cluster: check support check before scale-out the instance (#1659)

components/cluster/command/check.go

@@ -28,22 +28,34 @@ func newCheckCmd() *cobra.Command {
 		IdentityFile: path.Join(utils.UserHome(), ".ssh", "id_rsa"),
 	}
 	cmd := &cobra.Command{
-		Use:   "check <topology.yml | cluster-name>",
+		Use:   "check <topology.yml | cluster-name> [scale-out.yml]",
 		Short: "Perform preflight checks for the cluster.",
 		Long: `Perform preflight checks for the cluster. By default, it checks deploy servers
 before a cluster is deployed, the input is the topology.yaml for the cluster.
 If '--cluster' is set, it will perform checks for an existing cluster, the input
 is the cluster name. Some checks are ignore in this mode, such as port and dir
-conflict checks with other clusters`,
+conflict checks with other clusters
+If you want to check the scale-out topology, please use execute the following command
+'	check <cluster-name> <scale-out.yml> --cluster	'
+it will the new instances `,
 		RunE: func(cmd *cobra.Command, args []string) error {
-			if len(args) != 1 {
+			if len(args) != 1 && len(args) != 2 {
 				return cmd.Help()
 			}
+			scaleOutTopo := ""
 
 			if opt.ExistCluster {
 				clusterReport.ID = scrubClusterName(args[0])
 			}
-			return cm.CheckCluster(args[0], opt, gOpt)
+
+			if len(args) == 2 {
+				if !opt.ExistCluster {
+					return cmd.Help()
+				}
+				scaleOutTopo = args[1]
+			}
+
+			return cm.CheckCluster(args[0], scaleOutTopo, opt, gOpt)
 		},
 	}
 

pkg/cluster/manager/check.go

@@ -45,13 +45,14 @@ type CheckOptions struct {
 }
 
 // CheckCluster check cluster before deploying or upgrading
-func (m *Manager) CheckCluster(clusterOrTopoName string, opt CheckOptions, gOpt operator.Options) error {
+func (m *Manager) CheckCluster(clusterOrTopoName, scaleoutTopo string, opt CheckOptions, gOpt operator.Options) error {
 	var topo spec.Specification
 	ctx := ctxt.New(
 		context.Background(),
 		gOpt.Concurrency,
 		m.logger,
 	)
+	var currTopo *spec.Specification
 
 	if opt.ExistCluster { // check for existing cluster
 		clusterName := clusterOrTopoName
@@ -69,10 +70,28 @@ func (m *Manager) CheckCluster(clusterOrTopoName string, opt CheckOptions, gOpt
 		if err != nil {
 			return err
 		}
-		opt.User = metadata.User
-		opt.IdentityFile = m.specManager.Path(clusterName, "ssh", "id_rsa")
 
-		topo = *metadata.Topology
+		if scaleoutTopo != "" {
+			currTopo = metadata.Topology
+			// complete global configuration
+			topo.GlobalOptions = currTopo.GlobalOptions
+			topo.MonitoredOptions = currTopo.MonitoredOptions
+			topo.ServerConfigs = currTopo.ServerConfigs
+
+			if err := spec.ParseTopologyYaml(scaleoutTopo, &topo); err != nil {
+				return err
+			}
+			spec.ExpandRelativeDir(&topo)
+
+			// checkConflict after fillHostArch
+			// scaleOutTopo also is not exists instacne
+			opt.ExistCluster = false
+		} else {
+			opt.IdentityFile = m.specManager.Path(clusterName, "ssh", "id_rsa")
+			topo = *metadata.Topology
+			opt.User = metadata.User
+		}
+
 		topo.AdjustByVersion(metadata.Version)
 	} else { // check before cluster is deployed
 		topoFileName := clusterOrTopoName
@@ -82,15 +101,7 @@ func (m *Manager) CheckCluster(clusterOrTopoName string, opt CheckOptions, gOpt
 		}
 		spec.ExpandRelativeDir(&topo)
 
-		clusterList, err := m.specManager.GetAllClusters()
-		if err != nil {
-			return err
-		}
-		// use a dummy cluster name, the real cluster name is set during deploy
-		if err := spec.CheckClusterPortConflict(clusterList, "nonexist-dummy-tidb-cluster", &topo); err != nil {
-			return err
-		}
-		if err := spec.CheckClusterDirConflict(clusterList, "nonexist-dummy-tidb-cluster", &topo); err != nil {
+		if err := checkConflict(m, "nonexist-dummy-tidb-cluster", &topo); err != nil {
 			return err
 		}
 	}
@@ -115,6 +126,17 @@ func (m *Manager) CheckCluster(clusterOrTopoName string, opt CheckOptions, gOpt
 		return err
 	}
 
+	// Abort scale out operation if the merged topology is invalid
+	if currTopo != nil && scaleoutTopo != "" {
+		mergedTopo := currTopo.MergeTopo(&topo)
+		if err := mergedTopo.Validate(); err != nil {
+			return err
+		}
+		if err := checkConflict(m, clusterOrTopoName, mergedTopo); err != nil {
+			return err
+		}
+	}
+
 	if err := checkSystemInfo(ctx, sshConnProps, sshProxyProps, &topo, &gOpt, &opt); err != nil {
 		return err
 	}
@@ -186,6 +208,38 @@ func checkSystemInfo(
 					opt.Opr,
 				)
 			}
+
+			if !opt.ExistCluster {
+				t1 = t1.
+					CheckSys(
+						inst.GetHost(),
+						inst.DeployDir(),
+						task.ChecktypeIsExist,
+						topo,
+						opt.Opr,
+					).
+					CheckSys(
+						inst.GetHost(),
+						inst.DataDir(),
+						task.ChecktypeIsExist,
+						topo,
+						opt.Opr,
+					).
+					CheckSys(
+						inst.GetHost(),
+						inst.LogDir(),
+						task.ChecktypeIsExist,
+						topo,
+						opt.Opr,
+					).
+					CheckSys(
+						inst.GetHost(),
+						fmt.Sprintf("/etc/systemd/system/%s-%d.service", inst.ComponentName(), inst.GetPort()),
+						task.ChecktypeIsExist,
+						topo,
+						opt.Opr,
+					)
+			}
 			// if the data dir set in topology is relative, and the home dir of deploy user
 			// and the user run the check command is on different partitions, the disk detection
 			// may be using incorrect partition for validations.
@@ -199,6 +253,7 @@ func checkSystemInfo(
 						topo,
 						opt.Opr,
 					)
+
 				if opt.ExistCluster {
 					t1 = t1.CheckSys(
 						inst.GetHost(),
@@ -270,20 +325,6 @@ func checkSystemInfo(
 						topo,
 						opt.Opr,
 					).
-					// check for listening port
-					Shell(
-						inst.GetHost(),
-						"ss -lnt",
-						"",
-						false,
-					).
-					CheckSys(
-						inst.GetHost(),
-						"",
-						task.CheckTypePort,
-						topo,
-						opt.Opr,
-					).
 					// check for system limits
 					Shell(
 						inst.GetHost(),
@@ -328,6 +369,24 @@ func checkSystemInfo(
 						topo,
 						opt.Opr,
 					)
+
+				if !opt.ExistCluster {
+					t1 = t1.
+						// check for listening port
+						Shell(
+							inst.GetHost(),
+							"ss -lnt",
+							"",
+							false,
+						).
+						CheckSys(
+							inst.GetHost(),
+							"",
+							task.CheckTypePort,
+							topo,
+							opt.Opr,
+						)
+				}
 			}
 
 			checkSysTasks = append(
@@ -596,3 +655,17 @@ func (m *Manager) checkRegionsInfo(clusterName string, topo *spec.Specification,
 	}
 	return nil
 }
+
+// checkConflict checks cluster conflict
+func checkConflict(m *Manager, clusterName string, topo spec.Topology) error {
+	clusterList, err := m.specManager.GetAllClusters()
+	if err != nil {
+		return err
+	}
+	// use a dummy cluster name, the real cluster name is set during deploy
+	if err := spec.CheckClusterPortConflict(clusterList, clusterName, topo); err != nil {
+		return err
+	}
+	err = spec.CheckClusterDirConflict(clusterList, clusterName, topo)
+	return err
+}

pkg/cluster/manager/scale_out.go

@@ -166,14 +166,7 @@ func (m *Manager) ScaleOut(
 			}
 		}
 
-		clusterList, err := m.specManager.GetAllClusters()
-		if err != nil {
-			return err
-		}
-		if err := spec.CheckClusterPortConflict(clusterList, name, mergedTopo); err != nil {
-			return err
-		}
-		if err := spec.CheckClusterDirConflict(clusterList, name, mergedTopo); err != nil {
+		if err := checkConflict(m, name, mergedTopo); err != nil {
 			return err
 		}
 	}

pkg/cluster/operation/check.go

@@ -62,6 +62,7 @@ var (
 	CheckNameFio           = "fio"
 	CheckNameTHP           = "thp"
 	CheckNameDirPermission = "permission"
+	CheckNameDirExist      = "exist"
 )
 
 // CheckResult is the result of a check
@@ -866,3 +867,29 @@ func CheckDirPermission(ctx context.Context, e ctxt.Executor, user, path string)
 
 	return results
 }
+
+// CheckDirIsExist check if the directory exists
+func CheckDirIsExist(ctx context.Context, e ctxt.Executor, path string) []*CheckResult {
+	var results []*CheckResult
+
+	if path == "" {
+		return results
+	}
+
+	req, _, _ := e.Execute(ctx,
+		fmt.Sprintf(
+			"[ -e %s ] && echo 1",
+			path,
+		),
+		false)
+
+	if strings.ReplaceAll(string(req), "\n", "") == "1" {
+		results = append(results, &CheckResult{
+			Name: CheckNameDirExist,
+			Err:  fmt.Errorf("%s already exists", path),
+			Msg:  fmt.Sprintf("%s already exists", path),
+		})
+	}
+
+	return results
+}

pkg/cluster/task/check.go

@@ -36,6 +36,7 @@ var (
 	CheckTypePartitions   = "partitions"
 	CheckTypeFIO          = "fio"
 	CheckTypePermission   = "permission"
+	ChecktypeIsExist      = "exist"
 )
 
 // place the check utilities are stored
@@ -147,6 +148,13 @@ func (c *CheckSys) Execute(ctx context.Context) error {
 			return ErrNoExecutor
 		}
 		storeResults(ctx, c.host, operator.CheckDirPermission(ctx, e, c.topo.GlobalOptions.User, c.checkDir))
+	case ChecktypeIsExist:
+		e, ok := ctxt.GetInner(ctx).GetExecutor(c.host)
+		if !ok {
+			return ErrNoExecutor
+		}
+		// check partition mount options for data_dir
+		storeResults(ctx, c.host, operator.CheckDirIsExist(ctx, e, c.checkDir))
 	}
 
 	return nil