roachprod: prepare centralized - use centralized from roachprod lib

This patch starts using the roachprod centralized client from the
roachprod library when properly configured via environment variables.

Epic: none
Release note: None

Author: golgeek

pkg/BUILD.bazel

@@ -1812,6 +1812,7 @@ GO_TARGETS = [
     "//pkg/roachpb:roachpb_test",
     "//pkg/roachprod/blobfixture:blobfixture",
     "//pkg/roachprod/blobfixture:blobfixture_test",
+    "//pkg/roachprod/centralizedapi:centralizedapi",
     "//pkg/roachprod/cloud/types:types",
     "//pkg/roachprod/cloud:cloud",
     "//pkg/roachprod/cloud:cloud_test",

pkg/cmd/roachprod/cli/commands.go

@@ -439,9 +439,15 @@ hosts file.
 				}
 			} else {
 				machineType := func(clusterVMs vm.List) string {
+					if len(clusterVMs) == 0 {
+						return ""
+					}
 					return clusterVMs[0].MachineType
 				}
 				cpuArch := func(clusterVMs vm.List) string {
+					if len(clusterVMs) == 0 {
+						return ""
+					}
 					// Display CPU architecture and family.
 					if clusterVMs[0].CPUArch == "" {
 						// N.B. Either a local cluster or unsupported cloud provider.
@@ -501,6 +507,9 @@ hosts file.
 							}
 							timeRemaining := c.LifetimeRemaining().Round(time.Second)
 							formatTTL := func(ttl time.Duration) string {
+								if len(c.VMs) == 0 {
+									return color.HiRedString("")
+								}
 								if c.VMs[0].Preemptible {
 									return color.HiMagentaString(ttl.String())
 								} else {
@@ -528,7 +537,12 @@ hosts file.
 									color.HiGreenString(""))
 							}
 						} else {
-							fmt.Fprintf(tw, "\t(-)")
+							fmt.Fprintf(tw, "\t%s\t%s\t%s\t%s\t%s\t",
+								color.HiGreenString(""),
+								color.HiGreenString(""),
+								color.HiWhiteString(""),
+								color.HiWhiteString(""),
+								color.HiGreenString(""))
 						}
 						fmt.Fprintf(tw, "\n")
 					}

pkg/roachprod/BUILD.bazel

@@ -16,6 +16,7 @@ go_library(
         "//pkg/build",
         "//pkg/cli/exit",
         "//pkg/cmd/roachprod/grafana",
+        "//pkg/roachprod/centralizedapi",
         "//pkg/roachprod/cloud",
         "//pkg/roachprod/cloud/types",
         "//pkg/roachprod/config",

pkg/roachprod/centralizedapi/BUILD.bazel

@@ -0,0 +1,9 @@
+load("@io_bazel_rules_go//go:def.bzl", "go_library")
+
+go_library(
+    name = "centralizedapi",
+    srcs = ["centralizedapi.go"],
+    importpath = "github.com/cockroachdb/cockroach/pkg/roachprod/centralizedapi",
+    visibility = ["//visibility:public"],
+    deps = ["//pkg/cmd/roachprod-centralized/client"],
+)

pkg/roachprod/centralizedapi/centralizedapi.go

@@ -0,0 +1,29 @@
+// Copyright 2025 The Cockroach Authors.
+//
+// Use of this software is governed by the CockroachDB Software License
+// included in the /LICENSE file.
+
+package centralizedapi
+
+import (
+	"sync"
+
+	"github.com/cockroachdb/cockroach/pkg/cmd/roachprod-centralized/client"
+)
+
+var (
+	once           sync.Once
+	centralizedapi *client.Client
+)
+
+// GetCentralizedAPIClient returns a singleton instance of the centralized API client.
+func GetCentralizedAPIClient() *client.Client {
+	once.Do(func() {
+		client, err := client.NewClientWithConfig(client.LoadConfigFromEnv())
+		if err != nil {
+			panic(err)
+		}
+		centralizedapi = client
+	})
+	return centralizedapi
+}

pkg/roachprod/cloud/BUILD.bazel

@@ -11,6 +11,8 @@ go_library(
     visibility = ["//visibility:public"],
     deps = [
         "//pkg/cloud/amazon",
+        "//pkg/cmd/roachprod-centralized/client",
+        "//pkg/roachprod/centralizedapi",
         "//pkg/roachprod/cloud/types",
         "//pkg/roachprod/config",
         "//pkg/roachprod/logger",

pkg/roachprod/cloud/cluster_cloud.go

@@ -11,6 +11,8 @@ import (
 	"sort"
 	"time"
 
+	roachprodcentralized "github.com/cockroachdb/cockroach/pkg/cmd/roachprod-centralized/client"
+	"github.com/cockroachdb/cockroach/pkg/roachprod/centralizedapi"
 	cloudcluster "github.com/cockroachdb/cockroach/pkg/roachprod/cloud/types"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/logger"
 	"github.com/cockroachdb/cockroach/pkg/roachprod/ui"
@@ -140,6 +142,22 @@ func (c *Cloud) ListCloud(l *logger.Logger, options vm.ListOptions) error {
 
 	providers := append(c.providers, c.localProviders...)
 
+	apiClient := centralizedapi.GetCentralizedAPIClient()
+	if apiClient.IsEnabled() {
+		// Use the centralized roachprod service to list clusters.
+		response, err := apiClient.ListClusters(context.Background(), l, roachprodcentralized.ListClustersOptions{})
+		if err != nil {
+			l.Errorf("Error listing clusters using the centralized API: %s", err)
+		} else {
+			c.Clusters = response.Clusters
+			c.BadInstances = response.BadInstances
+
+			// Remote providers are already listed by the centralized service,
+			// so only use local providers in the local listing.
+			providers = c.localProviders
+		}
+	}
+
 	// List all VMs across all providers in parallel.
 	providerVMs := make(map[string]vm.List)
 	var g errgroup.Group
@@ -276,6 +294,14 @@ func CreateCluster(l *logger.Logger, opts []*ClusterCreateOpts) (*cloudcluster.C
 		Lifetime:  opts[0].CreateOpts.Lifetime,
 	}
 
+	apiClient := centralizedapi.GetCentralizedAPIClient()
+	if apiClient.IsEnabled() {
+		// Register the new cluster with the centralized roachprod service.
+		if err := apiClient.RegisterCluster(context.Background(), l, c); err != nil {
+			return nil, err
+		}
+	}
+
 	// Keep track of the total number of nodes created, as we append all cluster names
 	// with the node count.
 	var nodesCreated int
@@ -338,6 +364,14 @@ func CreateCluster(l *logger.Logger, opts []*ClusterCreateOpts) (*cloudcluster.C
 	// `roachprod.Start` expects nodes/vms to be in sorted order
 	sort.Sort(c.VMs)
 
+	if apiClient.IsEnabled() {
+		// Use the centralized roachprod service to register the final state
+		// of the cluster.
+		if err := apiClient.RegisterClusterUpdate(context.Background(), l, c); err != nil {
+			return nil, err
+		}
+	}
+
 	return c, nil
 }
 
@@ -378,6 +412,14 @@ func GrowCluster(l *logger.Logger, c *cloudcluster.Cluster, numNodes int) error
 	// `roachprod.Start` expects nodes/vms to be in sorted order
 	sort.Sort(c.VMs)
 
+	apiClient := centralizedapi.GetCentralizedAPIClient()
+	if apiClient.IsEnabled() {
+		// Use the centralized roachprod service to create the cluster.
+		if err := apiClient.RegisterClusterUpdate(context.Background(), l, c); err != nil {
+			return err
+		}
+	}
+
 	return nil
 }
 
@@ -409,6 +451,15 @@ func ShrinkCluster(l *logger.Logger, c *cloudcluster.Cluster, numNodes int) erro
 
 	// Update the list of VMs in the cluster.
 	c.VMs = c.VMs[:len(c.VMs)-numNodes]
+
+	apiClient := centralizedapi.GetCentralizedAPIClient()
+	if apiClient.IsEnabled() {
+		// Use the centralized roachprod service to create the cluster.
+		if err := apiClient.RegisterClusterUpdate(context.Background(), l, c); err != nil {
+			return err
+		}
+	}
+
 	return nil
 }
 
@@ -427,7 +478,10 @@ func DestroyCluster(l *logger.Logger, c *cloudcluster.Cluster) error {
 		publicRecords = append(publicRecords, v.PublicDNS)
 	}
 	dnsErr := vm.FanOutDNS(c.VMs, func(p vm.DNSProvider, vms vm.List) error {
-		publicRecordsErr := p.DeletePublicRecordsByName(context.Background(), publicRecords...)
+		var publicRecordsErr error
+		if !centralizedapi.GetCentralizedAPIClient().IsEnabled() {
+			publicRecordsErr = p.DeletePublicRecordsByName(context.Background(), publicRecords...)
+		}
 		srvRecordsErr := p.DeleteSRVRecordsBySubdomain(context.Background(), c.Name)
 		return errors.CombineErrors(publicRecordsErr, srvRecordsErr)
 	})
@@ -444,6 +498,15 @@ func DestroyCluster(l *logger.Logger, c *cloudcluster.Cluster) error {
 		return p.Delete(l, vms)
 	})
 	stopSpinner()
+
+	if clusterErr == nil {
+		apiClient := centralizedapi.GetCentralizedAPIClient()
+		if apiClient.IsEnabled() {
+			if err := apiClient.RegisterClusterDelete(context.Background(), l, c.Name); err != nil {
+				l.Printf("WARNING: failed to delete cluster %s from centralized service: %s", c.Name, err)
+			}
+		}
+	}
 	return errors.CombineErrors(dnsErr, clusterErr)
 }
 
@@ -458,5 +521,13 @@ func ExtendCluster(l *logger.Logger, c *cloudcluster.Cluster, extension time.Dur
 		return err
 	}
 	c.Lifetime = newLifetime
+
+	apiClient := centralizedapi.GetCentralizedAPIClient()
+	if apiClient.IsEnabled() {
+		// Use the centralized roachprod service to create the cluster.
+		if err := apiClient.RegisterClusterUpdate(context.Background(), l, c); err != nil {
+			return err
+		}
+	}
 	return nil
 }

pkg/roachprod/clusters_cache.go

@@ -159,7 +159,11 @@ func LoadClusters() error {
 		}
 
 		if len(c.VMs) == 0 {
-			return errors.Errorf("found no VMs in %s", clusterFilename(name))
+			// When roachprod-centralized is available, clusters are registered empty
+			// and updated later when the cluster is actually created to avoid race conditions
+			// in creation and background clusters syncs.
+			// Ignore empty clusters for now.
+			continue
 		}
 		if shouldIgnoreCluster(c) {
 			continue