sailfork / t3code-github / .repos/ alchemy-effect/ packages/ integration-tests/ test/ AWS/ AutoScaling/ LifecycleHookEventSource.test.ts

import * as AWS from "@alchemy.run/aws";
import { describe, expect } from "@alchemy.run/test";
import * as Test from "@alchemy.run/test/Alchemy";
import * as Core from "@alchemy.run/test/Core";
import * as autoscaling from "@distilled.cloud/aws/auto-scaling";
import * as Effect from "effect/Effect";
import * as HttpClient from "effect/http/HttpClient";
import * as HttpClientRequest from "effect/http/HttpClientRequest";
import * as Schedule from "effect/Schedule";
import LifecycleTestFunctionLive, {
  LifecycleTestFunction,
  lifecycleFleetAsgName,
  lifecycleHookName,
} from "./fixtures/lifecycle-handler.ts";

const testOptions = { providers: AWS.providers() };
const { test, beforeAll, afterAll } = Test.make(testOptions);
const sharedStack = Core.scratchStack(testOptions, "AutoScalingLifecycle");

const readinessPolicy = Schedule.max([Schedule.fixed("2 seconds"), Schedule.recurs(75)]);

let baseUrl: string;

// Deploys a Lambda that (a) creates a LAUNCHING lifecycle hook + EventBridge
// subscription via `consumeLifecycleActions` and (b) binds
// `CompleteLifecycleAction`. The fleet is sized to zero so no EC2 instances
// launch — this verifies the DX wiring (hook creation + binding IAM) without a
// >90s instance-launch round-trip. The full "instance pauses → Lambda completes
// CONTINUE → InService" flow is left to a live run with a non-zero desired
// capacity.
describe(
  "AutoScaling LifecycleHook event source + CompleteLifecycleAction",
  {
    tags: [
      "provider:aws",
      "provider:aws:autoscaling",
      "provider:aws:ec2",
      "provider:aws:lambda",
      "live",
    ],
  },
  () => {
    beforeAll(
      Effect.gen(function* () {
        yield* sharedStack.destroy();

        const { fn } = yield* sharedStack.deploy(
          Effect.gen(function* () {
            const fn = yield* LifecycleTestFunction;
            return { fn };
          }).pipe(Effect.provide(LifecycleTestFunctionLive)),
        );

        expect(fn.functionUrl).toBeTruthy();
        baseUrl = fn.functionUrl!.replace(/\/+$/, "");

        yield* HttpClient.get(`${baseUrl}/health`).pipe(
          Effect.flatMap((response) =>
            response.status === 200
              ? Effect.succeed(response)
              : Effect.fail(new Error(`Function not ready: ${response.status}`)),
          ),
          Effect.retry({ schedule: readinessPolicy }),
        );
      }),
      { timeout: 300_000 },
    );

    afterAll(
      Effect.gen(function* () {
        yield* sharedStack.destroy();
        // Assert the fixture ASG is fully gone — the suite must leave nothing.
        // The hook context has no AWS providers of its own, so provide them
        // explicitly for the out-of-band distilled call.
        const remaining = yield* Core.withProviders(
          autoscaling
            .describeAutoScalingGroups({ AutoScalingGroupNames: [lifecycleFleetAsgName] } as any)
            .pipe(
              Effect.map((r) => (r.AutoScalingGroups ?? []).length),
              Effect.repeat({
                until: (count) => count === 0,
                schedule: Schedule.spaced("3 seconds"),
                times: 10,
              }),
            ),
          testOptions,
          sharedStack.name,
        );
        expect(remaining).toBe(0);
      }),
      { timeout: 180_000 },
    );

    test.provider(
      "consumeLifecycleActions creates the launch lifecycle hook on the ASG",
      (_stack) =>
        Effect.gen(function* () {
          const hooks = yield* autoscaling.describeLifecycleHooks({
            AutoScalingGroupName: lifecycleFleetAsgName,
            LifecycleHookNames: [lifecycleHookName],
          } as any);
          const hook = hooks.LifecycleHooks?.[0];
          expect(hook?.LifecycleHookName).toEqual(lifecycleHookName);
          expect(hook?.LifecycleTransition).toEqual("autoscaling:EC2_INSTANCE_LAUNCHING");
          expect(hook?.HeartbeatTimeout).toEqual(300);
        }),
    );

    test.provider(
      "CompleteLifecycleAction binding is IAM-wired (no AccessDenied)",
      (_stack) =>
        Effect.gen(function* () {
          // A freshly attached IAM policy is eventually consistent — the first
          // invocations after deploy can still see AccessDenied. Poll (bounded)
          // until the call stops failing authorization, then assert.
          const body = yield* HttpClient.execute(
            HttpClientRequest.post(`${baseUrl}/complete-bogus`),
          ).pipe(
            Effect.flatMap((response) =>
              response.status === 200
                ? response.json
                : Effect.fail(new Error(`Function not ready: ${response.status}`)),
            ),
            Effect.map((json) => json as { ok: boolean; tag: string }),
            // `repeat` only re-runs successes: a transient 502 from the
            // Function URL (cold start right after deploy) must be retried
            // too, the way the /health probe is.
            Effect.retry({ schedule: Schedule.spaced("3 seconds"), times: 10 }),
            Effect.repeat({
              until: (b) => b.tag !== "AccessDenied" && b.tag !== "AccessDeniedException",
              schedule: Schedule.spaced("3 seconds"),
              times: 10,
            }),
          );
          // The SDK call reached AWS with valid credentials/policy: either it
          // succeeded, or it failed for a resource reason (no active lifecycle
          // action) — never an authorization failure.
          expect(["AccessDenied", "AccessDeniedException"]).not.toContain(body.tag);
        }),
      { timeout: 60_000 },
    );
  },
);