Cache remote clients in Connect

integration/teleterm_test.go

@@ -34,6 +34,7 @@ import (
 	"github.com/jonboulle/clockwork"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
+	"golang.org/x/sync/errgroup"
 	"google.golang.org/grpc"
 	"google.golang.org/grpc/credentials/insecure"
 
@@ -116,6 +117,12 @@ func TestTeleterm(t *testing.T) {
 		t.Parallel()
 		testDeleteConnectMyComputerNode(t, pack)
 	})
+
+	t.Run("TestClientCache", func(t *testing.T) {
+		t.Parallel()
+
+		testClientCache(t, pack, creds)
+	})
 }
 
 func testAddingRootCluster(t *testing.T, pack *dbhelpers.DatabasePack, creds *helpers.UserCreds) {
@@ -335,6 +342,79 @@ func testHeadlessWatcher(t *testing.T, pack *dbhelpers.DatabasePack, creds *help
 	)
 }
 
+func testClientCache(t *testing.T, pack *dbhelpers.DatabasePack, creds *helpers.UserCreds) {
+	ctx := context.Background()
+
+	tc := mustLogin(t, pack.Root.User.GetName(), pack, creds)
+
+	storageFakeClock := clockwork.NewFakeClockAt(time.Now())
+
+	storage, err := clusters.NewStorage(clusters.Config{
+		Dir:                tc.KeysDir,
+		Clock:              storageFakeClock,
+		InsecureSkipVerify: tc.InsecureSkipVerify,
+	})
+	require.NoError(t, err)
+
+	cluster, _, err := storage.Add(ctx, tc.WebProxyAddr)
+	require.NoError(t, err)
+
+	daemonService, err := daemon.New(daemon.Config{
+		Storage: storage,
+		CreateTshdEventsClientCredsFunc: func() (grpc.DialOption, error) {
+			return grpc.WithTransportCredentials(insecure.NewCredentials()), nil
+		},
+		KubeconfigsDir: t.TempDir(),
+		AgentsDir:      t.TempDir(),
+	})
+	require.NoError(t, err)
+	t.Cleanup(func() {
+		daemonService.Stop()
+	})
+
+	// Check if parallel calls trying to get a client will return the same one.
+	eg, egCtx := errgroup.WithContext(ctx)
+	blocker := make(chan struct{})
+	const concurrentCalls = 5
+	concurrentCallsForClient := make([]*client.ProxyClient, concurrentCalls)
+	for i := range concurrentCallsForClient {
+		client := &concurrentCallsForClient[i]
+		eg.Go(func() error {
+			<-blocker
+			c, err := daemonService.GetCachedClient(egCtx, cluster.URI)
+			*client = c
+			return err
+		})
+	}
+	// unblock the operation which is still in progress
+	close(blocker)
+	require.NoError(t, eg.Wait())
+	require.Subset(t, concurrentCallsForClient[:1], concurrentCallsForClient[1:])
+
+	// Since we have a client in the cache, it should be returned.
+	secondCallForClient, err := daemonService.GetCachedClient(ctx, cluster.URI)
+	require.NoError(t, err)
+	require.Equal(t, concurrentCallsForClient[0], secondCallForClient)
+
+	// Let's remove the client from the cache.
+	// The call to GetCachedClient will
+	// connect to proxy and return a new client.
+	err = daemonService.ClearCachedClientsForRoot(cluster.URI)
+	require.NoError(t, err)
+	thirdCallForClient, err := daemonService.GetCachedClient(ctx, cluster.URI)
+	require.NoError(t, err)
+	require.NotEqual(t, secondCallForClient, thirdCallForClient)
+
+	// After closing the client (from our or a remote side)
+	// it will be removed from the cache.
+	// The call to GetCachedClient will connect to proxy and return a new client.
+	err = thirdCallForClient.Close()
+	require.NoError(t, err)
+	fourthCallForClient, err := daemonService.GetCachedClient(ctx, cluster.URI)
+	require.NoError(t, err)
+	require.NotEqual(t, thirdCallForClient, fourthCallForClient)
+}
+
 func testCreateConnectMyComputerRole(t *testing.T, pack *dbhelpers.DatabasePack) {
 	systemUser, err := user.Current()
 	require.NoError(t, err)

lib/teleterm/apiserver/handler/handler_apps.go

@@ -34,7 +34,12 @@ func (s *Handler) GetApps(ctx context.Context, req *api.GetAppsRequest) (*api.Ge
 		return nil, trace.Wrap(err)
 	}
 
-	resp, err := cluster.GetApps(ctx, req)
+	proxyClient, err := s.DaemonService.GetCachedClient(ctx, cluster.URI)
+	if err != nil {
+		return nil, trace.Wrap(err)
+	}
+
+	resp, err := cluster.GetApps(ctx, proxyClient.CurrentCluster(), req)
 	if err != nil {
 		return nil, trace.Wrap(err)
 	}

lib/teleterm/apiserver/handler/handler_auth.go

@@ -37,6 +37,10 @@ func (s *Handler) Login(ctx context.Context, req *api.LoginRequest) (*api.EmptyR
 	// added by daemon.Service.ResolveClusterURI.
 	clusterClient.MFAPromptConstructor = nil
 
+	if err = s.DaemonService.ClearCachedClientsForRoot(cluster.URI); err != nil {
+		return nil, trace.Wrap(err)
+	}
+
 	if req.Params == nil {
 		return nil, trace.BadParameter("missing login parameters")
 	}
@@ -84,6 +88,10 @@ func (s *Handler) LoginPasswordless(stream api.TerminalService_LoginPasswordless
 	// daemon.Service.ResolveClusterURI.
 	clusterClient.MFAPromptConstructor = nil
 
+	if err := s.DaemonService.ClearCachedClientsForRoot(cluster.URI); err != nil {
+		return trace.Wrap(err)
+	}
+
 	// Start the prompt flow.
 	if err := cluster.PasswordlessLogin(stream.Context(), stream); err != nil {
 		return trace.Wrap(err)

lib/teleterm/apiserver/handler/handler_databases.go

@@ -28,14 +28,19 @@ import (
 	"github.com/gravitational/teleport/lib/teleterm/clusters"
 )
 
-// GetDatabases gets databses with filters and returns paginated results
+// GetDatabases gets databases with filters and returns paginated results
 func (s *Handler) GetDatabases(ctx context.Context, req *api.GetDatabasesRequest) (*api.GetDatabasesResponse, error) {
 	cluster, _, err := s.DaemonService.ResolveCluster(req.ClusterUri)
 	if err != nil {
 		return nil, trace.Wrap(err)
 	}
 
-	resp, err := cluster.GetDatabases(ctx, req)
+	proxyClient, err := s.DaemonService.GetCachedClient(ctx, cluster.URI)
+	if err != nil {
+		return nil, trace.Wrap(err)
+	}
+
+	resp, err := cluster.GetDatabases(ctx, proxyClient.CurrentCluster(), req)
 	if err != nil {
 		return nil, trace.Wrap(err)
 	}
@@ -63,7 +68,12 @@ func (s *Handler) ListDatabaseUsers(ctx context.Context, req *api.ListDatabaseUs
 		return nil, trace.Wrap(err)
 	}
 
-	dbUsers, err := cluster.GetAllowedDatabaseUsers(ctx, req.DbUri)
+	proxyClient, err := s.DaemonService.GetCachedClient(ctx, cluster.URI)
+	if err != nil {
+		return nil, trace.Wrap(err)
+	}
+
+	dbUsers, err := cluster.GetAllowedDatabaseUsers(ctx, proxyClient.CurrentCluster(), req.DbUri)
 	if err != nil {
 		return nil, trace.Wrap(err)
 	}

lib/teleterm/apiserver/middleware.go

@@ -20,11 +20,14 @@ package apiserver
 
 import (
 	"context"
+	"errors"
 
 	"github.com/gravitational/trace"
 	"github.com/gravitational/trace/trail"
 	"github.com/sirupsen/logrus"
 	"google.golang.org/grpc"
+
+	"github.com/gravitational/teleport/api/client"
 )
 
 // withErrorHandling is gRPC middleware that maps internal errors to proper gRPC error codes
@@ -38,6 +41,14 @@ func withErrorHandling(log logrus.FieldLogger) grpc.UnaryServerInterceptor {
 		resp, err := handler(ctx, req)
 		if err != nil {
 			log.WithError(err).Error("Request failed.")
+			// A stop gap solution that allows us to show a relogin modal when we
+			// receive an error from the server saying that the cert is expired.
+			// Read more: https://github.com/gravitational/teleport/pull/38202#discussion_r1497181659
+			// TODO(gzdunek): fix when addressing https://github.com/gravitational/teleport/issues/32550
+			if errors.Is(err, client.ErrClientCredentialsHaveExpired) {
+				return resp, trail.ToGRPC(err)
+			}
+
 			// do not return a full error stack on access denied errors
 			if trace.IsAccessDenied(err) {
 				return resp, trail.ToGRPC(trace.AccessDenied("access denied"))

lib/teleterm/clusters/cluster.go

@@ -83,7 +83,7 @@ func (c *Cluster) Connected() bool {
 // GetWithDetails makes requests to the auth server to return details of the current
 // Cluster that cannot be found on the disk only, including details about the user
 // and enabled enterprise features. This method requires a valid cert.
-func (c *Cluster) GetWithDetails(ctx context.Context) (*ClusterWithDetails, error) {
+func (c *Cluster) GetWithDetails(ctx context.Context, authClient auth.ClientI) (*ClusterWithDetails, error) {
 	var (
 		authPingResponse proto.PingResponse
 		caps             *types.AccessCapabilities
@@ -97,20 +97,8 @@ func (c *Cluster) GetWithDetails(ctx context.Context) (*ClusterWithDetails, erro
 		return nil, trace.Wrap(err)
 	}
 
+	//TODO(gzdunek): These calls should be done in parallel.
 	err = AddMetadataToRetryableError(ctx, func() error {
-		//nolint:staticcheck // SA1019. TODO(tross) update to use ClusterClient
-		proxyClient, err := c.clusterClient.ConnectToProxy(ctx)
-		if err != nil {
-			return trace.Wrap(err)
-		}
-		defer proxyClient.Close()
-
-		authClient, err := proxyClient.ConnectToCluster(ctx, c.clusterClient.SiteName)
-		if err != nil {
-			return trace.Wrap(err)
-		}
-		defer authClient.Close()
-
 		authPingResponse, err = authClient.Ping(ctx)
 		if err != nil {
 			return trace.Wrap(err)
@@ -218,12 +206,10 @@ func (c *Cluster) GetRoles(ctx context.Context) ([]*types.Role, error) {
 }
 
 // GetRequestableRoles returns the requestable roles for the currently logged-in user
-func (c *Cluster) GetRequestableRoles(ctx context.Context, req *api.GetRequestableRolesRequest) (*types.AccessCapabilities, error) {
+func (c *Cluster) GetRequestableRoles(ctx context.Context, req *api.GetRequestableRolesRequest, authClient auth.ClientI) (*types.AccessCapabilities, error) {
 	var (
-		authClient  auth.ClientI
-		proxyClient *client.ProxyClient
-		err         error
-		response    *types.AccessCapabilities
+		err      error
+		response *types.AccessCapabilities
 	)
 
 	resourceIds := make([]types.ResourceID, 0, len(req.GetResourceIds()))
@@ -237,19 +223,6 @@ func (c *Cluster) GetRequestableRoles(ctx context.Context, req *api.GetRequestab
 	}
 
 	err = AddMetadataToRetryableError(ctx, func() error {
-		//nolint:staticcheck // SA1019. TODO(tross) update to use ClusterClient
-		proxyClient, err = c.clusterClient.ConnectToProxy(ctx)
-		if err != nil {
-			return trace.Wrap(err)
-		}
-		defer proxyClient.Close()
-
-		authClient, err = proxyClient.ConnectToCluster(ctx, c.clusterClient.SiteName)
-		if err != nil {
-			return trace.Wrap(err)
-		}
-		defer authClient.Close()
-
 		response, err = authClient.GetAccessCapabilities(ctx, types.AccessCapabilitiesRequest{
 			ResourceIDs:      resourceIds,
 			RequestableRoles: true,