fix(db): make V74 self-healing on databases that ran the base branch

Dedup DELETE + note clamp before the DDL so the unique index and CHECK
cannot fail mid-migration (failed Flyway row -> backend boot loop) on a
DB that served writes under the old code (no dedup guard, 5000-char
notes). No-ops on a clean database.

Note: this changes V74's checksum — dev databases that already ran V74
on this branch need 'flyway repair' (or a fresh DB). CI and prod run
from V73 or clean and are unaffected.

Review round 3: Tobias (1).

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This commit is contained in:
Marcel
2026-06-11 07:51:20 +02:00
parent 2e9902e8a7
commit 98805f596a

View File

@@ -11,6 +11,22 @@
-- 2. CHECK on note length: mirrors chk_text_length on transcription_blocks.
-- 2000 is the spec'd limit — JourneyItemService.MAX_NOTE_LENGTH, the frontend
-- maxlength, and the i18n error message all agree (#793).
--
-- Defensive cleanup first: a database that served writes on the base branch
-- (no dedup guard, MAX_NOTE_LENGTH = 5000) can hold rows that would make the
-- DDL below fail mid-migration and boot-loop the backend on a failed Flyway
-- row. Both statements are no-ops on a clean database.
-- Keep the earliest-positioned row of each (geschichte, document) pair.
DELETE FROM journey_items a
USING journey_items b
WHERE a.geschichte_id = b.geschichte_id
AND a.document_id = b.document_id
AND a.document_id IS NOT NULL
AND a.position > b.position;
-- Clamp over-long notes written under the old 5000-char service limit.
UPDATE journey_items SET note = left(note, 2000) WHERE length(note) > 2000;
CREATE UNIQUE INDEX uq_journey_items_geschichte_document
ON journey_items (geschichte_id, document_id)