| 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 Stream from "effect/Stream"; |
| 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 { GcpEnvironment } from "../Environment.ts"; |
| 10 | import { createInternalLabels } from "../Labels.ts"; |
| 11 | import type { Providers } from "../Providers.ts"; |
| 12 | import { listLocations } from "./names.ts"; |
| 13 | import { waitForOperation } from "./operations.ts"; |
| 14 | import { |
| 15 | encodeOwnershipLine, |
| 16 | hasOwnershipMarker, |
| 17 | lastSegment, |
| 18 | locationOf, |
| 19 | locationParent, |
| 20 | ownedByAlchemy, |
| 21 | parseOwnership, |
| 22 | } from "./ownership.ts"; |
| 23 | |
| 24 | export type PipelineJobRequest = { |
| 25 | /** |
| 26 | * Location parent for created pipeline jobs |
| 27 | * (`projects/{project}/locations/{location}`). Defaults to the |
| 28 | * Schedule's location. |
| 29 | */ |
| 30 | parent?: string; |
| 31 | /** Optional pipeline job id. */ |
| 32 | pipelineJobId?: string; |
| 33 | /** Pipeline job to create on each run. */ |
| 34 | pipelineJob?: { |
| 35 | displayName?: string; |
| 36 | templateUri?: string; |
| 37 | pipelineSpec?: Record<string, unknown>; |
| 38 | serviceAccount?: string; |
| 39 | network?: string; |
| 40 | labels?: Record<string, string>; |
| 41 | runtimeConfig?: { |
| 42 | gcsOutputDirectory?: string; |
| 43 | parameterValues?: Record<string, unknown>; |
| 44 | }; |
| 45 | }; |
| 46 | }; |
| 47 | |
| 48 | export type NotebookExecutionJobRequest = { |
| 49 | parent?: string; |
| 50 | notebookExecutionJobId?: string; |
| 51 | notebookExecutionJob?: Record<string, unknown>; |
| 52 | }; |
| 53 | |
| 54 | export type ScheduleProps = { |
| 55 | /** |
| 56 | * Region (`us-central1`, …). Immutable — changing it replaces the |
| 57 | * Schedule. |
| 58 | * @default the stack's GCP region (`GCP.Region`, else the profile region, else `us-central1`) |
| 59 | */ |
| 60 | location?: string; |
| 61 | /** |
| 62 | * User-facing display name. Vertex AI Schedules have no labels field, |
| 63 | * so Alchemy ownership is stored in a `[alchemy …]` prefix. |
| 64 | */ |
| 65 | displayName?: string; |
| 66 | /** |
| 67 | * Cron schedule. Prefix with `CRON_TZ=${IANA_TIME_ZONE}` to set a |
| 68 | * timezone (for example `CRON_TZ=UTC 0 0 * * *`). |
| 69 | */ |
| 70 | cron: string; |
| 71 | /** |
| 72 | * Maximum number of runs that can be started concurrently. |
| 73 | * @default "1" |
| 74 | */ |
| 75 | maxConcurrentRunCount?: string; |
| 76 | /** |
| 77 | * Maximum number of active (non-terminal) pipeline runs. Only applies |
| 78 | * to `createPipelineJobRequest`. |
| 79 | */ |
| 80 | maxConcurrentActiveRunCount?: string; |
| 81 | /** |
| 82 | * Maximum number of runs before the schedule completes. |
| 83 | */ |
| 84 | maxRunCount?: string; |
| 85 | /** |
| 86 | * RFC3339 timestamp after which the first run may be scheduled. |
| 87 | */ |
| 88 | startTime?: string; |
| 89 | /** |
| 90 | * RFC3339 timestamp after which no new runs are scheduled. |
| 91 | */ |
| 92 | endTime?: string; |
| 93 | /** |
| 94 | * Queue new runs when `maxConcurrentRunCount` is reached instead of |
| 95 | * skipping them. |
| 96 | * @default false |
| 97 | */ |
| 98 | allowQueueing?: boolean; |
| 99 | /** |
| 100 | * Pipeline job created on each tick. Exactly one of |
| 101 | * `createPipelineJobRequest` or `createNotebookExecutionJobRequest` |
| 102 | * must be set. |
| 103 | */ |
| 104 | createPipelineJobRequest?: PipelineJobRequest; |
| 105 | /** |
| 106 | * Notebook execution job created on each tick. |
| 107 | */ |
| 108 | createNotebookExecutionJobRequest?: NotebookExecutionJobRequest; |
| 109 | /** |
| 110 | * When true, the schedule exists but does not start new runs (`PAUSED`). |
| 111 | * Synced via `schedules.pause` / `schedules.resume`. |
| 112 | * @default false |
| 113 | */ |
| 114 | paused?: boolean; |
| 115 | }; |
| 116 | |
| 117 | export type Schedule = Resource< |
| 118 | "GCP.AIPlatform.Schedule", |
| 119 | ScheduleProps, |
| 120 | { |
| 121 | /** Full resource name `projects/{project}/locations/{location}/schedules/{schedule}`. */ |
| 122 | name: string; |
| 123 | /** Schedule id (last path segment). */ |
| 124 | scheduleId: string; |
| 125 | /** Region id. */ |
| 126 | location: string; |
| 127 | /** Project id. */ |
| 128 | project: string; |
| 129 | /** User display name with the Alchemy ownership prefix stripped. */ |
| 130 | displayName: string | undefined; |
| 131 | /** Cron expression. */ |
| 132 | cron: string | undefined; |
| 133 | /** Server-reported state. */ |
| 134 | state: string | undefined; |
| 135 | /** Whether the schedule is paused. */ |
| 136 | paused: boolean; |
| 137 | /** Maximum concurrent starts. */ |
| 138 | maxConcurrentRunCount: string | undefined; |
| 139 | /** Maximum concurrent active pipeline runs. */ |
| 140 | maxConcurrentActiveRunCount: string | undefined; |
| 141 | /** Maximum run count. */ |
| 142 | maxRunCount: string | undefined; |
| 143 | /** RFC3339 start bound. */ |
| 144 | startTime: string | undefined; |
| 145 | /** RFC3339 end bound. */ |
| 146 | endTime: string | undefined; |
| 147 | /** Whether new runs may queue. */ |
| 148 | allowQueueing: boolean; |
| 149 | /** Number of runs started so far. */ |
| 150 | startedRunCount: string | undefined; |
| 151 | /** Next scheduled run (RFC3339). */ |
| 152 | nextRunTime: string | undefined; |
| 153 | /** RFC3339 creation timestamp. */ |
| 154 | createTime: string | undefined; |
| 155 | /** RFC3339 last-update timestamp. */ |
| 156 | updateTime: string | undefined; |
| 157 | }, |
| 158 | never, |
| 159 | Providers |
| 160 | >; |
| 161 | |
| 162 | /** |
| 163 | * A Vertex AI Schedule that periodically starts pipeline or notebook jobs. |
| 164 | * |
| 165 | * Schedules have no labels field — Alchemy stamps ownership into the |
| 166 | * display name. Location is immutable. Cron, run limits, and pause state |
| 167 | * update in place. |
| 168 | * |
| 169 | * ### Creating a Schedule |
| 170 | * **Example:** Daily paused pipeline schedule |
| 171 | * ```typescript |
| 172 | * const schedule = yield* GCP.AIPlatform.Schedule("Nightly", { |
| 173 | * cron: "CRON_TZ=UTC 0 8 * * *", |
| 174 | * paused: true, |
| 175 | * maxRunCount: "1", |
| 176 | * createPipelineJobRequest: { |
| 177 | * pipelineJob: { |
| 178 | * displayName: "nightly", |
| 179 | * templateUri: "https://us-kfp.pkg.dev/ml-pipeline/google-cloud-registry/hello-world/latest", |
| 180 | * }, |
| 181 | * }, |
| 182 | * }); |
| 183 | * ``` |
| 184 | * |
| 185 | * @resource |
| 186 | * @category AIPlatform |
| 187 | */ |
| 188 | export const Schedule = Resource<Schedule>("GCP.AIPlatform.Schedule"); |
| 189 | |
| 190 | export class ScheduleNotResolved extends Data.TaggedError( |
| 191 | "GCP.AIPlatform.ScheduleNotResolved", |
| 192 | )<{ |
| 193 | name: string; |
| 194 | }> {} |
| 195 | |
| 196 | const toAttrs = ( |
| 197 | schedule: aiplatform.GoogleCloudAiplatformV1Schedule, |
| 198 | project: string, |
| 199 | ) => { |
| 200 | const name = schedule.name ?? ""; |
| 201 | const parsed = parseOwnership(schedule.displayName); |
| 202 | return { |
| 203 | name, |
| 204 | scheduleId: lastSegment(name), |
| 205 | location: locationOf(name), |
| 206 | project, |
| 207 | displayName: parsed.text, |
| 208 | cron: schedule.cron, |
| 209 | state: schedule.state, |
| 210 | paused: schedule.state === "PAUSED", |
| 211 | maxConcurrentRunCount: schedule.maxConcurrentRunCount, |
| 212 | maxConcurrentActiveRunCount: schedule.maxConcurrentActiveRunCount, |
| 213 | maxRunCount: schedule.maxRunCount, |
| 214 | startTime: schedule.startTime, |
| 215 | endTime: schedule.endTime, |
| 216 | allowQueueing: schedule.allowQueueing === true, |
| 217 | startedRunCount: schedule.startedRunCount, |
| 218 | nextRunTime: schedule.nextRunTime, |
| 219 | createTime: schedule.createTime, |
| 220 | updateTime: schedule.updateTime, |
| 221 | }; |
| 222 | }; |
| 223 | |
| 224 | const getByName = (name: string) => |
| 225 | name.length === 0 |
| 226 | ? Effect.succeed(undefined) |
| 227 | : aiplatform |
| 228 | .getProjectsLocationsSchedules({ name }) |
| 229 | .pipe(Effect.catchTag("NotFound", () => Effect.succeed(undefined))); |
| 230 | |
| 231 | const listAt = (parent: string, project: string) => |
| 232 | aiplatform.listProjectsLocationsSchedules |
| 233 | .pages({ parent, pageSize: 100 }) |
| 234 | .pipe( |
| 235 | Stream.flatMap((page) => Stream.fromIterable(page.schedules ?? [])), |
| 236 | Stream.filter((schedule) => hasOwnershipMarker(schedule.displayName)), |
| 237 | Stream.map((schedule) => toAttrs(schedule, project)), |
| 238 | Stream.runCollect, |
| 239 | Effect.map((chunk) => Array.from(chunk)), |
| 240 | Effect.catchTag("NotFound", () => Effect.succeed([])), |
| 241 | ); |
| 242 | |
| 243 | const toPipelineRequest = ( |
| 244 | request: PipelineJobRequest | undefined, |
| 245 | parent: string, |
| 246 | ): aiplatform.GoogleCloudAiplatformV1CreatePipelineJobRequest | undefined => { |
| 247 | if (request === undefined) return undefined; |
| 248 | const job = request.pipelineJob; |
| 249 | return { |
| 250 | parent: request.parent ?? parent, |
| 251 | pipelineJobId: request.pipelineJobId, |
| 252 | pipelineJob: job |
| 253 | ? { |
| 254 | displayName: job.displayName, |
| 255 | templateUri: job.templateUri, |
| 256 | pipelineSpec: job.pipelineSpec, |
| 257 | serviceAccount: job.serviceAccount, |
| 258 | network: job.network, |
| 259 | labels: job.labels, |
| 260 | runtimeConfig: job.runtimeConfig, |
| 261 | } |
| 262 | : undefined, |
| 263 | }; |
| 264 | }; |
| 265 | |
| 266 | const toNotebookRequest = ( |
| 267 | request: NotebookExecutionJobRequest | undefined, |
| 268 | ): |
| 269 | | aiplatform.GoogleCloudAiplatformV1CreateNotebookExecutionJobRequest |
| 270 | | undefined => { |
| 271 | if (request === undefined) return undefined; |
| 272 | return { |
| 273 | parent: request.parent, |
| 274 | notebookExecutionJobId: request.notebookExecutionJobId, |
| 275 | notebookExecutionJob: |
| 276 | request.notebookExecutionJob as aiplatform.GoogleCloudAiplatformV1NotebookExecutionJob, |
| 277 | }; |
| 278 | }; |
| 279 | |
| 280 | export const ScheduleProvider = () => |
| 281 | Provider.succeed(Schedule, { |
| 282 | stables: ["name", "scheduleId", "location", "project", "createTime"], |
| 283 | |
| 284 | diff: Effect.fn(function* ({ news, olds, output }) { |
| 285 | if (!isResolved(news)) return undefined; |
| 286 | const previousLocation = olds?.location ?? output?.location; |
| 287 | const nextLocation = news.location ?? previousLocation; |
| 288 | if (previousLocation !== undefined && previousLocation !== nextLocation) { |
| 289 | return { action: "replace" as const, deleteFirst: false }; |
| 290 | } |
| 291 | return undefined; |
| 292 | }), |
| 293 | |
| 294 | read: Effect.fn(function* ({ id, olds, output }) { |
| 295 | const env = yield* GcpEnvironment.current; |
| 296 | const existing = yield* getByName(output?.name ?? ""); |
| 297 | if (existing !== undefined) { |
| 298 | const attrs = toAttrs(existing, env.project); |
| 299 | return (yield* ownedByAlchemy(id, existing.displayName)) |
| 300 | ? attrs |
| 301 | : Unowned(attrs); |
| 302 | } |
| 303 | const location = olds?.location ?? output?.location ?? env.region; |
| 304 | const parent = locationParent(env.project, location); |
| 305 | const ownership = yield* createInternalLabels(id); |
| 306 | const displayName = encodeOwnershipLine(ownership, olds?.displayName); |
| 307 | const found = yield* aiplatform.listProjectsLocationsSchedules |
| 308 | .pages({ parent, pageSize: 100 }) |
| 309 | .pipe( |
| 310 | Stream.flatMap((page) => Stream.fromIterable(page.schedules ?? [])), |
| 311 | Stream.filter((schedule) => schedule.displayName === displayName), |
| 312 | Stream.runHead, |
| 313 | Effect.map((option) => |
| 314 | option._tag === "Some" ? option.value : undefined, |
| 315 | ), |
| 316 | Effect.catchTag("NotFound", () => Effect.succeed(undefined)), |
| 317 | ); |
| 318 | if (found === undefined) return undefined; |
| 319 | const attrs = toAttrs(found, env.project); |
| 320 | return (yield* ownedByAlchemy(id, found.displayName)) |
| 321 | ? attrs |
| 322 | : Unowned(attrs); |
| 323 | }), |
| 324 | |
| 325 | list: () => |
| 326 | Effect.gen(function* () { |
| 327 | const env = yield* GcpEnvironment.current; |
| 328 | return (yield* Effect.forEach(listLocations(env.region), (location) => |
| 329 | listAt(locationParent(env.project, location), env.project), |
| 330 | )).flat(); |
| 331 | }), |
| 332 | |
| 333 | reconcile: Effect.fn(function* ({ id, news, output }) { |
| 334 | const env = yield* GcpEnvironment.current; |
| 335 | const location = news.location ?? output?.location ?? env.region; |
| 336 | const parent = locationParent(env.project, location); |
| 337 | const ownership = yield* createInternalLabels(id); |
| 338 | const displayName = encodeOwnershipLine(ownership, news.displayName); |
| 339 | const maxConcurrentRunCount = news.maxConcurrentRunCount ?? "1"; |
| 340 | const desiredPaused = news.paused === true; |
| 341 | const pipelineRequest = toPipelineRequest( |
| 342 | news.createPipelineJobRequest, |
| 343 | parent, |
| 344 | ); |
| 345 | const notebookRequest = toNotebookRequest( |
| 346 | news.createNotebookExecutionJobRequest, |
| 347 | ); |
| 348 | |
| 349 | let current = yield* getByName(output?.name ?? ""); |
| 350 | if (current === undefined) { |
| 351 | current = yield* aiplatform.listProjectsLocationsSchedules |
| 352 | .pages({ parent, pageSize: 100 }) |
| 353 | .pipe( |
| 354 | Stream.flatMap((page) => Stream.fromIterable(page.schedules ?? [])), |
| 355 | Stream.filter((schedule) => schedule.displayName === displayName), |
| 356 | Stream.runHead, |
| 357 | Effect.map((option) => |
| 358 | option._tag === "Some" ? option.value : undefined, |
| 359 | ), |
| 360 | Effect.catchTag("NotFound", () => Effect.succeed(undefined)), |
| 361 | ); |
| 362 | } |
| 363 | |
| 364 | if (current === undefined) { |
| 365 | const created = yield* aiplatform |
| 366 | .createProjectsLocationsSchedules({ |
| 367 | parent, |
| 368 | body: { |
| 369 | displayName, |
| 370 | cron: news.cron, |
| 371 | maxConcurrentRunCount, |
| 372 | maxConcurrentActiveRunCount: news.maxConcurrentActiveRunCount, |
| 373 | maxRunCount: news.maxRunCount, |
| 374 | startTime: news.startTime, |
| 375 | endTime: news.endTime, |
| 376 | allowQueueing: news.allowQueueing === true ? true : undefined, |
| 377 | createPipelineJobRequest: pipelineRequest, |
| 378 | createNotebookExecutionJobRequest: notebookRequest, |
| 379 | }, |
| 380 | }) |
| 381 | .pipe(Effect.catchTag("Conflict", () => Effect.succeed(undefined))); |
| 382 | current = created ?? undefined; |
| 383 | if (current === undefined) { |
| 384 | current = yield* aiplatform.listProjectsLocationsSchedules |
| 385 | .pages({ parent, pageSize: 100 }) |
| 386 | .pipe( |
| 387 | Stream.flatMap((page) => |
| 388 | Stream.fromIterable(page.schedules ?? []), |
| 389 | ), |
| 390 | Stream.filter((schedule) => schedule.displayName === displayName), |
| 391 | Stream.runHead, |
| 392 | Effect.map((option) => |
| 393 | option._tag === "Some" ? option.value : undefined, |
| 394 | ), |
| 395 | Effect.catchTag("NotFound", () => Effect.succeed(undefined)), |
| 396 | ); |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | if (current === undefined) { |
| 401 | return yield* new ScheduleNotResolved({ |
| 402 | name: output?.name ?? `${parent}/schedules/-`, |
| 403 | }); |
| 404 | } |
| 405 | |
| 406 | const name = current.name ?? ""; |
| 407 | const cronChanged = (current.cron ?? "") !== news.cron; |
| 408 | const displayChanged = (current.displayName ?? "") !== displayName; |
| 409 | const concurrentChanged = |
| 410 | (current.maxConcurrentRunCount ?? "") !== maxConcurrentRunCount; |
| 411 | const activeChanged = |
| 412 | (current.maxConcurrentActiveRunCount ?? "") !== |
| 413 | (news.maxConcurrentActiveRunCount ?? ""); |
| 414 | const maxRunChanged = |
| 415 | (current.maxRunCount ?? "") !== (news.maxRunCount ?? ""); |
| 416 | const startChanged = (current.startTime ?? "") !== (news.startTime ?? ""); |
| 417 | const endChanged = (current.endTime ?? "") !== (news.endTime ?? ""); |
| 418 | const queueChanged = |
| 419 | (current.allowQueueing === true) !== (news.allowQueueing === true); |
| 420 | |
| 421 | if ( |
| 422 | cronChanged || |
| 423 | displayChanged || |
| 424 | concurrentChanged || |
| 425 | activeChanged || |
| 426 | maxRunChanged || |
| 427 | startChanged || |
| 428 | endChanged || |
| 429 | queueChanged |
| 430 | ) { |
| 431 | current = yield* aiplatform.patchProjectsLocationsSchedules({ |
| 432 | name, |
| 433 | updateMask: [ |
| 434 | cronChanged ? "cron" : undefined, |
| 435 | displayChanged ? "display_name" : undefined, |
| 436 | concurrentChanged ? "max_concurrent_run_count" : undefined, |
| 437 | activeChanged ? "max_concurrent_active_run_count" : undefined, |
| 438 | maxRunChanged ? "max_run_count" : undefined, |
| 439 | startChanged ? "start_time" : undefined, |
| 440 | endChanged ? "end_time" : undefined, |
| 441 | queueChanged ? "allow_queueing" : undefined, |
| 442 | ] |
| 443 | .filter((field): field is string => field !== undefined) |
| 444 | .join(","), |
| 445 | body: { |
| 446 | name, |
| 447 | displayName, |
| 448 | cron: news.cron, |
| 449 | maxConcurrentRunCount, |
| 450 | maxConcurrentActiveRunCount: news.maxConcurrentActiveRunCount, |
| 451 | maxRunCount: news.maxRunCount, |
| 452 | startTime: news.startTime, |
| 453 | endTime: news.endTime, |
| 454 | allowQueueing: news.allowQueueing === true, |
| 455 | }, |
| 456 | }); |
| 457 | } |
| 458 | |
| 459 | const isPaused = current.state === "PAUSED"; |
| 460 | if (desiredPaused && !isPaused) { |
| 461 | yield* aiplatform.pauseProjectsLocationsSchedules({ name }); |
| 462 | current = (yield* getByName(name)) ?? current; |
| 463 | } else if (!desiredPaused && isPaused) { |
| 464 | yield* aiplatform.resumeProjectsLocationsSchedules({ |
| 465 | name, |
| 466 | body: { catchUp: false }, |
| 467 | }); |
| 468 | current = (yield* getByName(name)) ?? current; |
| 469 | } |
| 470 | |
| 471 | return toAttrs(current, env.project); |
| 472 | }), |
| 473 | |
| 474 | delete: Effect.fn(function* ({ output }) { |
| 475 | const existing = yield* getByName(output.name); |
| 476 | if (existing === undefined) return; |
| 477 | const operation = yield* aiplatform |
| 478 | .deleteProjectsLocationsSchedules({ name: output.name }) |
| 479 | .pipe(Effect.catchTag("NotFound", () => Effect.succeed(undefined))); |
| 480 | if (operation !== undefined) { |
| 481 | yield* waitForOperation(operation, { notFoundOk: true }); |
| 482 | } |
| 483 | }), |
| 484 | }); |