vttest: add gateway-initial-tablet-timeout flag to enable faster startup (#18887)

Signed-off-by: Armand Parajon <[email protected]>

Author: aparajon

go/cmd/vttestserver/cli/main.go

@@ -224,6 +224,8 @@ func New() (cmd *cobra.Command) {
 
 	utils.SetFlagDurationVar(cmd.Flags(), &config.VtgateTabletRefreshInterval, "tablet-refresh-interval", 10*time.Second, "Interval at which vtgate refreshes tablet information from topology server.")
 
+	utils.SetFlagDurationVar(cmd.Flags(), &config.VtgateGatewayInitialTabletTimeout, "gateway-initial-tablet-timeout", 30*time.Second, "At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type")
+
 	cmd.Flags().BoolVar(&doCreateTCPUser, "initialize-with-vt-dba-tcp", false, "If this flag is enabled, MySQL will be initialized with an additional user named vt_dba_tcp, who will have access via TCP/IP connection.")
 
 	utils.SetFlagBoolVar(cmd.Flags(), &config.NoScatter, "no-scatter", false, "when set to true, the planner will fail instead of producing a plan that includes scatter queries")

go/cmd/vttestserver/cli/main_test.go

@@ -254,6 +254,21 @@ func TestCanGetKeyspaces(t *testing.T) {
 	assertGetKeyspaces(ctx, t, clusterInstance)
 }
 
+func TestGatewayInitialTabletTimeout(t *testing.T) {
+	conf := config
+	defer resetConfig(conf)
+
+	// Start cluster with custom gateway tablet timeout and verify it starts successfully
+	cluster, err := startCluster("--gateway-initial-tablet-timeout=1s")
+	require.NoError(t, err)
+	defer cluster.TearDown()
+
+	// Verify the cluster is functional by getting keyspaces
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	assertGetKeyspaces(ctx, t, cluster)
+}
+
 func TestExternalTopoServerConsul(t *testing.T) {
 	conf := config
 	defer resetConfig(conf)

go/flags/endtoend/vtcombo.txt

@@ -132,6 +132,7 @@ Flags:
       --external-topo-server                                             Should vtcombo use an external topology server instead of starting its own in-memory topology server. If true, vtcombo will use the flags defined in topo/server.go to open topo server
       --foreign-key-mode string                                          This is to provide how to handle foreign key constraint in create/alter table. Valid values are: allow, disallow (default "allow")
       --gate-query-cache-memory int                                      gate server query cache size in bytes, maximum amount of memory to be cached. vtgate analyzes every incoming query and generate a query plan, these plans are being cached in a lru cache. This config controls the capacity of the lru cache. (default 33554432)
+      --gateway-initial-tablet-timeout duration                          At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type (default 30s)
       --gc-check-interval duration                                       Interval between garbage collection checks (default 1h0m0s)
       --gc-purge-check-interval duration                                 Interval between purge discovery checks (default 1m0s)
       --grpc-auth-mode string                                            Which auth plugin implementation to use (eg: static)

go/flags/endtoend/vttestserver.txt

@@ -43,6 +43,7 @@ Flags:
       --external-topo-implementation string                              the topology implementation to use for vtcombo process
       --extra-my-cnf string                                              extra files to add to the config, separated by ':'
       --foreign-key-mode string                                          This is to provide how to handle foreign key constraint in create/alter table. Valid values are: allow, disallow (default "allow")
+      --gateway-initial-tablet-timeout duration                          At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type (default 30s)
       --grpc-auth-mode string                                            Which auth plugin implementation to use (eg: static)
       --grpc-auth-mtls-allowed-substrings string                         List of substrings of at least one of the client certificate names (separated by colon).
       --grpc-auth-static-client-creds string                             When using grpc_static_auth in the server, this file provides the credentials to use to authenticate with server.

go/vt/vtgate/tabletgateway.go

@@ -68,16 +68,23 @@ var (
 	logCollations = logutil.NewThrottledLogger("CollationInconsistent", 1*time.Minute)
 )
 
+func registerTabletGatewayFlags(fs *pflag.FlagSet) {
+	utils.SetFlagStringVar(fs, &CellsToWatch, "cells-to-watch", "", "comma-separated list of cells for watching tablets")
+	utils.SetFlagDurationVar(fs, &initialTabletTimeout, "gateway-initial-tablet-timeout", 30*time.Second, "At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type")
+	fs.IntVar(&retryCount, "retry-count", 2, "retry count")
+	fs.BoolVar(&balancerEnabled, "enable-balancer", false, "(DEPRECATED: use --vtgate-balancer-mode instead) Enable the tablet balancer to evenly spread query load for a given tablet type")
+	fs.StringVar(&balancerModeFlag, "vtgate-balancer-mode", "", fmt.Sprintf("Tablet balancer mode (options: %s). Defaults to 'cell' which shuffles tablets in the local cell.", strings.Join(balancer.GetAvailableModeNames(), ", ")))
+	fs.StringSliceVar(&balancerVtgateCells, "balancer-vtgate-cells", []string{}, "Comma-separated list of cells that contain vttablets. For 'prefer-cell' mode, this is required. For 'random' mode, this is optional and filters tablets to those cells.")
+	fs.StringSliceVar(&balancerKeyspaces, "balancer-keyspaces", []string{}, "Comma-separated list of keyspaces for which to use the balancer (optional). If empty, applies to all keyspaces.")
+}
+
+func registerVtcomboTabletGatewayFlags(fs *pflag.FlagSet) {
+	utils.SetFlagDurationVar(fs, &initialTabletTimeout, "gateway-initial-tablet-timeout", 30*time.Second, "At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type")
+}
+
 func init() {
-	servenv.OnParseFor("vtgate", func(fs *pflag.FlagSet) {
-		utils.SetFlagStringVar(fs, &CellsToWatch, "cells-to-watch", "", "comma-separated list of cells for watching tablets")
-		utils.SetFlagDurationVar(fs, &initialTabletTimeout, "gateway-initial-tablet-timeout", 30*time.Second, "At startup, the tabletGateway will wait up to this duration to get at least one tablet per keyspace/shard/tablet type")
-		fs.IntVar(&retryCount, "retry-count", 2, "retry count")
-		fs.BoolVar(&balancerEnabled, "enable-balancer", false, "(DEPRECATED: use --vtgate-balancer-mode instead) Enable the tablet balancer to evenly spread query load for a given tablet type")
-		fs.StringVar(&balancerModeFlag, "vtgate-balancer-mode", "", fmt.Sprintf("Tablet balancer mode (options: %s). Defaults to 'cell' which shuffles tablets in the local cell.", strings.Join(balancer.GetAvailableModeNames(), ", ")))
-		fs.StringSliceVar(&balancerVtgateCells, "balancer-vtgate-cells", []string{}, "Comma-separated list of cells that contain vttablets. For 'prefer-cell' mode, this is required. For 'random' mode, this is optional and filters tablets to those cells.")
-		fs.StringSliceVar(&balancerKeyspaces, "balancer-keyspaces", []string{}, "Comma-separated list of keyspaces for which to use the balancer (optional). If empty, applies to all keyspaces.")
-	})
+	servenv.OnParseFor("vtgate", registerTabletGatewayFlags)
+	servenv.OnParseFor("vtcombo", registerVtcomboTabletGatewayFlags)
 }
 
 // TabletGateway implements the Gateway interface.

go/vt/vttest/local_cluster.go

@@ -160,6 +160,9 @@ type Config struct {
 
 	VtgateTabletRefreshInterval time.Duration
 
+	// Gateway initial tablet timeout - how long VTGate waits for tablets at startup
+	VtgateGatewayInitialTabletTimeout time.Duration
+
 	// Set the planner to fail on scatter queries
 	NoScatter bool
 }

go/vt/vttest/vtprocess.go

@@ -279,6 +279,14 @@ func VtcomboProcess(environment Environment, args *Config, mysql MySQLManager) (
 		vt.ExtraArgs = append(vt.ExtraArgs, fmt.Sprintf("--tablet-refresh-interval=%v", args.VtgateTabletRefreshInterval))
 	}
 
+	// If gateway initial tablet timeout is not defined then we will give it value of 30s (vtcombo default).
+	// Setting it to a lower value will reduce the time VTGate waits for tablets at startup.
+	if args.VtgateGatewayInitialTabletTimeout <= 0 {
+		vt.ExtraArgs = append(vt.ExtraArgs, fmt.Sprintf("--gateway-initial-tablet-timeout=%v", 30*time.Second))
+	} else {
+		vt.ExtraArgs = append(vt.ExtraArgs, fmt.Sprintf("--gateway-initial-tablet-timeout=%v", args.VtgateGatewayInitialTabletTimeout))
+	}
+
 	vt.ExtraArgs = append(vt.ExtraArgs, QueryServerArgs...)
 	vt.ExtraArgs = append(vt.ExtraArgs, environment.VtcomboArguments()...)