Wiz add subnet only, with validator's subset (ava-labs#1314)

* add flows to only create a devnet and only add a subnet into a devnet

* nit

* add validators filtering to node sync, node validate subnet, node devnet
wiz

* filter health check only ask for P chain

* fix panic

* address PR comments

cmd/nodecmd/create.go

@@ -16,6 +16,7 @@ import (
 	"strings"
 	"sync"
 
+	"github.com/ava-labs/avalanche-cli/cmd/flags"
 	"github.com/ava-labs/avalanche-cli/cmd/subnetcmd"
 	"github.com/ava-labs/avalanche-cli/pkg/ansible"
 	"github.com/ava-labs/avalanche-cli/pkg/ssh"
@@ -31,6 +32,7 @@ import (
 
 	"github.com/ava-labs/avalanche-cli/pkg/ux"
 	"github.com/spf13/cobra"
+	"golang.org/x/mod/semver"
 )
 
 const (
@@ -50,6 +52,7 @@ var (
 	numNodes                        []int
 	nodeType                        string
 	useLatestAvalanchegoVersion     bool
+	useCustomAvalanchegoVersion     string
 	useAvalanchegoVersionFromSubnet string
 	cmdLineGCPCredentialsPath       string
 	cmdLineGCPProjectName           string
@@ -86,6 +89,7 @@ will apply to all nodes in the cluster`,
 	cmd.Flags().IntSliceVar(&numNodes, "num-nodes", []int{}, "number of nodes to create per region(s). Use comma to separate multiple numbers for each region in the same order as --region flag")
 	cmd.Flags().StringVar(&nodeType, "node-type", "", "cloud instance type. Use 'default' to use recommended default instance type")
 	cmd.Flags().BoolVar(&useLatestAvalanchegoVersion, "latest-avalanchego-version", false, "install latest avalanchego version on node/s")
+	cmd.Flags().StringVar(&useCustomAvalanchegoVersion, "custom-avalanchego-version", "", "install given avalanchego version on node/s")
 	cmd.Flags().StringVar(&useAvalanchegoVersionFromSubnet, "avalanchego-version-from-subnet", "", "install latest avalanchego version, that is compatible with the given subnet, on node/s")
 	cmd.Flags().StringVar(&cmdLineGCPCredentialsPath, "gcp-credentials", "", "use given GCP credentials")
 	cmd.Flags().StringVar(&cmdLineGCPProjectName, "gcp-project", "", "use given GCP project")
@@ -97,8 +101,8 @@ will apply to all nodes in the cluster`,
 }
 
 func preCreateChecks() error {
-	if useLatestAvalanchegoVersion && useAvalanchegoVersionFromSubnet != "" {
-		return fmt.Errorf("could not use both latest avalanchego version and avalanchego version based on given subnet")
+	if !flags.EnsureMutuallyExclusive([]bool{useLatestAvalanchegoVersion, useAvalanchegoVersionFromSubnet != "", useCustomAvalanchegoVersion != ""}) {
+		return fmt.Errorf("latest avalanchego version, custom avalanchego version and avalanchego version based on given subnet, are mutually exclusive options")
 	}
 	if useAWS && useGCP {
 		return fmt.Errorf("could not use both AWS and GCP cloud options")
@@ -498,6 +502,11 @@ func getAvalancheGoVersion() (string, error) {
 	subnet := ""
 	if useLatestAvalanchegoVersion { //nolint: gocritic
 		version = "latest"
+	} else if useCustomAvalanchegoVersion != "" {
+		if !semver.IsValid(useCustomAvalanchegoVersion) {
+			return "", errors.New("custom avalanchego version must be a legal semantic version (ex: v1.1.1)")
+		}
+		version = useCustomAvalanchegoVersion
 	} else if useAvalanchegoVersionFromSubnet != "" {
 		subnet = useAvalanchegoVersionFromSubnet
 	} else {

cmd/nodecmd/stop.go

@@ -156,7 +156,7 @@ func stopNodes(_ *cobra.Command, args []string) error {
 	if len(nodeErrors) > 0 {
 		ux.Logger.PrintToUser("Failed nodes: ")
 		for node, nodeErr := range nodeErrors {
-			if strings.Contains(err.Error(), constants.ErrReleasingGCPStaticIP) {
+			if strings.Contains(nodeErr.Error(), constants.ErrReleasingGCPStaticIP) {
 				ux.Logger.PrintToUser(fmt.Sprintf("Node is stopped, but failed to release static ip address for node %s due to %s", node, nodeErr))
 			} else {
 				ux.Logger.PrintToUser(fmt.Sprintf("Failed to stop node %s due to %s", node, nodeErr))

cmd/nodecmd/sync.go

@@ -35,6 +35,8 @@ You can check the subnet bootstrap status by calling avalanche node status <clus
 		RunE:         syncSubnet,
 	}
 
+	cmd.Flags().StringSliceVar(&validators, "validators", []string{}, "sync subnet into given comma separated list of validators. defaults to all cluster nodes")
+
 	return cmd
 }
 
@@ -129,6 +131,12 @@ func syncSubnet(_ *cobra.Command, args []string) error {
 	if err != nil {
 		return err
 	}
+	if len(validators) != 0 {
+		hosts, err = filterHosts(hosts, validators)
+		if err != nil {
+			return err
+		}
+	}
 	defer disconnectHosts(hosts)
 	notBootstrappedNodes, err := checkHostsAreBootstrapped(hosts)
 	if err != nil {

cmd/nodecmd/validateSubnet.go

@@ -57,6 +57,8 @@ You can check the subnet sync status by calling avalanche node status <clusterNa
 	cmd.Flags().StringVar(&startTimeStr, "start-time", "", "UTC start time when this validator starts validating, in 'YYYY-MM-DD HH:MM:SS' format")
 	cmd.Flags().BoolVar(&defaultValidatorParams, "default-validator-params", false, "use default weight/start/duration params for subnet validator")
 
+	cmd.Flags().StringSliceVar(&validators, "validators", []string{}, "validate subnet for the given comma separated list of validators. defaults to all cluster nodes")
+
 	return cmd
 }
 
@@ -159,6 +161,12 @@ func validateSubnet(_ *cobra.Command, args []string) error {
 	if err != nil {
 		return err
 	}
+	if len(validators) != 0 {
+		hosts, err = filterHosts(hosts, validators)
+		if err != nil {
+			return err
+		}
+	}
 	defer disconnectHosts(hosts)
 
 	nodeIDMap, failedNodesMap := getNodeIDs(hosts)

cmd/nodecmd/wiz.go

@@ -17,8 +17,10 @@ import (
 
 	"github.com/ava-labs/avalanchego/ids"
 	"github.com/ava-labs/avalanchego/utils/logging"
+	"github.com/ava-labs/avalanchego/utils/set"
 	"github.com/ava-labs/avalanchego/vms/platformvm/status"
 	"github.com/spf13/cobra"
+	"golang.org/x/exp/slices"
 )
 
 const (
@@ -43,6 +45,7 @@ var (
 	nodeConf            string
 	subnetConf          string
 	chainConf           string
+	validators          []string
 )
 
 func newWizCmd() *cobra.Command {
@@ -54,7 +57,7 @@ func newWizCmd() *cobra.Command {
 The node wiz command creates a devnet and deploys, sync and validate a subnet into it. It creates the subnet if so needed.
 `,
 		SilenceUsage: true,
-		Args:         cobra.ExactArgs(2),
+		Args:         cobra.RangeArgs(1, 2),
 		RunE:         wiz,
 	}
 	cmd.Flags().BoolVar(&useStaticIP, "use-static-ip", true, "attach static Public IP on cloud servers")
@@ -82,20 +85,31 @@ The node wiz command creates a devnet and deploys, sync and validate a subnet in
 	cmd.Flags().StringVar(&nodeConf, "node-config", "", "path to avalanchego node configuration for subnet")
 	cmd.Flags().StringVar(&subnetConf, "subnet-config", "", "path to the subnet configuration for subnet")
 	cmd.Flags().StringVar(&chainConf, "chain-config", "", "path to the chain configuration for subnet")
+	cmd.Flags().BoolVar(&useLatestAvalanchegoVersion, "latest-avalanchego", false, "install latest avalanchego version on node/s")
+	cmd.Flags().StringVar(&useCustomAvalanchegoVersion, "avalanchego-version", "", "install given avalanchego version on node/s")
+	cmd.Flags().StringSliceVar(&validators, "validators", []string{}, "deploy subnet into given comma separated list of validators. defaults to all cluster nodes")
 	return cmd
 }
 
 func wiz(cmd *cobra.Command, args []string) error {
 	clusterName := args[0]
-	subnetName := args[1]
-	exists, err := clusterExists(clusterName)
+	subnetName := ""
+	if len(args) > 1 {
+		subnetName = args[1]
+	}
+	clusterAlreadyExists, err := clusterExists(clusterName)
 	if err != nil {
 		return err
 	}
-	if exists {
-		return fmt.Errorf("cluster %s already exists", clusterName)
+	if clusterAlreadyExists {
+		if err := checkClusterIsADevnet(clusterName); err != nil {
+			return err
+		}
 	}
-	if !app.SidecarExists(subnetName) || forceSubnetCreate {
+	if clusterAlreadyExists && subnetName == "" {
+		return fmt.Errorf("expecting to add subnet to existing cluster but no subnet-name was provided")
+	}
+	if subnetName != "" && (!app.SidecarExists(subnetName) || forceSubnetCreate) {
 		ux.Logger.PrintToUser("")
 		ux.Logger.PrintToUser(logging.Green.Wrap("Creating the subnet"))
 		ux.Logger.PrintToUser("")
@@ -126,18 +140,43 @@ func wiz(cmd *cobra.Command, args []string) error {
 			}
 		}
 	}
-	createDevnet = true
-	useAvalanchegoVersionFromSubnet = subnetName
-	ux.Logger.PrintToUser("")
-	ux.Logger.PrintToUser(logging.Green.Wrap("Creating the devnet"))
-	ux.Logger.PrintToUser("")
-	err = createNodes(cmd, []string{clusterName})
-	if err != nil {
-		return err
+
+	if !clusterAlreadyExists {
+		createDevnet = true
+		useAvalanchegoVersionFromSubnet = subnetName
+		ux.Logger.PrintToUser("")
+		ux.Logger.PrintToUser(logging.Green.Wrap("Creating the devnet..."))
+		ux.Logger.PrintToUser("")
+		if err := createNodes(cmd, []string{clusterName}); err != nil {
+			return err
+		}
+	} else {
+		ux.Logger.PrintToUser("")
+		ux.Logger.PrintToUser(logging.Green.Wrap("Adding subnet into existing devnet %s..."), clusterName)
+		ux.Logger.PrintToUser("")
 	}
+
+	// check all validators are found
+	if len(validators) != 0 {
+		hosts, err := ansible.GetInventoryFromAnsibleInventoryFile(app.GetAnsibleInventoryDirPath(clusterName))
+		if err != nil {
+			return err
+		}
+		_, err = filterHosts(hosts, validators)
+		if err != nil {
+			return err
+		}
+	}
+
 	if err := waitForHealthyCluster(clusterName, healthCheckTimeout, healthCheckPoolTime); err != nil {
 		return err
 	}
+	if subnetName == "" {
+		ux.Logger.PrintToUser("")
+		ux.Logger.PrintToUser(logging.Green.Wrap("Devnet %s has been created!"), clusterName)
+		return nil
+	}
+
 	ux.Logger.PrintToUser("")
 	ux.Logger.PrintToUser(logging.Green.Wrap("Deploying the subnet"))
 	ux.Logger.PrintToUser("")
@@ -175,7 +214,11 @@ func wiz(cmd *cobra.Command, args []string) error {
 		return err
 	}
 	ux.Logger.PrintToUser("")
-	ux.Logger.PrintToUser(logging.Green.Wrap("Devnet %s has been created and is validating subnet %s!"), clusterName, subnetName)
+	if clusterAlreadyExists {
+		ux.Logger.PrintToUser(logging.Green.Wrap("Devnet %s is now validating subnet %s"), clusterName, subnetName)
+	} else {
+		ux.Logger.PrintToUser(logging.Green.Wrap("Devnet %s is successfully created and is now validating subnet %s!"), clusterName, subnetName)
+	}
 	return nil
 }
 
@@ -228,6 +271,12 @@ func waitForClusterSubnetStatus(
 	if err != nil {
 		return err
 	}
+	if len(validators) != 0 {
+		hosts, err = filterHosts(hosts, validators)
+		if err != nil {
+			return err
+		}
+	}
 	defer disconnectHosts(hosts)
 	startTime := time.Now()
 	for {
@@ -275,3 +324,50 @@ func waitForClusterSubnetStatus(
 		time.Sleep(poolTime)
 	}
 }
+
+func checkClusterIsADevnet(clusterName string) error {
+	exists, err := clusterExists(clusterName)
+	if err != nil {
+		return err
+	}
+	if !exists {
+		return fmt.Errorf("cluster %q does not exists", clusterName)
+	}
+	clustersConfig, err := app.LoadClustersConfig()
+	if err != nil {
+		return err
+	}
+	if clustersConfig.Clusters[clusterName].Network.Kind != models.Devnet {
+		return fmt.Errorf("cluster %q is not a Devnet", clusterName)
+	}
+	return nil
+}
+
+func filterHosts(hosts []*models.Host, nodes []string) ([]*models.Host, error) {
+	indices := set.Set[int]{}
+	for _, node := range nodes {
+		added := false
+		for i, host := range hosts {
+			cloudID := host.GetCloudID()
+			ip := host.IP
+			nodeID, err := getNodeID(app.GetNodeInstanceDirPath(cloudID))
+			if err != nil {
+				return nil, err
+			}
+			if slices.Contains([]string{cloudID, ip, nodeID.String()}, node) {
+				added = true
+				indices.Add(i)
+			}
+		}
+		if !added {
+			return nil, fmt.Errorf("node %q not found", node)
+		}
+	}
+	filteredHosts := []*models.Host{}
+	for i, host := range hosts {
+		if indices.Contains(i) {
+			filteredHosts = append(filteredHosts, host)
+		}
+	}
+	return filteredHosts, nil
+}

pkg/ssh/ssh.go

@@ -294,7 +294,7 @@ func RunSSHCheckBootstrapped(host *models.Host) ([]byte, error) {
 // RunSSHCheckHealthy checks if node is healthy
 func RunSSHCheckHealthy(host *models.Host) ([]byte, error) {
 	// Craft and send the HTTP POST request
-	requestBody := "{\"jsonrpc\":\"2.0\", \"id\":1,\"method\":\"health.health\"}"
+	requestBody := "{\"jsonrpc\":\"2.0\", \"id\":1,\"method\":\"health.health\",\"params\": {\"tags\": [\"P\"]}}"
 	return PostOverSSH(host, "/ext/health", requestBody)
 }