Merge pull request #4027 from cockroachdb/spencerkimball/remove-init-…

…step

Remove explicit "init" step to start a cluster

acceptance/cluster/localcluster.go

@@ -233,17 +233,10 @@ func (l *LocalCluster) initCluster() {
 		}
 	}
 	create := func() (*Container, error) {
-		var entrypoint []string
-		if *cockroachImage == builderImage {
-			entrypoint = append(entrypoint, "/"+filepath.Base(*cockroachBinary))
-		} else if *cockroachEntry != "" {
-			entrypoint = append(entrypoint, *cockroachEntry)
-		}
 		return createContainer(l, dockerclient.ContainerConfig{
 			Image:      *cockroachImage,
 			Volumes:    vols,
-			Entrypoint: entrypoint,
-			Cmd:        []string{"init", "--stores=ssd=" + dataStr(0, 0)},
+			Entrypoint: []string{"/bin/true"},
 		})
 	}
 	c, err := create()
@@ -351,11 +344,6 @@ func (l *LocalCluster) createNodeCerts() {
 }
 
 func (l *LocalCluster) startNode(i int) *Container {
-	gossipNodes := []string{}
-	for j := 0; j < l.numLocal; j++ {
-		gossipNodes = append(gossipNodes, fmt.Sprintf("%s:%d", nodeStr(j), cockroachPort))
-	}
-
 	var stores = "ssd=" + dataStr(i, 0)
 	for j := 1; j < l.numStores; j++ {
 		stores += ",ssd=" + dataStr(i, j)
@@ -367,9 +355,13 @@ func (l *LocalCluster) startNode(i int) *Container {
 		"--certs=/certs",
 		"--host=" + nodeStr(i),
 		"--port=" + fmt.Sprintf("%d", cockroachPort),
-		"--gossip=" + strings.Join(gossipNodes, ","),
 		"--scan-max-idle-time=200ms", // set low to speed up tests
 	}
+	// Append --join flag for all nodes except first.
+	if i > 0 {
+		cmd = append(cmd, fmt.Sprintf("--join=%s:%d", nodeStr(0), cockroachPort))
+	}
+
 	var locallogDir string
 	if len(l.logDir) > 0 {
 		dockerlogDir := "/logs/" + nodeStr(i)

cli/cert.go

@@ -17,8 +17,9 @@
 package cli
 
 import (
+	"fmt"
+
 	"github.com/cockroachdb/cockroach/security"
-	"github.com/cockroachdb/cockroach/util"
 
 	"github.com/spf13/cobra"
 )
@@ -44,7 +45,7 @@ individual files in the directory specified by --certs (required).
 // to their corresponding files.
 func runCreateCACert(cmd *cobra.Command, args []string) error {
 	if err := security.RunCreateCACert(context.Certs, keySize); err != nil {
-		return util.Errorf("failed to generate CA certificate: %s", err)
+		return fmt.Errorf("failed to generate CA certificate: %s", err)
 	}
 	return nil
 }
@@ -68,7 +69,7 @@ At least one host should be passed in (either IP address of dns name).
 // to their corresponding files.
 func runCreateNodeCert(cmd *cobra.Command, args []string) error {
 	if err := security.RunCreateNodeCert(context.Certs, keySize, args); err != nil {
-		return util.Errorf("failed to generate node certificate: %s", err)
+		return fmt.Errorf("failed to generate node certificate: %s", err)
 	}
 	return nil
 }
@@ -95,7 +96,7 @@ func runCreateClientCert(cmd *cobra.Command, args []string) error {
 		return errMissingParams
 	}
 	if err := security.RunCreateClientCert(context.Certs, keySize, args[0]); err != nil {
-		return util.Errorf("failed to generate clent certificate: %s", err)
+		return fmt.Errorf("failed to generate clent certificate: %s", err)
 	}
 	return nil
 }

cli/cli.go

@@ -57,7 +57,6 @@ var cockroachCmd = &cobra.Command{
 
 func init() {
 	cockroachCmd.AddCommand(
-		initCmd,
 		startCmd,
 		certCmd,
 		exterminateCmd,

cli/cli_test.go

@@ -654,8 +654,7 @@ func TestFlagUsage(t *testing.T) {
   cockroach [command]
 
 Available Commands:
-  init        init new Cockroach cluster
-  start       start a node by joining the gossip network
+  start       start a node
   cert        create ca, node, and client certs
   exterminate destroy all data held by the node
   quit        drain and shutdown node

cli/flags.go

@@ -84,19 +84,16 @@ error occurs in any statement, the command exits with a
 non-zero status code and further statements are not
 executed. Only the results of the first SQL statement in each
 positional argument are printed on the standard output.`),
-	"gossip": wrapText(`
-A comma-separated list of gossip addresses or resolvers for gossip
-bootstrap. Each item in the list has an optional type:
-[type=]<address>. An unspecified type means ip address or dns.
-Type is one of:`) + `
+	"join": wrapText(`
+A comma-separated list of addresses to use when a new node is joining
+an existing cluster. For the first node in a cluster, --join should
+NOT be specified. Each address in the list has an optional type:
+[type=]<address>. An unspecified type means ip address or dns.  Type
+is one of:
 - tcp: (default if type is omitted): plain ip address or hostname.
-- unix: unix socket
-- lb: RPC load balancer forwarding to an arbitrary node
 - http-lb: HTTP load balancer: we query
-  http(s)://<address>/_status/details/local
-- self: for single node systems, specify --gossip=self (the
-  <address> is omitted).
-`,
+           http(s)://<address>/_status/details/local
+`),
 	"host": wrapText(`
 Database server host.`),
 	"insecure": wrapText(`
@@ -193,14 +190,6 @@ func initFlags(ctx *Context) {
 		pf.Var(pflagValue{f.Value}, normalizeStdFlagName(f.Name), f.Usage)
 	})
 
-	{
-		f := initCmd.Flags()
-		f.StringVar(&ctx.Stores, "stores", ctx.Stores, usage("stores"))
-		if err := initCmd.MarkFlagRequired("stores"); err != nil {
-			panic(err)
-		}
-	}
-
 	{
 		f := startCmd.Flags()
 		f.BoolVar(&ctx.EphemeralSingleNode, "dev", ctx.EphemeralSingleNode, usage("dev"))
@@ -219,8 +208,8 @@ func initFlags(ctx *Context) {
 		f.StringVar(&ctx.Certs, "certs", ctx.Certs, usage("certs"))
 		f.BoolVar(&ctx.Insecure, "insecure", ctx.Insecure, usage("insecure"))
 
-		// Gossip flags.
-		f.StringVar(&ctx.GossipBootstrap, "gossip", ctx.GossipBootstrap, usage("gossip"))
+		// Cluster joining flags.
+		f.StringVar(&ctx.JoinUsing, "join", ctx.JoinUsing, usage("join"))
 
 		// KV flags.
 		f.BoolVar(&ctx.Linearizable, "linearizable", ctx.Linearizable, usage("linearizable"))
@@ -232,9 +221,6 @@ func initFlags(ctx *Context) {
 		f.DurationVar(&ctx.ScanMaxIdleTime, "scan-max-idle-time", ctx.ScanMaxIdleTime, usage("scan-max-idle-time"))
 		f.DurationVar(&ctx.TimeUntilStoreDead, "time-until-store-dead", ctx.TimeUntilStoreDead, usage("time-until-store-dead"))
 
-		if err := startCmd.MarkFlagRequired("gossip"); err != nil {
-			panic(err)
-		}
 		if err := startCmd.MarkFlagRequired("stores"); err != nil {
 			panic(err)
 		}

cli/start.go

@@ -34,7 +34,6 @@ import (
 	"github.com/cockroachdb/cockroach/util"
 	"github.com/cockroachdb/cockroach/util/log"
 	"github.com/cockroachdb/cockroach/util/stop"
-	"github.com/cockroachdb/cockroach/util/uuid"
 
 	"github.com/spf13/cobra"
 )
@@ -77,77 +76,29 @@ func getJSON(hostport, path string, v interface{}) error {
 	return util.GetJSON(httpClient, context.HTTPRequestScheme(), hostport, path, v)
 }
 
-// initCmd command initializes a new Cockroach cluster.
-var initCmd = &cobra.Command{
-	Use:   "init --stores=...",
-	Short: "init new Cockroach cluster",
-	Long: `
-Initialize a new Cockroach cluster using the --stores flag to specify one or
-more storage locations. The first of these storage locations is used to
-bootstrap the first replica of the first range. If any of the storage locations
-are already part of a pre-existing cluster, the bootstrap will fail.
-`,
-	Example:      `  cockroach init --stores=ssd=/mnt/ssd1,ssd=/mnt/ssd2`,
-	SilenceUsage: true,
-	RunE:         runInit,
-}
-
-// runInit initializes the engine based on the first
-// store. The bootstrap engine may not be an in-memory type.
-func runInit(_ *cobra.Command, _ []string) error {
-	stopper := stop.NewStopper()
-	defer stopper.Stop()
-
-	// Default user for servers.
-	context.User = security.NodeUser
-
-	if err := context.InitStores(stopper); err != nil {
-		return util.Errorf("failed to initialize stores: %s", err)
-	}
-	if len(context.Engines) > 1 {
-		return util.Errorf("cannot bootstrap to more than one store")
-	}
-
-	return initCluster(stopper)
-}
-
-func initCluster(stopper *stop.Stopper) error {
-	// Generate a new UUID for cluster ID and bootstrap the cluster.
-	clusterID := uuid.NewUUID4().String()
-	if _, err := server.BootstrapCluster(clusterID, context.Engines, stopper); err != nil {
-		return util.Errorf("unable to bootstrap cluster: %s", err)
-	}
-
-	log.Infof("cockroach cluster %s has been initialized", clusterID)
-	return nil
-}
-
-// startCmd command starts nodes by joining the gossip network.
+// startCmd starts a node by initializing the stores and joining
+// the cluster.
 var startCmd = &cobra.Command{
 	Use:   "start",
-	Short: "start a node by joining the gossip network",
+	Short: "start a node",
 	Long: `
-Start a Cockroach node by joining the gossip network and exporting key ranges
-stored on physical device(s). The gossip network is joined by contacting one or
-more well-known hosts specified by the --gossip flag. Every node should be run
-with the same list of bootstrap hosts to guarantee a connected network. An
-alternate approach is to use a single host for --gossip and round-robin DNS.
+Start a CockroachDB node, which will export data from one or more
+storage devices, specified via the --stores flag.
 
-Each node exports data from one or more physical devices. These devices are
-specified via the --stores flag. This is a comma-separated list of paths to
-storage directories or for in-memory stores, the number of bytes. Although the
-paths should be specified to correspond uniquely to physical devices, this
-requirement isn't strictly enforced. See the --stores flag help description for
-additional details.`,
-	Example:      `  cockroach start --certs=<dir> --gossip=host1:port1[,...] --stores=ssd=/mnt/ssd1,...`,
+If no cluster exists yet and this is the first node, no additional
+flags are required. If the cluster already exists, and this node is
+uninitialized, specify the --join flag to point to any healthy node
+(or list of nodes) already part of the cluster.
+`,
+	Example:      `  cockroach start --certs=<dir> --stores=ssd=/mnt/ssd1,... [--join=host:port,[host:port]]`,
 	SilenceUsage: true,
 	RunE:         runStart,
 }
 
 // runStart starts the cockroach node using --stores as the list of
-// storage devices ("stores") on this machine and --gossip as the list
-// of "well-known" hosts used to join this node to the cockroach
-// cluster via the gossip network.
+// storage devices ("stores") on this machine and --join as the list
+// of other active nodes used to join this node to the cockroach
+// cluster, if this is its first time connecting.
 func runStart(_ *cobra.Command, _ []string) error {
 	info := util.GetBuildInfo()
 	log.Infof("[build] %s @ %s (%s)", info.Tag, info.Time, info.Vers)
@@ -160,36 +111,25 @@ func runStart(_ *cobra.Command, _ []string) error {
 		// for the default cluster-wide zone config.
 		config.DefaultZoneConfig.ReplicaAttrs = []roachpb.Attributes{{}}
 		context.Stores = "mem=1073741824"
-		context.GossipBootstrap = server.SelfGossipAddr
 	}
 
 	stopper := stop.NewStopper()
 	if err := context.InitStores(stopper); err != nil {
-		return util.Errorf("failed to initialize stores: %s", err)
-	}
-
-	if context.EphemeralSingleNode {
-		// A separate stopper for bootstrapping so that we can properly shutdown
-		// all of the stores.
-		initStopper := stop.NewStopper()
-		if err := initCluster(initStopper); err != nil {
-			return err
-		}
-		initStopper.Stop()
+		return fmt.Errorf("failed to initialize stores: %s", err)
 	}
 
 	if err := context.InitNode(); err != nil {
-		return util.Errorf("failed to initialize node: %s", err)
+		return fmt.Errorf("failed to initialize node: %s", err)
 	}
 
 	log.Info("starting cockroach node")
 	s, err := server.NewServer(&context.Context, stopper)
 	if err != nil {
-		return util.Errorf("failed to start Cockroach server: %s", err)
+		return fmt.Errorf("failed to start Cockroach server: %s", err)
 	}
 
-	if err := s.Start(false); err != nil {
-		return util.Errorf("cockroach server exited with error: %s", err)
+	if err := s.Start(); err != nil {
+		return fmt.Errorf("cockroach server exited with error: %s", err)
 	}
 
 	signalCh := make(chan os.Signal, 1)

gossip/client.go

@@ -188,7 +188,7 @@ func (c *client) handleGossip(g *Gossip, call *netrpc.Call) error {
 		return util.Errorf("received forward from node %d to %d (%s)", reply.NodeID, reply.AlternateNodeID, reply.AlternateAddr)
 	}
 
-	// If we have the sentinel gossip, we're considered connected.
+	// If we have the sentinel gossip we're considered connected.
 	g.checkHasConnected()
 
 	// Check whether this outgoing client is duplicating work already