fix(aws/autoscaling): harden AutoScalingGroup reconcile + add lifecycle convergence tests

[sam-goodwin]

Hardens the AutoScalingGroup resource as part of the per-resource hardening sweep on top of the reconcile migration.

Reconciler changes

- Effect.catch((error: any) =>
-   error?._tag === "AlreadyExistsFault"
-     ? Effect.void
-     : Effect.fail(error),
- ),
+ Effect.catchTag("AlreadyExistsFault", () => Effect.void),

Blanket Effect.catch swallowed every error. Scoped to AlreadyExistsFault only — race / restart-after-crash recovery — so auth, throttling, and validation errors propagate.

  existing = yield* describeGroup(autoScalingGroupName).pipe(
-   Effect.filterOrFail(Boolean, () => new Error(...)),
+   Effect.flatMap((group) =>
+     group ? Effect.succeed(group)
+           : Effect.fail(new AutoScalingGroupNotReadyAfterCreate()),
+   ),
    Effect.retry({
-     while: () => true,
+     while: (e) => e._tag === "AutoScalingGroupNotReadyAfterCreate",
      schedule: Schedule.recurs(8).pipe(...),
    }),
  );

The post-create read previously retried on any error including auth/validation. Now retries only the actual "still missing in describe" case.

+ .pipe(
+   Effect.retry({
+     while: (e) =>
+       e._tag === "ScalingActivityInProgressFault" ||
+       e._tag === "ResourceContentionFault",
+     schedule: Schedule.fixed("2 seconds").pipe(
+       Schedule.both(Schedule.recurs(15)),
+     ),
+   }),
+ );

updateAutoScalingGroup and deleteAutoScalingGroup now retry on transient ScalingActivityInProgressFault and ResourceContentionFault. Delete also retries on ResourceInUseFault. Attach/Detach LB target groups retry on the errors their distilled unions actually carry (ResourceContention, InstanceRefreshInProgress).

+ const attrs = toAttributes(group);
+ return (yield* hasAlchemyTags(id, attrs.tags))
+   ? attrs
+   : Unowned(attrs);

read now marks foreign ASGs as Unowned, gating takeover behind adopt(true) instead of silently re-tagging.

New lifecycle tests

packages/alchemy/test/AWS/AutoScaling/AutoScalingGroup.test.ts. Skipped unless TEST_LAUNCH_TEMPLATE_ID and TEST_SUBNET_IDS are set (real EC2 fleet required).

• redeploy with same props is a no-op
• reconcile resets desiredCapacity mutated out-of-band via setDesiredCapacity
• changing autoScalingGroupName triggers replace
• in-place modification of minSize/maxSize/desiredCapacity
• destroying an already-deleted ASG is a no-op
• adopt(true) re-tags a foreign ASG

Distilled patch

No patch needed for this pass. ScalingActivityInProgressFault is genuinely transient and a candidate for withRetryableError in distilled, but the reconciler-level retry handles it for now without depending on the AWS retry layer. InvalidInstanceLifecycleStateException is context-dependent and stays BadRequestError + ConflictError only.

[alchemy-version-bot]

Install the packages built from this commit:

alchemy

bun add alchemy@https://pkg.ing/alchemy/0f419cd

@alchemy.run/better-auth

bun add @alchemy.run/better-auth@https://pkg.ing/@alchemy.run/better-auth/0f419cd

@alchemy.run/pr-package

bun add @alchemy.run/pr-package@https://pkg.ing/@alchemy.run/pr-package/0f419cd

[alchemy-version-bot]

Website Preview Deployed

URL: https://alchemyeffectwebsite-worker-pr-227-eog3u7er6dtvr2om.testing-2b2.workers.dev

Built from commit 0f419cd.

---

This comment updates automatically with each push.