| 1 | import * as aiplatform from "@distilled.cloud/gcp/aiplatform_v1"; |
| 2 | import * as Data from "effect/Data"; |
| 3 | import * as Effect from "effect/Effect"; |
| 4 | import * as Schedule from "effect/Schedule"; |
| 5 | import { Unowned } from "../../AdoptPolicy.ts"; |
| 6 | import { isResolved } from "../../Diff.ts"; |
| 7 | import * as Provider from "../../Provider.ts"; |
| 8 | import { Resource } from "../../Resource.ts"; |
| 9 | import { tagRecord } from "../../Tags.ts"; |
| 10 | import { GcpEnvironment } from "../Environment.ts"; |
| 11 | import { createInternalLabels, hasAlchemyLabels, toLabels } from "../Labels.ts"; |
| 12 | import type { Providers } from "../Providers.ts"; |
| 13 | import { listLocations } from "./names.ts"; |
| 14 | import { resourceNameFromOperation, waitForOperation } from "./operations.ts"; |
| 15 | import { |
| 16 | AiPlatformNotResolved, |
| 17 | AiPlatformStillExists, |
| 18 | collectPages, |
| 19 | jsonEqual, |
| 20 | locationParent, |
| 21 | normalizeLocation, |
| 22 | parseResourceName, |
| 23 | toPhysicalId, |
| 24 | userLabels, |
| 25 | type EncryptionSpec, |
| 26 | type MachineSpec, |
| 27 | type NetworkSpec, |
| 28 | type PersistentDiskSpec, |
| 29 | } from "./shared.ts"; |
| 30 | |
| 31 | const COLLECTION = "notebookRuntimeTemplates"; |
| 32 | const DEFAULT_MACHINE_TYPE = "e2-standard-4"; |
| 33 | |
| 34 | export type NotebookEnvVar = { |
| 35 | /** Environment variable name. */ |
| 36 | name?: string; |
| 37 | /** Environment variable value. */ |
| 38 | value?: string; |
| 39 | }; |
| 40 | |
| 41 | export type ColabImage = { |
| 42 | /** Colab Enterprise release name (e.g. `colab-enterprise-release`). */ |
| 43 | releaseName?: string; |
| 44 | }; |
| 45 | |
| 46 | export type PostStartupScriptConfig = { |
| 47 | /** Inline post-startup script. */ |
| 48 | postStartupScript?: string; |
| 49 | /** Cloud Storage URI of a post-startup script. */ |
| 50 | postStartupScriptUrl?: string; |
| 51 | /** When the script runs (`RUN_ONCE`, `RUN_EVERY_START`). */ |
| 52 | postStartupScriptBehavior?: string; |
| 53 | }; |
| 54 | |
| 55 | export type NotebookSoftwareConfig = { |
| 56 | /** Google-managed Colab image. */ |
| 57 | colabImage?: ColabImage; |
| 58 | /** Environment variables passed to the runtime. */ |
| 59 | env?: NotebookEnvVar[]; |
| 60 | /** Post-startup script configuration. */ |
| 61 | postStartupScriptConfig?: PostStartupScriptConfig; |
| 62 | }; |
| 63 | |
| 64 | export type NotebookReservationAffinity = { |
| 65 | /** Affinity type (`NO_RESERVATION`, `ANY_RESERVATION`, `SPECIFIC_RESERVATION`). */ |
| 66 | reservationAffinityType?: string; |
| 67 | /** Label key of the reservation. */ |
| 68 | key?: string; |
| 69 | /** Reservation resource names. */ |
| 70 | values?: string[]; |
| 71 | }; |
| 72 | |
| 73 | export type NotebookRuntimeTemplateProps = { |
| 74 | /** |
| 75 | * Template id (the `{notebook_runtime_template}` segment). If omitted, |
| 76 | * a unique RFC1035 name is generated. Immutable — changing it replaces |
| 77 | * the template. |
| 78 | */ |
| 79 | notebookRuntimeTemplateId?: string; |
| 80 | /** |
| 81 | * Vertex AI location. Immutable — changing it replaces the template. |
| 82 | * `US-CENTRAL1` is accepted and normalized to `us-central1`. |
| 83 | * @default the stack's GCP region (`GCP.Region`, else the profile region, else `us-central1`) |
| 84 | */ |
| 85 | location?: string; |
| 86 | /** |
| 87 | * Display name (max 128 UTF-8 characters). Defaults to the template id. |
| 88 | */ |
| 89 | displayName?: string; |
| 90 | /** |
| 91 | * Human-readable description. Vertex AI cannot update it, so changing |
| 92 | * it replaces the template. |
| 93 | */ |
| 94 | description?: string; |
| 95 | /** |
| 96 | * User labels. Alchemy ownership labels are merged in automatically. |
| 97 | * Vertex AI cannot update them, so changing them replaces the template. |
| 98 | */ |
| 99 | labels?: Record<string, string>; |
| 100 | /** |
| 101 | * Runtime type. Immutable. |
| 102 | * `USER_DEFINED` or `ONE_CLICK`. |
| 103 | */ |
| 104 | notebookRuntimeType?: string; |
| 105 | /** |
| 106 | * Machine spec. Immutable — changing it replaces the template. |
| 107 | * @default { machineType: "e2-standard-4" } |
| 108 | */ |
| 109 | machineSpec?: MachineSpec; |
| 110 | /** |
| 111 | * Persistent data disk attached to the runtime. |
| 112 | */ |
| 113 | dataPersistentDiskSpec?: PersistentDiskSpec; |
| 114 | /** |
| 115 | * Network spec. |
| 116 | */ |
| 117 | networkSpec?: NetworkSpec; |
| 118 | /** |
| 119 | * Notebook software configuration. |
| 120 | */ |
| 121 | softwareConfig?: NotebookSoftwareConfig; |
| 122 | /** |
| 123 | * Customer-managed encryption key. |
| 124 | */ |
| 125 | encryptionSpec?: EncryptionSpec; |
| 126 | /** |
| 127 | * Compute Engine network tags. |
| 128 | */ |
| 129 | networkTags?: string[]; |
| 130 | /** |
| 131 | * Idle shutdown configuration. |
| 132 | */ |
| 133 | idleShutdownConfig?: { |
| 134 | idleTimeout?: string; |
| 135 | idleShutdownDisabled?: boolean; |
| 136 | }; |
| 137 | /** |
| 138 | * Shielded VM config. Immutable. |
| 139 | */ |
| 140 | shieldedVmConfig?: { enableSecureBoot?: boolean }; |
| 141 | /** |
| 142 | * Reservation affinity. |
| 143 | */ |
| 144 | reservationAffinity?: NotebookReservationAffinity; |
| 145 | }; |
| 146 | |
| 147 | export type NotebookRuntimeTemplate = Resource< |
| 148 | "GCP.AIPlatform.NotebookRuntimeTemplate", |
| 149 | NotebookRuntimeTemplateProps, |
| 150 | { |
| 151 | /** Full resource name. */ |
| 152 | name: string; |
| 153 | /** Template id (last path segment). */ |
| 154 | notebookRuntimeTemplateId: string; |
| 155 | /** Project id. */ |
| 156 | project: string; |
| 157 | /** Location id. */ |
| 158 | location: string; |
| 159 | /** Display name. */ |
| 160 | displayName: string | undefined; |
| 161 | /** Description. */ |
| 162 | description: string | undefined; |
| 163 | /** User labels (Alchemy ownership labels stripped). */ |
| 164 | labels: Record<string, string>; |
| 165 | /** Runtime type. */ |
| 166 | notebookRuntimeType: string | undefined; |
| 167 | /** Machine type. */ |
| 168 | machineType: string | undefined; |
| 169 | /** RFC3339 creation timestamp. */ |
| 170 | createTime: string | undefined; |
| 171 | /** RFC3339 last-update timestamp. */ |
| 172 | updateTime: string | undefined; |
| 173 | /** Etag for read-modify-write. */ |
| 174 | etag: string | undefined; |
| 175 | }, |
| 176 | never, |
| 177 | Providers |
| 178 | >; |
| 179 | |
| 180 | /** |
| 181 | * A Vertex AI notebook runtime template — machine, disk, network, and |
| 182 | * software defaults used to create Colab Enterprise runtimes. |
| 183 | * |
| 184 | * Only `displayName` and `softwareConfig` update in place; changing |
| 185 | * `notebookRuntimeTemplateId`, `location`, `machineSpec`, |
| 186 | * `notebookRuntimeType`, `description`, or `labels` replaces the template. |
| 187 | * |
| 188 | * ### Creating a Template |
| 189 | * **Example:** Generated name |
| 190 | * ```typescript |
| 191 | * const template = yield* GCP.AIPlatform.NotebookRuntimeTemplate("Runtime", { |
| 192 | * machineSpec: { machineType: "e2-standard-4" }, |
| 193 | * }); |
| 194 | * ``` |
| 195 | * |
| 196 | * **Example:** Named template with labels |
| 197 | * ```typescript |
| 198 | * const template = yield* GCP.AIPlatform.NotebookRuntimeTemplate("Runtime", { |
| 199 | * notebookRuntimeTemplateId: "colab-default", |
| 200 | * displayName: "colab default", |
| 201 | * labels: { env: "prod" }, |
| 202 | * machineSpec: { machineType: "e2-standard-4" }, |
| 203 | * networkSpec: { enableInternetAccess: true }, |
| 204 | * }); |
| 205 | * ``` |
| 206 | * |
| 207 | * @resource |
| 208 | * @category AIPlatform |
| 209 | */ |
| 210 | export const NotebookRuntimeTemplate = Resource<NotebookRuntimeTemplate>( |
| 211 | "GCP.AIPlatform.NotebookRuntimeTemplate", |
| 212 | ); |
| 213 | |
| 214 | export class NotebookRuntimeTemplateNotResolved extends Data.TaggedError( |
| 215 | "GCP.AIPlatform.NotebookRuntimeTemplateNotResolved", |
| 216 | )<{ |
| 217 | name: string; |
| 218 | }> {} |
| 219 | |
| 220 | const resourceName = (project: string, location: string, id: string) => |
| 221 | `${locationParent(project, location)}/${COLLECTION}/${id}`; |
| 222 | |
| 223 | const toAttrs = ( |
| 224 | template: aiplatform.GoogleCloudAiplatformV1NotebookRuntimeTemplate, |
| 225 | project: string, |
| 226 | ) => { |
| 227 | const name = template.name ?? ""; |
| 228 | const parsed = parseResourceName(name, COLLECTION); |
| 229 | return { |
| 230 | name, |
| 231 | notebookRuntimeTemplateId: parsed.id, |
| 232 | project: parsed.project || project, |
| 233 | location: parsed.location, |
| 234 | displayName: template.displayName, |
| 235 | description: template.description, |
| 236 | labels: userLabels(template.labels), |
| 237 | notebookRuntimeType: template.notebookRuntimeType, |
| 238 | machineType: template.machineSpec?.machineType, |
| 239 | createTime: template.createTime, |
| 240 | updateTime: template.updateTime, |
| 241 | etag: template.etag, |
| 242 | }; |
| 243 | }; |
| 244 | |
| 245 | const getByName = (name: string) => |
| 246 | aiplatform |
| 247 | .getProjectsLocationsNotebookRuntimeTemplates({ name }) |
| 248 | .pipe(Effect.catchTag("NotFound", () => Effect.succeed(undefined))); |
| 249 | |
| 250 | const waitUntilExists = (name: string) => |
| 251 | getByName(name).pipe( |
| 252 | Effect.filterOrFail( |
| 253 | ( |
| 254 | template, |
| 255 | ): template is aiplatform.GoogleCloudAiplatformV1NotebookRuntimeTemplate => |
| 256 | template !== undefined, |
| 257 | () => new AiPlatformNotResolved({ name }), |
| 258 | ), |
| 259 | Effect.retry({ |
| 260 | while: (error) => error._tag === "GCP.AIPlatform.NotResolved", |
| 261 | times: 8, |
| 262 | schedule: Schedule.spaced("2 seconds"), |
| 263 | }), |
| 264 | ); |
| 265 | |
| 266 | const waitUntilGone = (name: string) => |
| 267 | getByName(name).pipe( |
| 268 | Effect.filterOrFail( |
| 269 | (template) => template === undefined, |
| 270 | () => new AiPlatformStillExists({ name }), |
| 271 | ), |
| 272 | Effect.asVoid, |
| 273 | Effect.retry({ |
| 274 | while: (error) => error._tag === "GCP.AIPlatform.StillExists", |
| 275 | times: 10, |
| 276 | schedule: Schedule.spaced("2 seconds"), |
| 277 | }), |
| 278 | ); |
| 279 | |
| 280 | const desiredMachine = (spec: MachineSpec | undefined): MachineSpec => ({ |
| 281 | machineType: spec?.machineType ?? DEFAULT_MACHINE_TYPE, |
| 282 | acceleratorType: spec?.acceleratorType, |
| 283 | acceleratorCount: spec?.acceleratorCount, |
| 284 | gpuPartitionSize: spec?.gpuPartitionSize, |
| 285 | tpuTopology: spec?.tpuTopology, |
| 286 | }); |
| 287 | |
| 288 | export const NotebookRuntimeTemplateProvider = () => |
| 289 | Provider.succeed(NotebookRuntimeTemplate, { |
| 290 | stables: [ |
| 291 | "name", |
| 292 | "notebookRuntimeTemplateId", |
| 293 | "project", |
| 294 | "location", |
| 295 | "createTime", |
| 296 | ], |
| 297 | |
| 298 | diff: Effect.fn(function* ({ news, olds, output }) { |
| 299 | const env = yield* GcpEnvironment.current; |
| 300 | if (!isResolved(news)) return undefined; |
| 301 | const previousId = |
| 302 | olds?.notebookRuntimeTemplateId ?? output?.notebookRuntimeTemplateId; |
| 303 | const nextId = news.notebookRuntimeTemplateId ?? previousId; |
| 304 | const previousLocation = normalizeLocation( |
| 305 | olds?.location ?? output?.location, |
| 306 | env.region, |
| 307 | ); |
| 308 | const nextLocation = normalizeLocation( |
| 309 | news.location ?? output?.location, |
| 310 | env.region, |
| 311 | ); |
| 312 | const previousMachine = |
| 313 | olds?.machineSpec?.machineType ?? output?.machineType; |
| 314 | const nextMachine = desiredMachine(news.machineSpec).machineType; |
| 315 | const typeChanged = |
| 316 | (news.notebookRuntimeType ?? olds?.notebookRuntimeType) !== |
| 317 | (olds?.notebookRuntimeType ?? output?.notebookRuntimeType); |
| 318 | const replace = |
| 319 | (previousId !== undefined && |
| 320 | nextId !== undefined && |
| 321 | nextId !== previousId) || |
| 322 | previousLocation !== nextLocation || |
| 323 | (previousMachine !== undefined && previousMachine !== nextMachine) || |
| 324 | (olds?.notebookRuntimeType !== undefined && typeChanged) || |
| 325 | (olds !== undefined && |
| 326 | ((news.description ?? "") !== (olds.description ?? "") || |
| 327 | JSON.stringify(news.labels ?? {}) !== |
| 328 | JSON.stringify(olds.labels ?? {}))); |
| 329 | if (!replace) return undefined; |
| 330 | return { |
| 331 | action: "replace" as const, |
| 332 | deleteFirst: |
| 333 | previousLocation === nextLocation && |
| 334 | previousId !== undefined && |
| 335 | nextId === previousId, |
| 336 | }; |
| 337 | }), |
| 338 | |
| 339 | read: Effect.fn(function* ({ id, olds, output }) { |
| 340 | const env = yield* GcpEnvironment.current; |
| 341 | const templateId = yield* toPhysicalId( |
| 342 | id, |
| 343 | olds?.notebookRuntimeTemplateId, |
| 344 | output?.notebookRuntimeTemplateId, |
| 345 | ); |
| 346 | const location = normalizeLocation( |
| 347 | olds?.location ?? output?.location, |
| 348 | env.region, |
| 349 | ); |
| 350 | const name = |
| 351 | output?.name ?? resourceName(env.project, location, templateId); |
| 352 | const existing = yield* getByName(name); |
| 353 | if (existing === undefined) return undefined; |
| 354 | const attrs = toAttrs(existing, env.project); |
| 355 | return (yield* hasAlchemyLabels(id, tagRecord(existing.labels))) |
| 356 | ? attrs |
| 357 | : Unowned(attrs); |
| 358 | }), |
| 359 | |
| 360 | list: () => |
| 361 | Effect.gen(function* () { |
| 362 | const env = yield* GcpEnvironment.current; |
| 363 | const pages = (yield* Effect.forEach( |
| 364 | listLocations(env.region), |
| 365 | (location) => |
| 366 | collectPages( |
| 367 | aiplatform.listProjectsLocationsNotebookRuntimeTemplates.pages({ |
| 368 | parent: locationParent(env.project, location), |
| 369 | pageSize: 100, |
| 370 | }), |
| 371 | ).pipe(Effect.catchTag("NotFound", () => Effect.succeed([]))), |
| 372 | )).flat(); |
| 373 | return pages.flatMap((page) => |
| 374 | (page.notebookRuntimeTemplates ?? []) |
| 375 | .filter((template) => |
| 376 | Object.keys(template.labels ?? {}).some((key) => |
| 377 | key.startsWith("alchemy-"), |
| 378 | ), |
| 379 | ) |
| 380 | .map((template) => toAttrs(template, env.project)), |
| 381 | ); |
| 382 | }), |
| 383 | |
| 384 | reconcile: Effect.fn(function* ({ id, news, output }) { |
| 385 | const env = yield* GcpEnvironment.current; |
| 386 | const templateId = yield* toPhysicalId( |
| 387 | id, |
| 388 | news.notebookRuntimeTemplateId, |
| 389 | output?.notebookRuntimeTemplateId, |
| 390 | ); |
| 391 | const location = normalizeLocation( |
| 392 | news.location ?? output?.location, |
| 393 | env.region, |
| 394 | ); |
| 395 | const name = resourceName(env.project, location, templateId); |
| 396 | const desiredLabels = { |
| 397 | ...toLabels(news.labels), |
| 398 | ...(yield* createInternalLabels(id)), |
| 399 | }; |
| 400 | const displayName = news.displayName ?? templateId; |
| 401 | const machineSpec = desiredMachine(news.machineSpec); |
| 402 | |
| 403 | let current = yield* getByName(output?.name ?? name); |
| 404 | |
| 405 | if (current === undefined) { |
| 406 | const created = yield* aiplatform |
| 407 | .createProjectsLocationsNotebookRuntimeTemplates({ |
| 408 | parent: locationParent(env.project, location), |
| 409 | notebookRuntimeTemplateId: templateId, |
| 410 | body: { |
| 411 | displayName, |
| 412 | description: news.description, |
| 413 | labels: desiredLabels, |
| 414 | notebookRuntimeType: news.notebookRuntimeType, |
| 415 | machineSpec, |
| 416 | dataPersistentDiskSpec: news.dataPersistentDiskSpec, |
| 417 | networkSpec: news.networkSpec, |
| 418 | softwareConfig: news.softwareConfig, |
| 419 | encryptionSpec: news.encryptionSpec, |
| 420 | networkTags: news.networkTags, |
| 421 | idleShutdownConfig: news.idleShutdownConfig, |
| 422 | shieldedVmConfig: news.shieldedVmConfig, |
| 423 | reservationAffinity: news.reservationAffinity, |
| 424 | }, |
| 425 | }) |
| 426 | .pipe(Effect.catchTag("Conflict", () => Effect.succeed(undefined))); |
| 427 | if (created !== undefined) { |
| 428 | yield* waitForOperation(created, { alreadyExistsOk: true }); |
| 429 | } |
| 430 | const createdName = |
| 431 | resourceNameFromOperation(created ?? {}) ?? output?.name ?? name; |
| 432 | current = yield* waitUntilExists(createdName); |
| 433 | } |
| 434 | |
| 435 | if (current === undefined) { |
| 436 | return yield* new NotebookRuntimeTemplateNotResolved({ name }); |
| 437 | } |
| 438 | |
| 439 | const observedName = current.name ?? name; |
| 440 | // Only display name and software config are updatable; label and |
| 441 | // description changes replace the template (see `diff`). |
| 442 | const displayChanged = (current.displayName ?? "") !== displayName; |
| 443 | const softwareChanged = |
| 444 | news.softwareConfig !== undefined && |
| 445 | !jsonEqual(current.softwareConfig, news.softwareConfig); |
| 446 | |
| 447 | if (displayChanged || softwareChanged) { |
| 448 | const updateMask = [ |
| 449 | displayChanged ? "display_name" : undefined, |
| 450 | softwareChanged ? "software_config" : undefined, |
| 451 | ].filter((field): field is string => field !== undefined); |
| 452 | const patched = |
| 453 | yield* aiplatform.patchProjectsLocationsNotebookRuntimeTemplates({ |
| 454 | name: observedName, |
| 455 | updateMask: updateMask.join(","), |
| 456 | body: { |
| 457 | name: observedName, |
| 458 | displayName, |
| 459 | softwareConfig: news.softwareConfig, |
| 460 | etag: current.etag, |
| 461 | }, |
| 462 | }); |
| 463 | current = patched ?? current; |
| 464 | } |
| 465 | |
| 466 | return toAttrs(current, env.project); |
| 467 | }), |
| 468 | |
| 469 | delete: Effect.fn(function* ({ output }) { |
| 470 | const operation = yield* aiplatform |
| 471 | .deleteProjectsLocationsNotebookRuntimeTemplates({ |
| 472 | name: output.name, |
| 473 | }) |
| 474 | .pipe(Effect.catchTag("NotFound", () => Effect.succeed(undefined))); |
| 475 | if (operation !== undefined) { |
| 476 | yield* waitForOperation(operation, { notFoundOk: true }); |
| 477 | } |
| 478 | yield* waitUntilGone(output.name); |
| 479 | }), |
| 480 | }); |