ministackorg/ministack v1.3.8

on GitHub

What's Changed

[1.3.8] — 2026-04-21

Added

• S3 s3:TestEvent on PutBucketNotificationConfiguration — configuring bucket notifications now delivers a flat s3:TestEvent payload (no Records wrapper) to every SQS / SNS / Lambda destination, matching real AWS S3 behaviour so tooling that listens for the test event on bucket setup works locally. Contributed by @nigel-campbell (#416)

Fixed

• Lambda persistence crash on warm start with any event source mapping — lambda_svc.restore_state() called _ensure_poller() when restoring ESMs, but the module-level invocation ran at line 170 while _ensure_poller was defined ~3,500 lines later. Warm starts with PERSIST_STATE=1 and an ESM in lambda.json raised NameError: name '_ensure_poller' is not defined on every Lambda request until the state file was deleted. The module-level load/restore is now at the bottom of the file, after every helper it may call. Reported by @whittin3. Fixes #422
• DynamoDB SSEDescription used the request shape instead of the response shape — CreateTable stored the caller's SSESpecification (Enabled/KMSMasterKeyId) directly as the response SSEDescription, which is missing the Status field Terraform v6 waits on; UpdateTable silently ignored SSESpecification. Warm-boot terraform apply on any encrypted table hung forever with unexpected state '', wanted target 'DISABLED, ENABLED'. Fixed by converting spec → description (Status, SSEType, KMSMasterKeyArn) at create + update, plus a one-shot migration in restore_state for legacy persisted tables. Reported by @whittin3. Fixes #418
• test_s3_put_notification_sends_test_event was flaky under parallel load — background thread delivering s3:TestEvent raced the test's poll. PutBucketNotification now delivers the test event synchronously (matching AWS effective behaviour) before returning; removes the race and also fixes multi-tenant delivery where the worker thread lost the caller's account contextvar.

Changed

• Hardened persisted-state restore across every service — 35 service modules wrapped their module-level load_state() + restore_state() invocation in try/except so a corrupt or schema-incompatible {service}.json logs and continues with a fresh store instead of breaking the service at import. Audited every restore_state / load_persisted_state for forward references (0 other violators) and unsafe data[key] subscript accesses (all guarded). This closes the entire class of bug around #422.