docs(adr): ADR-038 — domain event drives note-less journey-item cleanup on document delete (#805)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

docs/adr/038-domain-event-driven-journey-item-cleanup.md

@@ -0,0 +1,118 @@
+# ADR-038 — Domain event drives note-less journey-item cleanup on document delete
+
+**Status:** Accepted
+**Date:** 2026-06-11
+**Issue:** #805 (P1 — deleting a document linked via a note-less journey_item 500s at DB constraint)
+
+## Context
+
+Two constraints in V72 encode contradictory rules for a journey item that has a
+`document_id` but no `note`:
+
+- **`fk_journey_items_document` → `ON DELETE SET NULL`** — when a document is deleted,
+  Postgres nulls out `document_id`.
+- **`chk_journey_item_not_empty`** — requires at least one of `document_id` or `note`
+  to be non-null.
+
+A note-less item (`document_id` set, `note IS NULL`) satisfies the CHECK while the
+document exists. Deleting the document causes Postgres to attempt `SET NULL`, which
+would leave both columns null — a direct CHECK violation. Postgres aborts the
+transaction with a 500 that bypasses `GlobalExceptionHandler`.
+
+The natural fix — delete note-less items inside `DocumentService.deleteDocument` before
+`deleteById` runs — cannot call `JourneyItemService` directly: `JourneyItemService`
+already injects `DocumentService`, and Spring Framework 7 (used by Spring Boot 4)
+**fully prohibits constructor-injection cycles**. The application will not start if such
+a cycle is introduced.
+
+## Decision
+
+`DocumentService.deleteDocument` publishes a **`DocumentDeletingEvent`** (plain record,
+payload: `documentId` UUID only) via `ApplicationEventPublisher` **before**
+`documentRepository.deleteById`. A dedicated `@Component`
+`JourneyItemDocumentDeleteListener` in the `geschichte.journeyitem` package consumes
+this event and calls `journeyItemRepository.deleteNoteLessByDocumentId(documentId)`
+directly — bypassing `JourneyItemService` to avoid re-introducing the cycle and to
+suppress the per-item `JOURNEY_ITEM_REMOVED` audit emission (see audit decision below).
+
+### Load-bearing listener-phase choice: plain `@EventListener`
+
+The listener is annotated with `@EventListener` (not
+`@TransactionalEventListener(AFTER_COMMIT)`, not `@Async`). **This choice is
+load-bearing:**
+
+- **`AFTER_COMMIT` would break the fix entirely.** `AFTER_COMMIT` fires *after* the
+  surrounding transaction has committed. By that point, `documentRepository.deleteById`
+  has already executed and Postgres has already tried `ON DELETE SET NULL` — the
+  constraint violation fires before the listener ever runs.
+- **`@Async` would break rollback atomicity (AC-5).** An async listener runs on a
+  separate thread in its own transaction. If `deleteDocument` subsequently rolls back
+  (e.g. due to an unrelated failure), the listener's deletes are in a committed async
+  transaction and cannot be undone.
+- **Plain `@EventListener` runs synchronously in the publisher's thread and
+  transaction.** The listener's JPQL delete and the `deleteById` are a single atomic
+  unit: if either fails, both roll back together.
+
+### Repository method
+
+```java
+@Modifying(clearAutomatically = true)
+@Query("DELETE FROM JourneyItem i WHERE i.document.id = :documentId AND (i.note IS NULL OR i.note = '')")
+int deleteNoteLessByDocumentId(@Param("documentId") UUID documentId);
+```
+
+`i.document.id` (the real association path) is used instead of `i.documentId`: the
+transient `getDocumentId()` getter on `JourneyItem` makes Spring Data unable to resolve
+a derived query path (same trap documented at `JourneyItemRepository:33-44`).
+
+`clearAutomatically = true` invalidates the L1 cache so subsequent reads in the same
+session do not return stale entities.
+
+The predicate `(note IS NULL OR note = '')` covers the `note = ''` edge case that the
+service layer can never produce (normalizeNote converts blank strings to null), but that
+may exist via raw SQL inserts or legacy data. Whitespace-only notes (`note = '   '`)
+do not match and are preserved as note-carrying placeholders.
+
+### Audit decision
+
+The listener calls the repository directly rather than routing through
+`JourneyItemService.delete`. This deliberately bypasses the `JOURNEY_ITEM_REMOVED`
+audit emission: a document used in multiple journeys would otherwise produce N audit
+rows for a single user action. The `DOCUMENT_DELETED` entry written by `deleteDocument`
+is the sole audit record for the operation.
+
+### Boundary: documents must not depend on journey
+
+The event direction is `document → journey`, never the reverse. `DocumentService`
+publishes events it knows nothing about the consumers of; `JourneyItemService`'s
+dependency on `DocumentService` is unchanged and remains the only cross-domain
+reference. This direction is the prerequisite for the cycle constraint to hold.
+
+## Alternatives rejected
+
+- **DB trigger on `journey_items`** — trigger logic is opaque to Java developers,
+  invisible to code review, and not covered by the JPA test harness.
+- **RESTRICT instead of SET NULL** — breaks the existing note-carrying placeholder
+  UX: deleting a document with a note-carrying journey item would 409 instead of
+  preserving the item as a placeholder.
+- **Relax `chk_journey_item_not_empty`** — the constraint enforces a real invariant
+  (every item must have at least document or note). Removing it would allow empty rows.
+- **`@Lazy` on the `JourneyItemService → DocumentService` injection** — Spring Boot 4 /
+  Spring Framework 7 prohibits constructor-injection cycles regardless of `@Lazy`.
+- **Make `DocumentService` call `JourneyItemService`** — introduces the prohibited
+  cycle. Rejected at design time.
+
+## Consequences
+
+- **No schema change** — no new Flyway migration, no `db-orm.puml` /
+  `db-relationships.puml` update.
+- This is the **first custom domain event** in the codebase. No prior
+  `ApplicationEventPublisher` usage existed in `main/`. New cross-domain cleanup that
+  cannot use direct service calls should follow this pattern.
+- All tests that delete documents and then assert journey-item state **must route
+  through `DocumentService.deleteDocument`**, not `documentRepository.deleteById`.
+  The existing `JourneyItemIntegrationTest` tests that covered the note-carrying
+  placeholder UX have been updated accordingly.
+- The `DOCUMENT_DELETED` `AuditKind` was added as part of this fix to give AC-7's
+  audit assertion a positive check (absence-only assertions pass vacuously if all
+  auditing regresses).