Error(w, "Method not allowed", http.StatusMethodNotAllowed) |
| 147 | + return |
| 148 | + } |
| 149 | + |
| 150 | + ctx := r.Context() |
| 151 | + log := klog.FromContext(ctx) |
| 152 | + |
| 153 | + var route proxy.Route |
| 154 | + if err := json.NewDecoder(r.Body).Decode(&route); err != nil { |
| 155 | + log.Error(err, "Failed to decode refresh request") |
| 156 | + http.Error(w, fmt.Sprintf("Failed to decode request: %v", err), http.StatusBadRequest) |
| 157 | + return |
| 158 | + } |
| 159 | + |
| 160 | + log.V(consts.DebugLogLevel).Info("Received route refresh", "route", route) |
| 161 | + |
| 162 | + // Handle based on state |
| 163 | + if route.State == v1alpha1.SandboxStateRunning { |
| 164 | + // Update the route |
| 165 | + if registry.UpdateWithVersion(route.ID, route.IP, route.ResourceVersion) { |
| 166 | + log.Info("Route updated via refresh", "id", route.ID, "ip", route.IP) |
| 167 | + } else { |
| 168 | + log.V(consts.DebugLogLevel).Info("Route update skipped due to older resourceVersion", "id", route.ID) |
| 169 | + } |
| 170 | + } else { |
| 171 | + // Delete the route if the sandbox is dead |
| 172 | + registry.Delete(route.ID) |
| 173 | + } |
| 174 | + |
| 175 | + w.WriteHeader(http.StatusNoContent) |
| 176 | +} |
0 commit comments