fix: update conn derpmap every 5s in single tailnet (#9176)

coderd/apidoc/docs.go

@@ -11976,6 +11976,9 @@ const docTemplate = `{
                 "app_security_key": {
                     "type": "string"
                 },
+                "derp_map": {
+                    "$ref": "#/definitions/tailcfg.DERPMap"
+                },
                 "derp_mesh_key": {
                     "type": "string"
                 },

coderd/apidoc/swagger.json

@@ -10937,6 +10937,9 @@
         "app_security_key": {
           "type": "string"
         },
+        "derp_map": {
+          "$ref": "#/definitions/tailcfg.DERPMap"
+        },
         "derp_mesh_key": {
           "type": "string"
         },

coderd/coderd.go

@@ -404,7 +404,7 @@ func New(options *Options) *API {
 		api.agentProvider, err = NewServerTailnet(api.ctx,
 			options.Logger,
 			options.DERPServer,
-			options.BaseDERPMap,
+			api.DERPMap,
 			func(context.Context) (tailnet.MultiAgentConn, error) {
 				return (*api.TailnetCoordinator.Load()).ServeMultiAgent(uuid.New()), nil
 			},

coderd/tailnet.go

@@ -44,15 +44,16 @@ func NewServerTailnet(
 	ctx context.Context,
 	logger slog.Logger,
 	derpServer *derp.Server,
-	derpMap *tailcfg.DERPMap,
+	derpMapFn func() *tailcfg.DERPMap,
 	getMultiAgent func(context.Context) (tailnet.MultiAgentConn, error),
 	cache *wsconncache.Cache,
 	traceProvider trace.TracerProvider,
 ) (*ServerTailnet, error) {
 	logger = logger.Named("servertailnet")
+	originalDerpMap := derpMapFn()
 	conn, err := tailnet.NewConn(&tailnet.Options{
 		Addresses: []netip.Prefix{netip.PrefixFrom(tailnet.IP(), 128)},
-		DERPMap:   derpMap,
+		DERPMap:   originalDerpMap,
 		Logger:    logger,
 	})
 	if err != nil {
@@ -60,9 +61,32 @@ func NewServerTailnet(
 	}
 
 	serverCtx, cancel := context.WithCancel(ctx)
+	derpMapUpdaterClosed := make(chan struct{})
+	go func() {
+		defer close(derpMapUpdaterClosed)
+
+		ticker := time.NewTicker(5 * time.Second)
+		defer ticker.Stop()
+
+		for {
+			select {
+			case <-serverCtx.Done():
+				return
+			case <-ticker.C:
+			}
+
+			newDerpMap := derpMapFn()
+			if !tailnet.CompareDERPMaps(originalDerpMap, newDerpMap) {
+				conn.SetDERPMap(newDerpMap)
+				originalDerpMap = newDerpMap
+			}
+		}
+	}()
+
 	tn := &ServerTailnet{
 		ctx:                  serverCtx,
 		cancel:               cancel,
+		derpMapUpdaterClosed: derpMapUpdaterClosed,
 		logger:               logger,
 		tracer:               traceProvider.Tracer(tracing.TracerName),
 		conn:                 conn,
@@ -237,8 +261,9 @@ func (s *ServerTailnet) reinitCoordinator() {
 }
 
 type ServerTailnet struct {
-	ctx    context.Context
-	cancel func()
+	ctx                  context.Context
+	cancel               func()
+	derpMapUpdaterClosed chan struct{}
 
 	logger        slog.Logger
 	tracer        trace.Tracer
@@ -403,5 +428,6 @@ func (s *ServerTailnet) Close() error {
 	_ = s.cache.Close()
 	_ = s.conn.Close()
 	s.transport.CloseIdleConnections()
+	<-s.derpMapUpdaterClosed
 	return nil
 }

coderd/tailnet_test.go

@@ -15,6 +15,7 @@ import (
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"go.opentelemetry.io/otel/trace"
+	"tailscale.com/tailcfg"
 
 	"cdr.dev/slog"
 	"cdr.dev/slog/sloggers/slogtest"
@@ -230,7 +231,7 @@ func setupAgent(t *testing.T, agentAddresses []netip.Prefix) (uuid.UUID, agent.A
 		context.Background(),
 		logger,
 		derpServer,
-		manifest.DERPMap,
+		func() *tailcfg.DERPMap { return manifest.DERPMap },
 		func(context.Context) (tailnet.MultiAgentConn, error) { return coord.ServeMultiAgent(uuid.New()), nil },
 		cache,
 		trace.NewNoopTracerProvider(),