|
| 1 | +// Copyright 2018 The Cockroach Authors. |
| 2 | +// |
| 3 | +// Use of this software is governed by the CockroachDB Software License |
| 4 | +// included in the /LICENSE file. |
| 5 | + |
| 6 | +package gce |
| 7 | + |
| 8 | +import ( |
| 9 | + "context" |
| 10 | + "fmt" |
| 11 | + "strings" |
| 12 | + |
| 13 | + compute "cloud.google.com/go/compute/apiv1" |
| 14 | + computepb "cloud.google.com/go/compute/apiv1/computepb" |
| 15 | + "github.com/cockroachdb/cockroach/pkg/roachprod/logger" |
| 16 | + "github.com/cockroachdb/cockroach/pkg/roachprod/ui" |
| 17 | + "github.com/cockroachdb/cockroach/pkg/roachprod/vm" |
| 18 | + "github.com/cockroachdb/cockroach/pkg/util/syncutil" |
| 19 | + "github.com/cockroachdb/errors" |
| 20 | + "google.golang.org/protobuf/proto" |
| 21 | +) |
| 22 | + |
| 23 | +// buildInstanceProperties creates a computepb.InstanceProperties struct for the |
| 24 | +// GCP Compute SDK. This is the SDK equivalent of computeInstanceArgs(). |
| 25 | +func (p *Provider) buildInstanceProperties( |
| 26 | + l *logger.Logger, |
| 27 | + opts vm.CreateOpts, |
| 28 | + providerOpts *ProviderOpts, |
| 29 | + startupScriptContent string, |
| 30 | + zone string, |
| 31 | + labels map[string]string, |
| 32 | +) (*computepb.InstanceProperties, error) { |
| 33 | + // Determine which image to use. |
| 34 | + image := providerOpts.Image |
| 35 | + imageProject := defaultImageProject |
| 36 | + |
| 37 | + if providerOpts.useArmAMI() { |
| 38 | + image = ARM64Image |
| 39 | + } |
| 40 | + |
| 41 | + if opts.Arch == string(vm.ArchFIPS) { |
| 42 | + image = FIPSImage |
| 43 | + imageProject = FIPSImageProject |
| 44 | + } |
| 45 | + |
| 46 | + // Build the boot disk configuration. |
| 47 | + disks := []*computepb.AttachedDisk{ |
| 48 | + { |
| 49 | + AutoDelete: proto.Bool(true), |
| 50 | + Boot: proto.Bool(true), |
| 51 | + InitializeParams: &computepb.AttachedDiskInitializeParams{ |
| 52 | + DiskSizeGb: proto.Int64(int64(opts.OsVolumeSize)), |
| 53 | + DiskType: proto.String(providerOpts.BootDiskType), |
| 54 | + SourceImage: proto.String(fmt.Sprintf("projects/%s/global/images/%s", imageProject, image)), |
| 55 | + }, |
| 56 | + }, |
| 57 | + } |
| 58 | + |
| 59 | + // Add additional disks (SSDs or persistent disks). |
| 60 | + if !providerOpts.BootDiskOnly { |
| 61 | + if opts.SSDOpts.UseLocalSSD { |
| 62 | + // Local SSDs are physically attached to the host machine. |
| 63 | + for i := 0; i < providerOpts.SSDCount; i++ { |
| 64 | + disks = append(disks, &computepb.AttachedDisk{ |
| 65 | + AutoDelete: proto.Bool(true), |
| 66 | + Type: proto.String(computepb.AttachedDisk_SCRATCH.String()), |
| 67 | + Mode: proto.String(computepb.AttachedDisk_READ_WRITE.String()), |
| 68 | + InitializeParams: &computepb.AttachedDiskInitializeParams{ |
| 69 | + DiskType: proto.String("local-ssd"), |
| 70 | + }, |
| 71 | + Interface: proto.String(computepb.AttachedDisk_NVME.String()), |
| 72 | + }) |
| 73 | + } |
| 74 | + } else { |
| 75 | + // Persistent disks are network-attached storage. |
| 76 | + for i := 0; i < providerOpts.PDVolumeCount; i++ { |
| 77 | + disks = append(disks, &computepb.AttachedDisk{ |
| 78 | + AutoDelete: proto.Bool(true), |
| 79 | + InitializeParams: &computepb.AttachedDiskInitializeParams{ |
| 80 | + DiskSizeGb: proto.Int64(int64(providerOpts.PDVolumeSize)), |
| 81 | + DiskType: proto.String(providerOpts.PDVolumeType), |
| 82 | + }, |
| 83 | + }) |
| 84 | + } |
| 85 | + } |
| 86 | + } |
| 87 | + |
| 88 | + // Configure networking. |
| 89 | + if len(zone) < 3 { |
| 90 | + return nil, errors.Newf("invalid zone %q: must be at least 3 characters", zone) |
| 91 | + } |
| 92 | + region := zone[:len(zone)-2] |
| 93 | + project := p.GetProject() |
| 94 | + networkInterfaces := []*computepb.NetworkInterface{ |
| 95 | + { |
| 96 | + Subnetwork: proto.String(fmt.Sprintf("projects/%s/regions/%s/subnetworks/default", project, region)), |
| 97 | + AccessConfigs: []*computepb.AccessConfig{ |
| 98 | + { |
| 99 | + Name: proto.String("External NAT"), |
| 100 | + Type: proto.String(computepb.AccessConfig_ONE_TO_ONE_NAT.String()), |
| 101 | + }, |
| 102 | + }, |
| 103 | + }, |
| 104 | + } |
| 105 | + |
| 106 | + // Configure scheduling. |
| 107 | + scheduling := &computepb.Scheduling{} |
| 108 | + if providerOpts.preemptible { |
| 109 | + scheduling.Preemptible = proto.Bool(true) |
| 110 | + scheduling.OnHostMaintenance = proto.String(computepb.Scheduling_TERMINATE.String()) |
| 111 | + scheduling.AutomaticRestart = proto.Bool(false) |
| 112 | + } else if providerOpts.UseSpot { |
| 113 | + scheduling.ProvisioningModel = proto.String(computepb.Scheduling_SPOT.String()) |
| 114 | + } else if providerOpts.TerminateOnMigration { |
| 115 | + scheduling.OnHostMaintenance = proto.String(computepb.Scheduling_TERMINATE.String()) |
| 116 | + } else { |
| 117 | + scheduling.OnHostMaintenance = proto.String(computepb.Scheduling_MIGRATE.String()) |
| 118 | + } |
| 119 | + |
| 120 | + // Configure the service account. |
| 121 | + var serviceAccounts []*computepb.ServiceAccount |
| 122 | + sa := providerOpts.ServiceAccount |
| 123 | + if sa == "" && p.GetProject() == p.defaultProject { |
| 124 | + sa = providerOpts.defaultServiceAccount |
| 125 | + } |
| 126 | + if sa != "" { |
| 127 | + serviceAccounts = []*computepb.ServiceAccount{ |
| 128 | + { |
| 129 | + Email: proto.String(sa), |
| 130 | + Scopes: []string{"https://www.googleapis.com/auth/cloud-platform"}, |
| 131 | + }, |
| 132 | + } |
| 133 | + } |
| 134 | + |
| 135 | + // Configure metadata (including the startup script). |
| 136 | + metadata := &computepb.Metadata{ |
| 137 | + Items: []*computepb.Items{ |
| 138 | + { |
| 139 | + Key: proto.String("startup-script"), |
| 140 | + Value: proto.String(startupScriptContent), |
| 141 | + }, |
| 142 | + }, |
| 143 | + } |
| 144 | + |
| 145 | + // Build the final InstanceProperties. |
| 146 | + props := &computepb.InstanceProperties{ |
| 147 | + MachineType: proto.String(providerOpts.MachineType), |
| 148 | + Disks: disks, |
| 149 | + NetworkInterfaces: networkInterfaces, |
| 150 | + Scheduling: scheduling, |
| 151 | + ServiceAccounts: serviceAccounts, |
| 152 | + Metadata: metadata, |
| 153 | + } |
| 154 | + |
| 155 | + if providerOpts.MinCPUPlatform != "" { |
| 156 | + props.MinCpuPlatform = proto.String(providerOpts.MinCPUPlatform) |
| 157 | + } |
| 158 | + |
| 159 | + // Set the labels on the instance properties. |
| 160 | + props.Labels = labels |
| 161 | + |
| 162 | + return props, nil |
| 163 | +} |
| 164 | + |
| 165 | +// parseLabelsString converts a "key=value,key2=value2" string into a map. |
| 166 | +func parseLabelsString(labels string) (map[string]string, error) { |
| 167 | + result := make(map[string]string) |
| 168 | + if labels == "" { |
| 169 | + return result, nil |
| 170 | + } |
| 171 | + |
| 172 | + pairs := strings.Split(labels, ",") |
| 173 | + for _, pair := range pairs { |
| 174 | + kv := strings.Split(pair, "=") |
| 175 | + if len(kv) != 2 { |
| 176 | + return nil, errors.Newf("invalid label format: %q (expected key=value)", pair) |
| 177 | + } |
| 178 | + result[kv[0]] = kv[1] |
| 179 | + } |
| 180 | + return result, nil |
| 181 | +} |
| 182 | + |
| 183 | +// bulkInsertInstances uses the GCP Compute SDK to create multiple VMs in a |
| 184 | +// single zone using the BulkInsert API. |
| 185 | +// |
| 186 | +// Benefits over CLI-based approach: |
| 187 | +// - CLI: 100 VMs = ~4 concurrent gcloud CLI calls with batching |
| 188 | +// - BulkInsert: 100 VMs = 1 API call per zone |
| 189 | +func (p *Provider) bulkInsertInstances( |
| 190 | + ctx context.Context, |
| 191 | + l *logger.Logger, |
| 192 | + zone string, |
| 193 | + hostNames []string, |
| 194 | + instanceProps *computepb.InstanceProperties, |
| 195 | +) error { |
| 196 | + project := p.GetProject() |
| 197 | + |
| 198 | + client, err := compute.NewInstancesRESTClient(ctx) |
| 199 | + if err != nil { |
| 200 | + return errors.Wrap(err, "failed to create GCE compute client") |
| 201 | + } |
| 202 | + defer func() { _ = client.Close() }() |
| 203 | + |
| 204 | + // Build the per-instance properties map for exact VM naming. |
| 205 | + perInstanceProps := make(map[string]*computepb.BulkInsertInstanceResourcePerInstanceProperties) |
| 206 | + for _, name := range hostNames { |
| 207 | + perInstanceProps[name] = &computepb.BulkInsertInstanceResourcePerInstanceProperties{ |
| 208 | + Name: proto.String(name), |
| 209 | + } |
| 210 | + } |
| 211 | + |
| 212 | + // Build the BulkInsert request. |
| 213 | + req := &computepb.BulkInsertInstanceRequest{ |
| 214 | + Project: project, |
| 215 | + Zone: zone, |
| 216 | + BulkInsertInstanceResourceResource: &computepb.BulkInsertInstanceResource{ |
| 217 | + Count: proto.Int64(int64(len(hostNames))), |
| 218 | + MinCount: proto.Int64(int64(len(hostNames))), |
| 219 | + InstanceProperties: instanceProps, |
| 220 | + PerInstanceProperties: perInstanceProps, |
| 221 | + }, |
| 222 | + } |
| 223 | + |
| 224 | + // Execute the BulkInsert request (async operation). |
| 225 | + op, err := client.BulkInsert(ctx, req) |
| 226 | + if err != nil { |
| 227 | + return errors.Wrapf(err, "BulkInsert request failed for zone %s", zone) |
| 228 | + } |
| 229 | + |
| 230 | + // Wait for the operation to complete with a spinner for progress feedback. |
| 231 | + err = func() error { |
| 232 | + defer ui.NewDefaultSpinner(l, fmt.Sprintf("BulkInsert: creating %d instances in %s", len(hostNames), zone)).Start()() |
| 233 | + return op.Wait(ctx) |
| 234 | + }() |
| 235 | + if err != nil { |
| 236 | + return errors.Wrapf(err, "BulkInsert operation failed for zone %s", zone) |
| 237 | + } |
| 238 | + l.Printf("BulkInsert: successfully created %d instances in zone %s", len(hostNames), zone) |
| 239 | + return nil |
| 240 | +} |
| 241 | + |
| 242 | +// createInstancesSDK creates VMs using the GCP Compute SDK's BulkInsert API. |
| 243 | +// This is an alternative to the CLI-based approach that is more efficient |
| 244 | +// for creating large numbers of VMs. |
| 245 | +func (p *Provider) createInstancesSDK( |
| 246 | + l *logger.Logger, |
| 247 | + opts vm.CreateOpts, |
| 248 | + providerOpts *ProviderOpts, |
| 249 | + labels string, |
| 250 | + zoneToHostNames map[string][]string, |
| 251 | + usedZones []string, |
| 252 | +) (vm.List, error) { |
| 253 | + project := p.GetProject() |
| 254 | + |
| 255 | + // Compute the extraMountOpts for the startup script. |
| 256 | + extraMountOpts := "" |
| 257 | + if !providerOpts.BootDiskOnly { |
| 258 | + if opts.SSDOpts.UseLocalSSD { |
| 259 | + extraMountOpts = "discard" |
| 260 | + if opts.SSDOpts.NoExt4Barrier { |
| 261 | + extraMountOpts = fmt.Sprintf("%s,nobarrier", extraMountOpts) |
| 262 | + } |
| 263 | + } else { |
| 264 | + extraMountOpts = "discard" |
| 265 | + } |
| 266 | + } |
| 267 | + |
| 268 | + // Generate the startup script content as a string. |
| 269 | + startupScriptContent, err := generateStartupScriptContent( |
| 270 | + extraMountOpts, |
| 271 | + opts.SSDOpts.FileSystem, |
| 272 | + providerOpts.UseMultipleDisks, |
| 273 | + opts.Arch == string(vm.ArchFIPS), |
| 274 | + providerOpts.EnableCron, |
| 275 | + providerOpts.BootDiskOnly, |
| 276 | + ) |
| 277 | + if err != nil { |
| 278 | + return nil, errors.Wrap(err, "failed to generate startup script content") |
| 279 | + } |
| 280 | + |
| 281 | + // Parse the labels string into a map. |
| 282 | + labelsMap, err := parseLabelsString(labels) |
| 283 | + if err != nil { |
| 284 | + return nil, errors.Wrap(err, "failed to parse labels") |
| 285 | + } |
| 286 | + |
| 287 | + // Create VMs using BulkInsert - one API call per zone. |
| 288 | + l.Printf("Creating instances using BulkInsert API, distributed across [%s]", |
| 289 | + strings.Join(usedZones, ", ")) |
| 290 | + |
| 291 | + g := newLimitedErrorGroup() |
| 292 | + for zone, zoneHosts := range zoneToHostNames { |
| 293 | + zone := zone |
| 294 | + zoneHosts := zoneHosts |
| 295 | + |
| 296 | + g.Go(func() error { |
| 297 | + // Build the InstanceProperties for this zone (includes labels). |
| 298 | + instanceProps, err := p.buildInstanceProperties( |
| 299 | + l, opts, providerOpts, startupScriptContent, zone, labelsMap, |
| 300 | + ) |
| 301 | + if err != nil { |
| 302 | + return errors.Wrapf(err, "failed to build instance properties for zone %s", zone) |
| 303 | + } |
| 304 | + |
| 305 | + // Call BulkInsert for this zone. |
| 306 | + err = p.bulkInsertInstances( |
| 307 | + context.Background(), |
| 308 | + l, |
| 309 | + zone, |
| 310 | + zoneHosts, |
| 311 | + instanceProps, |
| 312 | + ) |
| 313 | + if err != nil { |
| 314 | + return errors.Wrapf(err, "BulkInsert failed for zone %s", zone) |
| 315 | + } |
| 316 | + |
| 317 | + return nil |
| 318 | + }) |
| 319 | + } |
| 320 | + |
| 321 | + if err := g.Wait(); err != nil { |
| 322 | + return nil, err |
| 323 | + } |
| 324 | + |
| 325 | + // Fetch the created VMs to populate vmList using parallel list calls per zone. |
| 326 | + var vmList vm.List |
| 327 | + var vmListMutex syncutil.Mutex |
| 328 | + |
| 329 | + g = newLimitedErrorGroup() |
| 330 | + for zone, zoneHosts := range zoneToHostNames { |
| 331 | + zone := zone |
| 332 | + zoneHosts := zoneHosts |
| 333 | + |
| 334 | + g.Go(func() error { |
| 335 | + // Build a set of hostnames for quick lookup. |
| 336 | + hostNameSet := make(map[string]struct{}, len(zoneHosts)) |
| 337 | + for _, name := range zoneHosts { |
| 338 | + hostNameSet[name] = struct{}{} |
| 339 | + } |
| 340 | + |
| 341 | + // Build name filter for server-side filtering. |
| 342 | + nameFilter := fmt.Sprintf("zone:%s AND name:(%s)", zone, strings.Join(zoneHosts, " ")) |
| 343 | + |
| 344 | + listArgs := []string{ |
| 345 | + "compute", "instances", "list", |
| 346 | + "--filter", nameFilter, |
| 347 | + "--project", project, |
| 348 | + "--format", "json", |
| 349 | + } |
| 350 | + |
| 351 | + var instances []jsonVM |
| 352 | + if err := runJSONCommand(listArgs, &instances); err != nil { |
| 353 | + return errors.Wrapf(err, "failed to list VMs in zone %s", zone) |
| 354 | + } |
| 355 | + |
| 356 | + // Filter to only include our created VMs (safety check). |
| 357 | + vmListMutex.Lock() |
| 358 | + defer vmListMutex.Unlock() |
| 359 | + for _, instance := range instances { |
| 360 | + if _, ok := hostNameSet[instance.Name]; ok { |
| 361 | + v := instance.toVM(project, p.dnsProvider.PublicDomain()) |
| 362 | + vmList = append(vmList, *v) |
| 363 | + } |
| 364 | + } |
| 365 | + return nil |
| 366 | + }) |
| 367 | + } |
| 368 | + |
| 369 | + if err := g.Wait(); err != nil { |
| 370 | + return nil, err |
| 371 | + } |
| 372 | + |
| 373 | + return vmList, nil |
| 374 | +} |
0 commit comments