| 1 | import * as amp from "@distilled.cloud/aws/amp"; |
| 2 | import type * as Duration from "effect/Duration"; |
| 3 | import * as Effect from "effect/Effect"; |
| 4 | import * as Option from "effect/Option"; |
| 5 | import * as Schedule from "effect/Schedule"; |
| 6 | import * as Stream from "effect/Stream"; |
| 7 | import { Unowned } from "../../AdoptPolicy.ts"; |
| 8 | import { isResolved } from "../../Diff.ts"; |
| 9 | import * as Provider from "../../Provider.ts"; |
| 10 | import { Resource } from "../../Resource.ts"; |
| 11 | import { createInternalTags, hasAlchemyTags } from "../../Tags.ts"; |
| 12 | import { toWireSeconds } from "../../Util/Duration.ts"; |
| 13 | import type { Providers } from "../Providers.ts"; |
| 14 | import { syncAmpTags, toTagRecord } from "./internal.ts"; |
| 15 | |
| 16 | /** |
| 17 | * Tolerance band around the expected value inside which deviations are not |
| 18 | * flagged — either an absolute `amount` or a `ratio` of the expected value. |
| 19 | */ |
| 20 | export type AnomalyDetectorIgnoreNearExpected = |
| 21 | | { amount: number } |
| 22 | | { ratio: number }; |
| 23 | |
| 24 | /** What the detector does with evaluation windows that have no data. */ |
| 25 | export type AnomalyDetectorMissingDataAction = |
| 26 | | { markAsAnomaly: boolean } |
| 27 | | { skip: boolean }; |
| 28 | |
| 29 | export interface AnomalyDetectorProps { |
| 30 | /** |
| 31 | * Id of the AMP workspace the detector runs in. Changing the workspace |
| 32 | * replaces the detector. |
| 33 | */ |
| 34 | workspaceId: string; |
| 35 | /** |
| 36 | * Human-readable alias for the detector, unique within the workspace. |
| 37 | * Changing the alias replaces the detector. |
| 38 | */ |
| 39 | alias: string; |
| 40 | /** |
| 41 | * The PromQL query producing the time series the Random Cut Forest |
| 42 | * algorithm scores for anomalies. Updating it retrains the detector. |
| 43 | */ |
| 44 | query: string; |
| 45 | /** |
| 46 | * Number of consecutive data points combined into one sample (shingle) |
| 47 | * for scoring. If omitted, the service default is used. |
| 48 | */ |
| 49 | shingleSize?: number; |
| 50 | /** |
| 51 | * Number of points sampled per Random Cut Forest tree. If omitted, the |
| 52 | * service default is used. |
| 53 | */ |
| 54 | sampleSize?: number; |
| 55 | /** |
| 56 | * Suppress anomalies above the expected value that fall within this |
| 57 | * tolerance band. |
| 58 | */ |
| 59 | ignoreNearExpectedFromAbove?: AnomalyDetectorIgnoreNearExpected; |
| 60 | /** |
| 61 | * Suppress anomalies below the expected value that fall within this |
| 62 | * tolerance band. |
| 63 | */ |
| 64 | ignoreNearExpectedFromBelow?: AnomalyDetectorIgnoreNearExpected; |
| 65 | /** |
| 66 | * How often the detector evaluates the query. Accepts any |
| 67 | * `Duration.Input` (e.g. `"1 minute"`, `Duration.minutes(1)`; a bare |
| 68 | * number is milliseconds); the wire unit is whole seconds |
| 69 | * (`evaluationIntervalInSeconds`). If omitted, the service default is |
| 70 | * used. |
| 71 | */ |
| 72 | evaluationInterval?: Duration.Input; |
| 73 | /** |
| 74 | * What to do when an evaluation window has no data: mark it as an |
| 75 | * anomaly (`{ markAsAnomaly: true }`) or skip it (`{ skip: true }`). |
| 76 | */ |
| 77 | missingDataAction?: AnomalyDetectorMissingDataAction; |
| 78 | /** |
| 79 | * Extra labels attached to the anomaly metrics the detector emits. |
| 80 | */ |
| 81 | labels?: Record<string, string>; |
| 82 | /** |
| 83 | * User-defined tags for the detector. |
| 84 | */ |
| 85 | tags?: Record<string, string>; |
| 86 | } |
| 87 | |
| 88 | export interface AnomalyDetector extends Resource< |
| 89 | "AWS.AMP.AnomalyDetector", |
| 90 | AnomalyDetectorProps, |
| 91 | { |
| 92 | workspaceId: string; |
| 93 | anomalyDetectorId: string; |
| 94 | anomalyDetectorArn: string; |
| 95 | alias: string; |
| 96 | status: string; |
| 97 | }, |
| 98 | never, |
| 99 | Providers |
| 100 | > {} |
| 101 | |
| 102 | /** |
| 103 | * A Random Cut Forest anomaly detector inside an Amazon Managed Service for |
| 104 | * Prometheus workspace — continuously evaluates a PromQL query and emits |
| 105 | * anomaly scores as new metrics in the same workspace. |
| 106 | * |
| 107 | * ### Creating an Anomaly Detector |
| 108 | * **Example:** Detect Anomalies on a Request-Rate Query |
| 109 | * ```typescript |
| 110 | * const workspace = yield* AMP.Workspace("Metrics", {}); |
| 111 | * const detector = yield* AMP.AnomalyDetector("RequestSpikes", { |
| 112 | * workspaceId: workspace.workspaceId, |
| 113 | * alias: "request-spikes", |
| 114 | * query: 'rate(http_requests_total{job="api"}[5m])', |
| 115 | * evaluationInterval: "1 minute", |
| 116 | * missingDataAction: { skip: true }, |
| 117 | * }); |
| 118 | * ``` |
| 119 | * |
| 120 | * @resource |
| 121 | */ |
| 122 | export const AnomalyDetector = Resource<AnomalyDetector>( |
| 123 | "AWS.AMP.AnomalyDetector", |
| 124 | ); |
| 125 | |
| 126 | export const AnomalyDetectorProvider = () => |
| 127 | Provider.effect( |
| 128 | AnomalyDetector, |
| 129 | Effect.gen(function* () { |
| 130 | const toAttrs = ( |
| 131 | workspaceId: string, |
| 132 | detector: amp.AnomalyDetectorDescription, |
| 133 | ) => ({ |
| 134 | workspaceId, |
| 135 | anomalyDetectorId: detector.anomalyDetectorId, |
| 136 | anomalyDetectorArn: detector.arn, |
| 137 | alias: detector.alias, |
| 138 | status: detector.status.statusCode, |
| 139 | }); |
| 140 | |
| 141 | /** Describe a detector by id; typed not-found → undefined. */ |
| 142 | const describe = Effect.fn(function* ( |
| 143 | workspaceId: string, |
| 144 | anomalyDetectorId: string, |
| 145 | ) { |
| 146 | const response = yield* amp |
| 147 | .describeAnomalyDetector({ workspaceId, anomalyDetectorId }) |
| 148 | .pipe( |
| 149 | Effect.catchTag("ResourceNotFoundException", () => |
| 150 | Effect.succeed(undefined), |
| 151 | ), |
| 152 | ); |
| 153 | return response?.anomalyDetector; |
| 154 | }); |
| 155 | |
| 156 | /** |
| 157 | * Find a detector by its alias (the deterministic identity within a |
| 158 | * workspace) — used when no id is cached in `output`. |
| 159 | */ |
| 160 | const findByAlias = Effect.fn(function* ( |
| 161 | workspaceId: string, |
| 162 | alias: string, |
| 163 | ) { |
| 164 | const summary = yield* amp.listAnomalyDetectors |
| 165 | .items({ workspaceId, alias }) |
| 166 | .pipe( |
| 167 | Stream.filter((s) => s.alias === alias), |
| 168 | Stream.runHead, |
| 169 | Effect.map(Option.getOrUndefined), |
| 170 | Effect.catchTag("ResourceNotFoundException", () => |
| 171 | Effect.succeed(undefined), |
| 172 | ), |
| 173 | ); |
| 174 | return summary === undefined |
| 175 | ? undefined |
| 176 | : yield* describe(workspaceId, summary.anomalyDetectorId); |
| 177 | }); |
| 178 | |
| 179 | /** |
| 180 | * Bounded best-effort wait toward ACTIVE. Training/backfill can |
| 181 | * outlast the window; the last observed description is returned and |
| 182 | * a still-training detector converges on a later reconcile. |
| 183 | */ |
| 184 | const waitSettled = Effect.fn(function* ( |
| 185 | workspaceId: string, |
| 186 | anomalyDetectorId: string, |
| 187 | ) { |
| 188 | return yield* amp |
| 189 | .describeAnomalyDetector({ workspaceId, anomalyDetectorId }) |
| 190 | .pipe( |
| 191 | Effect.map((r) => r.anomalyDetector), |
| 192 | Effect.repeat({ |
| 193 | schedule: Schedule.max([ |
| 194 | Schedule.fixed("3 seconds"), |
| 195 | Schedule.recurs(20), |
| 196 | ]), |
| 197 | until: (d): boolean => |
| 198 | d.status.statusCode !== "CREATING" && |
| 199 | d.status.statusCode !== "UPDATING", |
| 200 | }), |
| 201 | ); |
| 202 | }); |
| 203 | |
| 204 | const canonical = (value: unknown) => |
| 205 | JSON.stringify(value, (_, v: unknown) => |
| 206 | v !== null && typeof v === "object" && !Array.isArray(v) |
| 207 | ? Object.fromEntries( |
| 208 | Object.entries(v as Record<string, unknown>).sort(([a], [b]) => |
| 209 | a.localeCompare(b), |
| 210 | ), |
| 211 | ) |
| 212 | : v, |
| 213 | ); |
| 214 | |
| 215 | /** |
| 216 | * Detect drift between desired props and the observed detector. |
| 217 | * Only user-specified fields participate — omitted optional fields |
| 218 | * defer to service defaults and never force a (full-retrain) put. |
| 219 | */ |
| 220 | const configDrifts = ( |
| 221 | news: AnomalyDetectorProps, |
| 222 | observed: amp.AnomalyDetectorDescription, |
| 223 | ): boolean => { |
| 224 | const rcf = observed.configuration?.randomCutForest; |
| 225 | const drift = |
| 226 | news.query !== rcf?.query || |
| 227 | (news.shingleSize !== undefined && |
| 228 | news.shingleSize !== rcf?.shingleSize) || |
| 229 | (news.sampleSize !== undefined && |
| 230 | news.sampleSize !== rcf?.sampleSize) || |
| 231 | (news.ignoreNearExpectedFromAbove !== undefined && |
| 232 | canonical(news.ignoreNearExpectedFromAbove) !== |
| 233 | canonical(rcf?.ignoreNearExpectedFromAbove)) || |
| 234 | (news.ignoreNearExpectedFromBelow !== undefined && |
| 235 | canonical(news.ignoreNearExpectedFromBelow) !== |
| 236 | canonical(rcf?.ignoreNearExpectedFromBelow)) || |
| 237 | (news.evaluationInterval !== undefined && |
| 238 | toWireSeconds(news.evaluationInterval) !== |
| 239 | observed.evaluationIntervalInSeconds) || |
| 240 | (news.missingDataAction !== undefined && |
| 241 | canonical(news.missingDataAction) !== |
| 242 | canonical(observed.missingDataAction)) || |
| 243 | (news.labels !== undefined && |
| 244 | canonical(news.labels) !== canonical(toTagRecord(observed.labels))); |
| 245 | return drift; |
| 246 | }; |
| 247 | |
| 248 | const toConfiguration = (news: AnomalyDetectorProps) => ({ |
| 249 | randomCutForest: { |
| 250 | query: news.query, |
| 251 | shingleSize: news.shingleSize, |
| 252 | sampleSize: news.sampleSize, |
| 253 | ignoreNearExpectedFromAbove: news.ignoreNearExpectedFromAbove, |
| 254 | ignoreNearExpectedFromBelow: news.ignoreNearExpectedFromBelow, |
| 255 | }, |
| 256 | }); |
| 257 | |
| 258 | return { |
| 259 | stables: [ |
| 260 | "workspaceId", |
| 261 | "anomalyDetectorId", |
| 262 | "anomalyDetectorArn", |
| 263 | "alias", |
| 264 | ], |
| 265 | |
| 266 | diff: Effect.fn(function* ({ olds, news }) { |
| 267 | if (!isResolved(news)) return undefined; |
| 268 | if ( |
| 269 | olds?.workspaceId !== news.workspaceId || |
| 270 | olds?.alias !== news.alias |
| 271 | ) { |
| 272 | return { action: "replace" } as const; |
| 273 | } |
| 274 | }), |
| 275 | |
| 276 | read: Effect.fn(function* ({ id, olds, output }) { |
| 277 | const workspaceId = output?.workspaceId ?? olds?.workspaceId; |
| 278 | if (!workspaceId) return undefined; |
| 279 | const detector = |
| 280 | output?.anomalyDetectorId !== undefined |
| 281 | ? yield* describe(workspaceId, output.anomalyDetectorId) |
| 282 | : olds?.alias !== undefined |
| 283 | ? yield* findByAlias(workspaceId, olds.alias) |
| 284 | : undefined; |
| 285 | if (detector === undefined) return undefined; |
| 286 | const attrs = toAttrs(workspaceId, detector); |
| 287 | const tags = toTagRecord(detector.tags); |
| 288 | return (yield* hasAlchemyTags(id, tags)) ? attrs : Unowned(attrs); |
| 289 | }), |
| 290 | |
| 291 | reconcile: Effect.fn(function* ({ id, news, output, session }) { |
| 292 | const workspaceId = news!.workspaceId; |
| 293 | const internalTags = yield* createInternalTags(id); |
| 294 | const desiredTags = { ...internalTags, ...news!.tags }; |
| 295 | |
| 296 | // 1. Observe — by cached id, falling back to the alias. |
| 297 | let detector = |
| 298 | output?.anomalyDetectorId !== undefined |
| 299 | ? yield* describe(workspaceId, output.anomalyDetectorId) |
| 300 | : yield* findByAlias(workspaceId, news!.alias); |
| 301 | |
| 302 | // 2. Ensure — create if missing. |
| 303 | if (detector === undefined) { |
| 304 | const created = yield* amp.createAnomalyDetector({ |
| 305 | workspaceId, |
| 306 | alias: news!.alias, |
| 307 | configuration: toConfiguration(news!), |
| 308 | evaluationIntervalInSeconds: toWireSeconds( |
| 309 | news!.evaluationInterval, |
| 310 | ), |
| 311 | missingDataAction: news!.missingDataAction, |
| 312 | labels: news!.labels, |
| 313 | tags: desiredTags, |
| 314 | }); |
| 315 | detector = yield* waitSettled( |
| 316 | workspaceId, |
| 317 | created.anomalyDetectorId, |
| 318 | ); |
| 319 | } else if (configDrifts(news!, detector)) { |
| 320 | // 3. Sync — `putAnomalyDetector` triggers a full retrain, so it |
| 321 | // is only called when an observed aspect actually drifts. |
| 322 | yield* amp.putAnomalyDetector({ |
| 323 | workspaceId, |
| 324 | anomalyDetectorId: detector.anomalyDetectorId, |
| 325 | configuration: toConfiguration(news!), |
| 326 | evaluationIntervalInSeconds: toWireSeconds( |
| 327 | news!.evaluationInterval, |
| 328 | ), |
| 329 | missingDataAction: news!.missingDataAction, |
| 330 | labels: news!.labels, |
| 331 | }); |
| 332 | detector = yield* waitSettled( |
| 333 | workspaceId, |
| 334 | detector.anomalyDetectorId, |
| 335 | ); |
| 336 | } |
| 337 | |
| 338 | // 3b. Sync tags — diff against OBSERVED cloud tags. |
| 339 | yield* syncAmpTags(detector.arn, desiredTags); |
| 340 | |
| 341 | yield* session.note(`${workspaceId}/${detector.anomalyDetectorId}`); |
| 342 | return toAttrs(workspaceId, detector); |
| 343 | }), |
| 344 | |
| 345 | delete: Effect.fn(function* ({ output }) { |
| 346 | yield* amp |
| 347 | .deleteAnomalyDetector({ |
| 348 | workspaceId: output.workspaceId, |
| 349 | anomalyDetectorId: output.anomalyDetectorId, |
| 350 | }) |
| 351 | .pipe( |
| 352 | Effect.catchTag("ResourceNotFoundException", () => Effect.void), |
| 353 | // A detector mid-training rejects deletion; retry briefly. |
| 354 | Effect.retry({ |
| 355 | while: (e) => e._tag === "ConflictException", |
| 356 | schedule: Schedule.max([ |
| 357 | Schedule.fixed("4 seconds"), |
| 358 | Schedule.recurs(25), |
| 359 | ]), |
| 360 | }), |
| 361 | ); |
| 362 | }), |
| 363 | |
| 364 | // Sub-resource keyed by its parent workspace — not independently |
| 365 | // enumerable across the account. |
| 366 | list: () => Effect.succeed([]), |
| 367 | }; |
| 368 | }), |
| 369 | ); |