The best way to control Claude Code, Codex and more from any computer or phone
Mirrored from https://github.com/Camo67/t3code.git
Synced Nov 21, 06:00 UTC
| 1 | import * as emr from "@distilled.cloud/aws/emr"; |
| 2 | import { Unowned } from "alchemy/AdoptPolicy"; |
| 3 | import { isResolved } from "alchemy/Diff"; |
| 4 | import { createPhysicalName } from "alchemy/PhysicalName"; |
| 5 | import * as Provider from "alchemy/Provider"; |
| 6 | import { Resource } from "alchemy/Resource"; |
| 7 | import { createInternalTags, diffTags, hasAlchemyTags } from "alchemy/Tags"; |
| 8 | import { toWireSeconds } from "alchemy/Util/Duration"; |
| 9 | import * as Data from "effect/Data"; |
| 10 | import type * as Duration from "effect/Duration"; |
| 11 | import * as Effect from "effect/Effect"; |
| 12 | import * as Option from "effect/Option"; |
| 13 | import * as Schedule from "effect/Schedule"; |
| 14 | import * as Stream from "effect/Stream"; |
| 15 | import type { Providers } from "../Providers.ts"; |
| 16 | |
| 17 | export class EMRClusterOperationFailed extends Data.TaggedError("EMRClusterOperationFailed")<{ |
| 18 | readonly message: string; |
| 19 | readonly cause?: unknown; |
| 20 | }> {} |
| 21 | |
| 22 | export class EMRClusterNotReady extends Data.TaggedError("EMRClusterNotReady")<{ |
| 23 | readonly message: string; |
| 24 | readonly cause?: unknown; |
| 25 | }> {} |
| 26 | |
| 27 | export class EMRClusterUnexpectedResponse extends Data.TaggedError("EMRClusterUnexpectedResponse")<{ |
| 28 | readonly message: string; |
| 29 | readonly cause?: unknown; |
| 30 | }> {} |
| 31 | |
| 32 | /** |
| 33 | * EC2 instance topology for an EMR {@link Cluster} — a master (primary) |
| 34 | * instance group plus an optional core instance group, using the uniform |
| 35 | * instance-groups configuration. |
| 36 | */ |
| 37 | export interface ClusterInstancesConfig { |
| 38 | /** |
| 39 | * EC2 instance type of the single master (primary) node. Changing this |
| 40 | * replaces the cluster. |
| 41 | * @default "m5.xlarge" |
| 42 | */ |
| 43 | masterInstanceType?: string; |
| 44 | /** |
| 45 | * EC2 instance type of the core nodes. Changing this replaces the cluster. |
| 46 | * @default "m5.xlarge" |
| 47 | */ |
| 48 | coreInstanceType?: string; |
| 49 | /** |
| 50 | * Number of core nodes. `0` launches a master-only cluster. Resizing |
| 51 | * between non-zero counts is applied in place; crossing the `0` boundary |
| 52 | * (adding or removing the core group entirely) replaces the cluster. |
| 53 | * @default 1 |
| 54 | */ |
| 55 | coreInstanceCount?: number; |
| 56 | /** |
| 57 | * VPC subnet the cluster instances launch in. Changing the subnet replaces |
| 58 | * the cluster. |
| 59 | * @default an EMR-chosen subnet of the account's default VPC |
| 60 | */ |
| 61 | ec2SubnetId?: string; |
| 62 | /** |
| 63 | * EC2 key pair for SSH access to the master node. Changing this replaces |
| 64 | * the cluster. |
| 65 | */ |
| 66 | ec2KeyName?: string; |
| 67 | /** |
| 68 | * Whether the cluster stays alive (transitions to `WAITING`) after all |
| 69 | * steps complete rather than auto-terminating. Updateable in place. |
| 70 | * @default true |
| 71 | */ |
| 72 | keepJobFlowAliveWhenNoSteps?: boolean; |
| 73 | /** |
| 74 | * Whether the cluster's EC2 instances are locked against termination by |
| 75 | * API calls or user intervention. Updateable in place (and always lifted |
| 76 | * before the provider terminates the cluster on destroy). |
| 77 | * @default false |
| 78 | */ |
| 79 | terminationProtected?: boolean; |
| 80 | /** |
| 81 | * Whether EMR automatically replaces unhealthy core/task nodes. |
| 82 | * Updateable in place. |
| 83 | */ |
| 84 | unhealthyNodeReplacement?: boolean; |
| 85 | /** |
| 86 | * EMR-managed security group for the master node. Changing this replaces |
| 87 | * the cluster. |
| 88 | * @default a security group created by EMR |
| 89 | */ |
| 90 | emrManagedMasterSecurityGroup?: string; |
| 91 | /** |
| 92 | * EMR-managed security group for core/task nodes. Changing this replaces |
| 93 | * the cluster. |
| 94 | * @default a security group created by EMR |
| 95 | */ |
| 96 | emrManagedSlaveSecurityGroup?: string; |
| 97 | /** |
| 98 | * Additional security groups applied to the master node. Changing these |
| 99 | * replaces the cluster. |
| 100 | */ |
| 101 | additionalMasterSecurityGroups?: string[]; |
| 102 | /** |
| 103 | * Additional security groups applied to core/task nodes. Changing these |
| 104 | * replaces the cluster. |
| 105 | */ |
| 106 | additionalSlaveSecurityGroups?: string[]; |
| 107 | } |
| 108 | |
| 109 | export interface ClusterProps { |
| 110 | /** |
| 111 | * Display name of the cluster. If omitted, a deterministic physical name is |
| 112 | * generated. Changing the name replaces the cluster. |
| 113 | */ |
| 114 | clusterName?: string; |
| 115 | /** |
| 116 | * Amazon EMR release, e.g. `"emr-7.5.0"`. Changing the release replaces the |
| 117 | * cluster. |
| 118 | */ |
| 119 | releaseLabel: string; |
| 120 | /** |
| 121 | * Applications installed on the cluster, e.g. `["Spark", "Hadoop"]`. |
| 122 | * Changing the set replaces the cluster. |
| 123 | */ |
| 124 | applications?: string[]; |
| 125 | /** |
| 126 | * EC2 instance topology (master/core instance groups, subnet, key pair, |
| 127 | * security groups). |
| 128 | */ |
| 129 | instances?: ClusterInstancesConfig; |
| 130 | /** |
| 131 | * IAM service role EMR assumes to manage cluster resources, e.g. a role |
| 132 | * trusting `elasticmapreduce.amazonaws.com`. Changing the role replaces |
| 133 | * the cluster. |
| 134 | */ |
| 135 | serviceRole: string; |
| 136 | /** |
| 137 | * IAM instance profile (also called the job-flow role or EC2 instance |
| 138 | * profile) assumed by the cluster's EC2 instances. Changing it replaces |
| 139 | * the cluster. |
| 140 | */ |
| 141 | jobFlowRole: string; |
| 142 | /** |
| 143 | * S3 URI where cluster logs are written, e.g. `"s3://my-bucket/logs/"`. |
| 144 | * Changing it replaces the cluster. |
| 145 | */ |
| 146 | logUri?: string; |
| 147 | /** |
| 148 | * Name of an EMR {@link SecurityConfiguration} applied at launch. Changing |
| 149 | * it replaces the cluster. |
| 150 | */ |
| 151 | securityConfiguration?: string; |
| 152 | /** |
| 153 | * Application configuration overrides (raw EMR `Configuration` objects, |
| 154 | * e.g. `[{ Classification: "spark-defaults", Properties: {...} }]`). |
| 155 | * Changing them replaces the cluster. |
| 156 | */ |
| 157 | configurations?: emr.Configuration[]; |
| 158 | /** |
| 159 | * Whether all IAM principals in the account can see the cluster. |
| 160 | * Updateable in place. |
| 161 | * @default true |
| 162 | */ |
| 163 | visibleToAllUsers?: boolean; |
| 164 | /** |
| 165 | * Size of the EBS root volume (GiB) of each instance. Changing it replaces |
| 166 | * the cluster. |
| 167 | */ |
| 168 | ebsRootVolumeSize?: number; |
| 169 | /** |
| 170 | * Custom AMI ID for the cluster instances. Changing it replaces the |
| 171 | * cluster. |
| 172 | */ |
| 173 | customAmiId?: string; |
| 174 | /** |
| 175 | * Number of steps that may execute concurrently (1-256). Updateable in |
| 176 | * place. |
| 177 | * @default 1 |
| 178 | */ |
| 179 | stepConcurrencyLevel?: number; |
| 180 | /** |
| 181 | * Auto-termination policy — the cluster terminates itself after being idle |
| 182 | * for the configured duration (60 seconds to 7 days). Updateable in place; |
| 183 | * removing the prop detaches the policy. |
| 184 | */ |
| 185 | autoTerminationPolicy?: { |
| 186 | /** |
| 187 | * Idle time after which the cluster auto-terminates — e.g. `"1 hour"` or |
| 188 | * `Duration.hours(1)`. Sent to AWS as whole seconds. |
| 189 | */ |
| 190 | idleTimeout: Duration.Input; |
| 191 | }; |
| 192 | /** |
| 193 | * User-defined tags for the cluster. |
| 194 | */ |
| 195 | tags?: Record<string, string>; |
| 196 | } |
| 197 | |
| 198 | export interface Cluster extends Resource< |
| 199 | "AWS.EMR.Cluster", |
| 200 | ClusterProps, |
| 201 | { |
| 202 | /** The ID of the cluster (e.g. `j-2AXXXXXXGAPLF`). */ |
| 203 | clusterId: string; |
| 204 | /** The ARN of the cluster. */ |
| 205 | clusterArn: string; |
| 206 | /** The name of the cluster. */ |
| 207 | clusterName: string; |
| 208 | /** The cluster state (e.g. `STARTING`, `RUNNING`, `WAITING`). */ |
| 209 | state: string; |
| 210 | /** The public DNS name of the primary node, when reachable. */ |
| 211 | masterPublicDnsName: string | undefined; |
| 212 | /** The tags applied to the cluster. */ |
| 213 | tags: Record<string, string>; |
| 214 | }, |
| 215 | never, |
| 216 | Providers |
| 217 | > {} |
| 218 | |
| 219 | /** |
| 220 | * A provisioned Amazon EMR cluster (job flow) running open-source big-data |
| 221 | * frameworks such as Apache Spark and Hadoop on EC2 instances. |
| 222 | * |
| 223 | * Clusters take roughly 10-15 minutes to reach `WAITING` and bill per |
| 224 | * instance-hour while they exist. Each cluster needs an EMR service role and |
| 225 | * an EC2 instance profile (job-flow role); destroy clusters you are not |
| 226 | * using, or set `autoTerminationPolicy` as a safety net. |
| 227 | * ### Creating a Cluster |
| 228 | * **Example:** Spark Cluster in a Default-VPC Subnet |
| 229 | * ```typescript |
| 230 | * const cluster = yield* Cluster("Analytics", { |
| 231 | * releaseLabel: "emr-7.5.0", |
| 232 | * applications: ["Spark", "Hadoop"], |
| 233 | * serviceRole: serviceRole.roleName, |
| 234 | * jobFlowRole: instanceProfile.instanceProfileName, |
| 235 | * logUri: Output.interpolate`s3://${logsBucket.bucketName}/logs/`, |
| 236 | * instances: { |
| 237 | * masterInstanceType: "m5.xlarge", |
| 238 | * coreInstanceType: "m5.xlarge", |
| 239 | * coreInstanceCount: 1, |
| 240 | * ec2SubnetId: subnetId, |
| 241 | * }, |
| 242 | * }); |
| 243 | * ``` |
| 244 | * |
| 245 | * **Example:** Cluster with an Auto-Termination Safety Net |
| 246 | * ```typescript |
| 247 | * const cluster = yield* Cluster("Batch", { |
| 248 | * releaseLabel: "emr-7.5.0", |
| 249 | * applications: ["Spark"], |
| 250 | * serviceRole: serviceRole.roleName, |
| 251 | * jobFlowRole: instanceProfile.instanceProfileName, |
| 252 | * autoTerminationPolicy: { idleTimeout: "1 hour" }, |
| 253 | * stepConcurrencyLevel: 4, |
| 254 | * }); |
| 255 | * ``` |
| 256 | * |
| 257 | * ### Applying a Security Configuration |
| 258 | * **Example:** Cluster with Encryption Settings |
| 259 | * ```typescript |
| 260 | * const config = yield* SecurityConfiguration("Encryption", { |
| 261 | * securityConfiguration: { |
| 262 | * EncryptionConfiguration: { |
| 263 | * EnableInTransitEncryption: false, |
| 264 | * EnableAtRestEncryption: false, |
| 265 | * }, |
| 266 | * }, |
| 267 | * }); |
| 268 | * const cluster = yield* Cluster("Secure", { |
| 269 | * releaseLabel: "emr-7.5.0", |
| 270 | * serviceRole: serviceRole.roleName, |
| 271 | * jobFlowRole: instanceProfile.instanceProfileName, |
| 272 | * securityConfiguration: config.securityConfigurationName, |
| 273 | * }); |
| 274 | * ``` |
| 275 | * |
| 276 | * @resource |
| 277 | */ |
| 278 | export const Cluster = Resource<Cluster>("AWS.EMR.Cluster"); |
| 279 | |
| 280 | // States in which a cluster is (still) a live, converging resource. |
| 281 | const ACTIVE_STATES = ["STARTING", "BOOTSTRAPPING", "RUNNING", "WAITING"] as const; |
| 282 | |
| 283 | // Terminal states: the cluster is gone or irreversibly going away. |
| 284 | const TERMINAL_STATES = new Set(["TERMINATING", "TERMINATED", "TERMINATED_WITH_ERRORS"]); |
| 285 | |
| 286 | const toTagRecord = ( |
| 287 | tags: ReadonlyArray<{ Key?: string; Value?: string }> | undefined, |
| 288 | ): Record<string, string> => |
| 289 | Object.fromEntries( |
| 290 | (tags ?? []) |
| 291 | .filter( |
| 292 | (tag): tag is { Key: string; Value: string } => |
| 293 | typeof tag.Key === "string" && typeof tag.Value === "string", |
| 294 | ) |
| 295 | .map((tag) => [tag.Key, tag.Value]), |
| 296 | ); |
| 297 | |
| 298 | const sameStringSet = ( |
| 299 | a: readonly string[] | undefined, |
| 300 | b: readonly string[] | undefined, |
| 301 | ): boolean => { |
| 302 | const left = [...(a ?? [])].sort(); |
| 303 | const right = [...(b ?? [])].sort(); |
| 304 | return left.length === right.length && left.every((v, i) => v === right[i]); |
| 305 | }; |
| 306 | |
| 307 | /** Build the master/core instance-group configuration for RunJobFlow. */ |
| 308 | const buildInstanceGroups = ( |
| 309 | instances: ClusterInstancesConfig | undefined, |
| 310 | ): emr.InstanceGroupConfig[] => { |
| 311 | const coreCount = instances?.coreInstanceCount ?? 1; |
| 312 | return [ |
| 313 | { |
| 314 | Name: "Master", |
| 315 | InstanceRole: "MASTER", |
| 316 | InstanceType: instances?.masterInstanceType ?? "m5.xlarge", |
| 317 | InstanceCount: 1, |
| 318 | }, |
| 319 | ...(coreCount > 0 |
| 320 | ? [ |
| 321 | { |
| 322 | Name: "Core", |
| 323 | InstanceRole: "CORE" as const, |
| 324 | InstanceType: instances?.coreInstanceType ?? "m5.xlarge", |
| 325 | InstanceCount: coreCount, |
| 326 | }, |
| 327 | ] |
| 328 | : []), |
| 329 | ]; |
| 330 | }; |
| 331 | |
| 332 | export const ClusterProvider = () => |
| 333 | Provider.effect( |
| 334 | Cluster, |
| 335 | Effect.gen(function* () { |
| 336 | const toName = (id: string, props: Partial<ClusterProps>) => |
| 337 | props.clusterName |
| 338 | ? Effect.succeed(props.clusterName) |
| 339 | : createPhysicalName({ id, maxLength: 80 }); |
| 340 | |
| 341 | // Describe a cluster by id; a terminal or missing cluster reads as |
| 342 | // undefined — a terminated EMR cluster cannot be revived or updated. |
| 343 | const readCluster = Effect.fn(function* (clusterId: string) { |
| 344 | const response = yield* emr |
| 345 | .describeCluster({ ClusterId: clusterId }) |
| 346 | .pipe(Effect.catchTag("ClusterNotFound", () => Effect.undefined)); |
| 347 | const cluster = response?.Cluster; |
| 348 | if (!cluster?.Id) return undefined; |
| 349 | if (TERMINAL_STATES.has(cluster.Status?.State ?? "")) return undefined; |
| 350 | return cluster; |
| 351 | }); |
| 352 | |
| 353 | // Cluster names are not unique in EMR; the deterministic physical name |
| 354 | // is. Find the live (non-terminal) cluster carrying our name — used |
| 355 | // when state was lost and we only know the derived name. |
| 356 | const findClusterByName = Effect.fn(function* (name: string) { |
| 357 | const summary = yield* emr.listClusters.items({ ClusterStates: [...ACTIVE_STATES] }).pipe( |
| 358 | Stream.filter((summary) => summary.Name === name), |
| 359 | Stream.runHead, |
| 360 | Effect.map(Option.getOrUndefined), |
| 361 | ); |
| 362 | return summary?.Id ? yield* readCluster(summary.Id) : undefined; |
| 363 | }); |
| 364 | |
| 365 | // Bounded readiness wait. A Spark/Hadoop cluster typically reaches |
| 366 | // WAITING in 10-15 minutes; budget ~30 min (90 * 20s). A terminal |
| 367 | // state stops the wait immediately (non-retryable). |
| 368 | const waitForReady = Effect.fn(function* (clusterId: string) { |
| 369 | const policy = Schedule.max([Schedule.fixed("20 seconds"), Schedule.recurs(90)]); |
| 370 | return yield* emr.describeCluster({ ClusterId: clusterId }).pipe( |
| 371 | Effect.flatMap( |
| 372 | Effect.fnUntraced(function* (response) { |
| 373 | const cluster = response.Cluster; |
| 374 | const state = cluster?.Status?.State ?? "STARTING"; |
| 375 | if (TERMINAL_STATES.has(state)) { |
| 376 | const reason = cluster?.Status?.StateChangeReason; |
| 377 | return yield* new EMRClusterOperationFailed({ |
| 378 | message: |
| 379 | `EMR cluster '${clusterId}' reached terminal state '${state}'` + |
| 380 | ` (${reason?.Code ?? "unknown"}: ${reason?.Message ?? "no message"})`, |
| 381 | }); |
| 382 | } |
| 383 | if (state !== "WAITING" && state !== "RUNNING") { |
| 384 | return yield* new EMRClusterNotReady({ |
| 385 | message: `EMR cluster '${clusterId}' not ready (state: ${state})`, |
| 386 | }); |
| 387 | } |
| 388 | return cluster!; |
| 389 | }), |
| 390 | ), |
| 391 | Effect.retry({ |
| 392 | while: (e) => e instanceof Error && !e.message.includes("terminal state"), |
| 393 | schedule: policy, |
| 394 | }), |
| 395 | ); |
| 396 | }); |
| 397 | |
| 398 | const toAttrs = Effect.fn(function* (cluster: emr.Cluster) { |
| 399 | if (!cluster.Id || !cluster.ClusterArn || !cluster.Name) { |
| 400 | return yield* new EMRClusterUnexpectedResponse({ |
| 401 | message: `EMR cluster '${cluster.Id}' is missing its ARN or name`, |
| 402 | }); |
| 403 | } |
| 404 | return { |
| 405 | clusterId: cluster.Id, |
| 406 | clusterArn: cluster.ClusterArn, |
| 407 | clusterName: cluster.Name, |
| 408 | state: cluster.Status?.State ?? "STARTING", |
| 409 | masterPublicDnsName: cluster.MasterPublicDnsName, |
| 410 | tags: toTagRecord(cluster.Tags), |
| 411 | }; |
| 412 | }); |
| 413 | |
| 414 | return { |
| 415 | stables: ["clusterId", "clusterArn", "clusterName"], |
| 416 | |
| 417 | diff: Effect.fn(function* ({ id, olds, news }) { |
| 418 | if (!isResolved(news)) return undefined; |
| 419 | const n = news; |
| 420 | const o = olds; |
| 421 | if (n === undefined || o === undefined) return undefined; |
| 422 | if ((yield* toName(id, o)) !== (yield* toName(id, n))) { |
| 423 | return { action: "replace" } as const; |
| 424 | } |
| 425 | // Create-only cluster-level properties force a replacement. |
| 426 | if (n.releaseLabel !== o.releaseLabel) { |
| 427 | return { action: "replace" } as const; |
| 428 | } |
| 429 | if (!sameStringSet(n.applications, o.applications)) { |
| 430 | return { action: "replace" } as const; |
| 431 | } |
| 432 | if (n.serviceRole !== o.serviceRole) { |
| 433 | return { action: "replace" } as const; |
| 434 | } |
| 435 | if (n.jobFlowRole !== o.jobFlowRole) { |
| 436 | return { action: "replace" } as const; |
| 437 | } |
| 438 | if ((n.logUri ?? undefined) !== (o.logUri ?? undefined)) { |
| 439 | return { action: "replace" } as const; |
| 440 | } |
| 441 | if ((n.securityConfiguration ?? undefined) !== (o.securityConfiguration ?? undefined)) { |
| 442 | return { action: "replace" } as const; |
| 443 | } |
| 444 | if (JSON.stringify(n.configurations ?? []) !== JSON.stringify(o.configurations ?? [])) { |
| 445 | return { action: "replace" } as const; |
| 446 | } |
| 447 | if ((n.ebsRootVolumeSize ?? undefined) !== (o.ebsRootVolumeSize ?? undefined)) { |
| 448 | return { action: "replace" } as const; |
| 449 | } |
| 450 | if ((n.customAmiId ?? undefined) !== (o.customAmiId ?? undefined)) { |
| 451 | return { action: "replace" } as const; |
| 452 | } |
| 453 | // Create-only instance topology. |
| 454 | const ni = n.instances ?? {}; |
| 455 | const oi = o.instances ?? {}; |
| 456 | if ( |
| 457 | (ni.masterInstanceType ?? "m5.xlarge") !== (oi.masterInstanceType ?? "m5.xlarge") || |
| 458 | (ni.coreInstanceType ?? "m5.xlarge") !== (oi.coreInstanceType ?? "m5.xlarge") || |
| 459 | (ni.ec2SubnetId ?? undefined) !== (oi.ec2SubnetId ?? undefined) || |
| 460 | (ni.ec2KeyName ?? undefined) !== (oi.ec2KeyName ?? undefined) || |
| 461 | (ni.emrManagedMasterSecurityGroup ?? undefined) !== |
| 462 | (oi.emrManagedMasterSecurityGroup ?? undefined) || |
| 463 | (ni.emrManagedSlaveSecurityGroup ?? undefined) !== |
| 464 | (oi.emrManagedSlaveSecurityGroup ?? undefined) || |
| 465 | !sameStringSet(ni.additionalMasterSecurityGroups, oi.additionalMasterSecurityGroups) || |
| 466 | !sameStringSet(ni.additionalSlaveSecurityGroups, oi.additionalSlaveSecurityGroups) |
| 467 | ) { |
| 468 | return { action: "replace" } as const; |
| 469 | } |
| 470 | // The core group can resize in place, but crossing the 0 boundary |
| 471 | // adds/removes the whole group — replacement. |
| 472 | const newCore = ni.coreInstanceCount ?? 1; |
| 473 | const oldCore = oi.coreInstanceCount ?? 1; |
| 474 | if ((newCore === 0) !== (oldCore === 0)) { |
| 475 | return { action: "replace" } as const; |
| 476 | } |
| 477 | }), |
| 478 | |
| 479 | read: Effect.fn(function* ({ id, olds, output }) { |
| 480 | const cluster = output?.clusterId |
| 481 | ? yield* readCluster(output.clusterId) |
| 482 | : yield* findClusterByName(yield* toName(id, olds ?? {})); |
| 483 | if (!cluster) return undefined; |
| 484 | const attrs = yield* toAttrs(cluster); |
| 485 | return (yield* hasAlchemyTags(id, attrs.tags)) ? attrs : Unowned(attrs); |
| 486 | }), |
| 487 | |
| 488 | reconcile: Effect.fn(function* ({ id, news, output, session }) { |
| 489 | const props = news!; |
| 490 | const name = output?.clusterName ?? (yield* toName(id, props)); |
| 491 | const internalTags = yield* createInternalTags(id); |
| 492 | const desiredTags = { ...internalTags, ...props.tags }; |
| 493 | |
| 494 | // 1. Observe — cloud state is authoritative; output is only an id |
| 495 | // cache. A terminated cluster reads as missing. |
| 496 | let observed = output?.clusterId |
| 497 | ? yield* readCluster(output.clusterId) |
| 498 | : yield* findClusterByName(name); |
| 499 | |
| 500 | // 2. Ensure — launch the job flow if missing. Create errors |
| 501 | // (invalid roles, subnet, release) propagate directly. |
| 502 | let clusterId = observed?.Id; |
| 503 | if (clusterId === undefined) { |
| 504 | const instances = props.instances ?? {}; |
| 505 | const created = yield* emr |
| 506 | .runJobFlow({ |
| 507 | Name: name, |
| 508 | ReleaseLabel: props.releaseLabel, |
| 509 | Applications: (props.applications ?? []).map((Name) => ({ |
| 510 | Name, |
| 511 | })), |
| 512 | Instances: { |
| 513 | InstanceGroups: buildInstanceGroups(props.instances), |
| 514 | Ec2SubnetId: instances.ec2SubnetId, |
| 515 | Ec2KeyName: instances.ec2KeyName, |
| 516 | KeepJobFlowAliveWhenNoSteps: instances.keepJobFlowAliveWhenNoSteps ?? true, |
| 517 | TerminationProtected: instances.terminationProtected ?? false, |
| 518 | UnhealthyNodeReplacement: instances.unhealthyNodeReplacement, |
| 519 | EmrManagedMasterSecurityGroup: instances.emrManagedMasterSecurityGroup, |
| 520 | EmrManagedSlaveSecurityGroup: instances.emrManagedSlaveSecurityGroup, |
| 521 | AdditionalMasterSecurityGroups: instances.additionalMasterSecurityGroups, |
| 522 | AdditionalSlaveSecurityGroups: instances.additionalSlaveSecurityGroups, |
| 523 | }, |
| 524 | ServiceRole: props.serviceRole, |
| 525 | JobFlowRole: props.jobFlowRole, |
| 526 | LogUri: props.logUri, |
| 527 | SecurityConfiguration: props.securityConfiguration, |
| 528 | Configurations: props.configurations, |
| 529 | VisibleToAllUsers: props.visibleToAllUsers ?? true, |
| 530 | EbsRootVolumeSize: props.ebsRootVolumeSize, |
| 531 | CustomAmiId: props.customAmiId, |
| 532 | StepConcurrencyLevel: props.stepConcurrencyLevel, |
| 533 | AutoTerminationPolicy: props.autoTerminationPolicy |
| 534 | ? { |
| 535 | IdleTimeout: toWireSeconds(props.autoTerminationPolicy.idleTimeout), |
| 536 | } |
| 537 | : undefined, |
| 538 | Tags: Object.entries(desiredTags).map(([Key, Value]) => ({ |
| 539 | Key, |
| 540 | Value, |
| 541 | })), |
| 542 | }) |
| 543 | .pipe( |
| 544 | // A freshly created EC2 instance profile is not yet visible to |
| 545 | // EMR (IAM eventual consistency) — RunJobFlow rejects it with |
| 546 | // "Invalid InstanceProfile" for several seconds. Retry bounded. |
| 547 | Effect.retry({ |
| 548 | while: (e) => e._tag === "EmrInstanceProfileNotReady", |
| 549 | schedule: Schedule.spaced("5 seconds"), |
| 550 | times: 10, |
| 551 | }), |
| 552 | ); |
| 553 | if (!created.JobFlowId) { |
| 554 | return yield* new EMRClusterUnexpectedResponse({ |
| 555 | message: `RunJobFlow for '${name}' returned no JobFlowId`, |
| 556 | }); |
| 557 | } |
| 558 | clusterId = created.JobFlowId; |
| 559 | } |
| 560 | |
| 561 | // Provisioning surfaces as STARTING/BOOTSTRAPPING; wait (bounded) |
| 562 | // for WAITING/RUNNING so modify calls don't race bootstrap. |
| 563 | observed = yield* waitForReady(clusterId); |
| 564 | |
| 565 | // 3. Sync — per mutable aspect, diff OBSERVED state against the |
| 566 | // desired props and apply only the delta. |
| 567 | const desiredVisible = props.visibleToAllUsers ?? true; |
| 568 | if ((observed.VisibleToAllUsers ?? true) !== desiredVisible) { |
| 569 | yield* emr.setVisibleToAllUsers({ |
| 570 | JobFlowIds: [clusterId], |
| 571 | VisibleToAllUsers: desiredVisible, |
| 572 | }); |
| 573 | } |
| 574 | |
| 575 | const desiredKeepAlive = props.instances?.keepJobFlowAliveWhenNoSteps ?? true; |
| 576 | const observedKeepAlive = !(observed.AutoTerminate ?? false); |
| 577 | if (observedKeepAlive !== desiredKeepAlive) { |
| 578 | yield* emr.setKeepJobFlowAliveWhenNoSteps({ |
| 579 | JobFlowIds: [clusterId], |
| 580 | KeepJobFlowAliveWhenNoSteps: desiredKeepAlive, |
| 581 | }); |
| 582 | } |
| 583 | |
| 584 | const desiredProtected = props.instances?.terminationProtected; |
| 585 | if ( |
| 586 | desiredProtected !== undefined && |
| 587 | (observed.TerminationProtected ?? false) !== desiredProtected |
| 588 | ) { |
| 589 | yield* emr.setTerminationProtection({ |
| 590 | JobFlowIds: [clusterId], |
| 591 | TerminationProtected: desiredProtected, |
| 592 | }); |
| 593 | } |
| 594 | |
| 595 | const desiredUnhealthy = props.instances?.unhealthyNodeReplacement; |
| 596 | if ( |
| 597 | desiredUnhealthy !== undefined && |
| 598 | (observed.UnhealthyNodeReplacement ?? false) !== desiredUnhealthy |
| 599 | ) { |
| 600 | yield* emr.setUnhealthyNodeReplacement({ |
| 601 | JobFlowIds: [clusterId], |
| 602 | UnhealthyNodeReplacement: desiredUnhealthy, |
| 603 | }); |
| 604 | } |
| 605 | |
| 606 | if ( |
| 607 | props.stepConcurrencyLevel !== undefined && |
| 608 | (observed.StepConcurrencyLevel ?? 1) !== props.stepConcurrencyLevel |
| 609 | ) { |
| 610 | yield* emr.modifyCluster({ |
| 611 | ClusterId: clusterId, |
| 612 | StepConcurrencyLevel: props.stepConcurrencyLevel, |
| 613 | }); |
| 614 | } |
| 615 | |
| 616 | // Auto-termination policy: observe, then put/remove the delta. |
| 617 | const observedPolicy = yield* emr.getAutoTerminationPolicy({ ClusterId: clusterId }).pipe( |
| 618 | Effect.map((r) => r.AutoTerminationPolicy), |
| 619 | Effect.orElseSucceed(() => undefined), |
| 620 | ); |
| 621 | if (props.autoTerminationPolicy !== undefined) { |
| 622 | const desiredIdleTimeout = toWireSeconds(props.autoTerminationPolicy.idleTimeout); |
| 623 | if (observedPolicy?.IdleTimeout !== desiredIdleTimeout) { |
| 624 | yield* emr.putAutoTerminationPolicy({ |
| 625 | ClusterId: clusterId, |
| 626 | AutoTerminationPolicy: { |
| 627 | IdleTimeout: desiredIdleTimeout, |
| 628 | }, |
| 629 | }); |
| 630 | } |
| 631 | } else if (observedPolicy !== undefined) { |
| 632 | yield* emr.removeAutoTerminationPolicy({ ClusterId: clusterId }); |
| 633 | } |
| 634 | |
| 635 | // Core instance-group resize (in place between non-zero counts). |
| 636 | const desiredCore = props.instances?.coreInstanceCount ?? 1; |
| 637 | if (desiredCore > 0) { |
| 638 | const core = yield* emr.listInstanceGroups.items({ ClusterId: clusterId }).pipe( |
| 639 | Stream.filter((g) => g.InstanceGroupType === "CORE"), |
| 640 | Stream.runHead, |
| 641 | Effect.map(Option.getOrUndefined), |
| 642 | ); |
| 643 | if (core?.Id !== undefined && (core.RequestedInstanceCount ?? 0) !== desiredCore) { |
| 644 | yield* emr.modifyInstanceGroups({ |
| 645 | ClusterId: clusterId, |
| 646 | InstanceGroups: [{ InstanceGroupId: core.Id, InstanceCount: desiredCore }], |
| 647 | }); |
| 648 | } |
| 649 | } |
| 650 | |
| 651 | // 3b. Sync tags — diff against OBSERVED cloud tags. |
| 652 | const observedTags = toTagRecord(observed.Tags); |
| 653 | const { removed, upsert } = diffTags(observedTags, desiredTags); |
| 654 | if (upsert.length > 0) { |
| 655 | yield* emr.addTags({ ResourceId: clusterId, Tags: upsert }); |
| 656 | } |
| 657 | if (removed.length > 0) { |
| 658 | yield* emr.removeTags({ ResourceId: clusterId, TagKeys: removed }); |
| 659 | } |
| 660 | |
| 661 | // 4. Return fresh attributes. |
| 662 | const final = yield* readCluster(clusterId); |
| 663 | yield* session.note(clusterId); |
| 664 | return yield* toAttrs(final ?? observed); |
| 665 | }), |
| 666 | |
| 667 | delete: Effect.fn(function* ({ output }) { |
| 668 | const clusterId = output.clusterId; |
| 669 | // A cluster that is already terminal (or gone) is success. |
| 670 | const observed = yield* readCluster(clusterId); |
| 671 | if (observed === undefined) return; |
| 672 | // Lift termination protection so TerminateJobFlows succeeds. |
| 673 | if (observed.TerminationProtected === true) { |
| 674 | yield* emr |
| 675 | .setTerminationProtection({ |
| 676 | JobFlowIds: [clusterId], |
| 677 | TerminationProtected: false, |
| 678 | }) |
| 679 | .pipe(Effect.catchTag("JobFlowNotFound", () => Effect.void)); |
| 680 | } |
| 681 | yield* emr |
| 682 | .terminateJobFlows({ JobFlowIds: [clusterId] }) |
| 683 | .pipe(Effect.catchTag("JobFlowNotFound", () => Effect.void)); |
| 684 | // Termination is irreversible once initiated; wait (bounded) for |
| 685 | // the state to leave the active set. Full instance teardown takes |
| 686 | // 5-10 more minutes server-side and needs no babysitting. |
| 687 | yield* emr.describeCluster({ ClusterId: clusterId }).pipe( |
| 688 | Effect.flatMap((response) => { |
| 689 | const state = response.Cluster?.Status?.State ?? "TERMINATED"; |
| 690 | return TERMINAL_STATES.has(state) |
| 691 | ? Effect.void |
| 692 | : Effect.fail( |
| 693 | new EMRClusterNotReady({ |
| 694 | message: `EMR cluster '${clusterId}' still active (state: ${state})`, |
| 695 | }), |
| 696 | ); |
| 697 | }), |
| 698 | Effect.catchTag("ClusterNotFound", () => Effect.void), |
| 699 | Effect.retry({ |
| 700 | schedule: Schedule.max([Schedule.fixed("10 seconds"), Schedule.recurs(30)]), |
| 701 | }), |
| 702 | ); |
| 703 | }), |
| 704 | |
| 705 | list: () => |
| 706 | emr.listClusters.items({ ClusterStates: [...ACTIVE_STATES] }).pipe( |
| 707 | Stream.runCollect, |
| 708 | Effect.map((chunk) => |
| 709 | Array.from(chunk).flatMap((summary) => (summary.Id ? [summary.Id] : [])), |
| 710 | ), |
| 711 | Effect.flatMap( |
| 712 | Effect.forEach((clusterId) => readCluster(clusterId), { |
| 713 | concurrency: 4, |
| 714 | }), |
| 715 | ), |
| 716 | Effect.map((clusters) => |
| 717 | clusters.flatMap((cluster) => (cluster === undefined ? [] : [cluster])), |
| 718 | ), |
| 719 | Effect.flatMap( |
| 720 | Effect.forEach((cluster) => toAttrs(cluster), { |
| 721 | concurrency: 4, |
| 722 | }), |
| 723 | ), |
| 724 | ), |
| 725 | }; |
| 726 | }), |
| 727 | ); |