chores: updated

[KodeSage]

Description

Locks in the schema-level guarantees of the device_prekeys table (signed + one-time prekeys in one table, discriminated by key_type) with regression coverage.

The table itself already landed with the unified prekey work: apps/backend/src/db/schema.ts defines every column, index, and constraint the issue calls for, and apps/backend/drizzle/0000_stale_mandarin.sql migrates it. Verified against the issue's acceptance criteria:

Requirement	Enforced by
id uuid PK, keyId integer, publicKey text, signature nullable, consumed boolean default false, createdAt	column definitions
keyType pgEnum (signed / one_time)	prekey_type enum
deviceId uuid FK, cascade delete	device_prekeys_device_id_devices_id_fk
Unique (deviceId, keyType, keyId)	device_prekeys_device_type_keyid_idx
Partial index on (deviceId) WHERE key_type='one_time' AND consumed=false	device_prekeys_one_time_available_idx
Exactly one active signed prekey per device	partial unique device_prekeys_signed_device_idx WHERE key_type='signed'
Signed prekey rows require a non-null signature	DB check device_prekeys_signed_requires_signature
Consuming a one-time prekey flips consumed=true atomically	SELECT … FOR UPDATE SKIP LOCKED inside a transaction, routes/users.ts

The gap this PR closes: nothing pinned those invariants. Every one of them lives in DDL rather than route code, so an edit to schema.ts could silently drop the check constraint or a partial index and no test would fail.

Changes:

1. New devicePrekeys.schema.test.ts — 15 tests asserting the table shape and invariants against the live Drizzle definition via getTableConfig. Partial-index and check-constraint predicates are serialized through PgDialect and compared as SQL text, so weakening a predicate fails the build. A final block scans the generated migration DDL, which catches the case where a constraint exists in TypeScript but was never generated into a migration and so never reaches Postgres.
2. Tightened the atomic-consume assertion in users.bundle.test.ts — it previously only asserted that some update ran, which would still pass if the row lock or SKIP LOCKED were removed. It now asserts the transaction, for('update', { skipLocked: true }), and set({ consumed: true }) specifically, pinning both the atomicity and the flip-don't-delete behaviour that keeps bundle-fetch history auditable.

Note on the FK target: the issue specifies deviceId → user_devices.id. It points at devices.id instead, which is correct rather than a deviation — user_devices was merged into devices, now the single canonical device registry that messages.senderDeviceId, messageEnvelopes.recipientDeviceId, and pushSubscriptions.deviceId all reference. Left as is.

Type of change

• Bug fix
• New feature
• Documentation update
• Other — test coverage (no production code changed; the diff is test files only)

Issue numbers

Fixes #154

Testing

• Full backend suite green: 413 tests across 42 files.
• The new tests were checked against a negative control — temporarily deleting the signed-prekey partial unique index and the check constraint from schema.ts failed exactly the three relevant tests (allows exactly one signed prekey per device, allows many one-time prekeys per device, requires a signature on signed prekey rows), confirming the assertions are not vacuous. schema.ts was then restored and is unmodified in this PR.
• ESLint and Prettier clean on both touched files.
• tsc --noEmit reports 9 errors, all pre-existing on main (in roomManager, uploads, stellarListener) and none in the files touched here.

Checklist

• I have read the contributing guidelines
• I have tested my changes locally
• My code follows the project's coding standards

[drips-wave]

@KodeSage Great news! 🎉 Based on an automated assessment of this PR, the linked Wave issue(s) no longer count against your application limits.

You can now already apply to more issues while waiting for a review of this PR. Keep up the great work! 🚀

Learn more about application limits