[rooms] fix slow engagement counters

apps/rooms/migrations/0031_room_engagement_counters.sql

@@ -0,0 +1,27 @@
+-- Cheer and favorite counters on `room`, as columns — the cheer/favorite twins of
+-- `visits` (0011).
+--
+-- Generated from packages/domain/src/rooms-db.ts (ROOM_SCHEMA_DDL) — keep in sync.
+--
+-- Until now `Stats.CheerCount`/`FavoriteCount` were DERIVED on every read by aggregating
+-- the per-player `interaction` rows (`SELECT room_id, SUM(cheered), SUM(favorited) …
+-- GROUP BY room_id`). 0021's covering index made the per-room form cheap, but the browse
+-- and discovery feeds rank EVERY listable room, so they ran the unfiltered form — a walk
+-- of the whole index per call — and the discovery page runs the hot feed once per
+-- category row. That was the bulk of the database's reads.
+--
+-- Now each counter is kept current by the toggles themselves: the same batch that flips a
+-- player's `interaction.cheered`/`favorited` applies a `cheers = cheers ± 1` to the room,
+-- so a cheer still shows immediately and concurrent toggles can't lose each other. The
+-- per-player rows stay the record of WHO cheered; the columns are the count of them.
+--
+-- The backfill below is the one remaining full aggregate, run once here. `room_id` is a
+-- virtual generated column, so the correlated subquery resolves it per row; the sums come
+-- off `idx_interaction_room` without touching the table.
+
+ALTER TABLE room ADD COLUMN cheers INTEGER NOT NULL DEFAULT 0;
+ALTER TABLE room ADD COLUMN favorites INTEGER NOT NULL DEFAULT 0;
+
+UPDATE room SET
+	cheers = (SELECT COALESCE(SUM(cheered), 0) FROM interaction WHERE interaction.room_id = room.room_id),
+	favorites = (SELECT COALESCE(SUM(favorited), 0) FROM interaction WHERE interaction.room_id = room.room_id);

apps/rooms/src/openapi.ts

@@ -149,11 +149,12 @@ export const RoomTagDto = z.object({
 })
 
 /**
- * A room's engagement counters. `CheerCount`/`FavoriteCount` are aggregated from the
- * per-player `interaction` rows on every read. `VisitCount` is the room's lifetime
- * visits — the `room.visits` column, bumped by the `match` worker on every successful
- * matchmake into the room. Nothing records distinct visitors, so `VisitorCount` stays
- * at 0.
+ * A room's engagement counters. `CheerCount`/`FavoriteCount` are the `room.cheers`/
+ * `favorites` columns, moved by the cheer/favorite toggles as they flip the per-player
+ * `interaction` row (so a count is how many players currently have the flag set).
+ * `VisitCount` is the room's lifetime visits — the `room.visits` column, bumped by the
+ * `match` worker on every successful matchmake into the room. Nothing records distinct
+ * visitors, so `VisitorCount` stays at 0.
  */
 export const RoomStatsDto = z.object({
 	CheerCount: z.int(),

apps/rooms/src/test/integration/api.test.ts

@@ -4769,7 +4769,7 @@ describe('rooms endpoints', () => {
 		expect(otherGet).toMatchObject({ Cheered: false, Favorited: false })
 	})
 
-	it('room Stats aggregate cheers/favorites from the interaction table', async () => {
+	it('room Stats count cheers/favorites on the room row, kept by the toggles', async () => {
 		type Stats = {
 			CheerCount: number
 			FavoriteCount: number
@@ -4818,6 +4818,21 @@ describe('rooms endpoints', () => {
 			VisitCount: 0,
 		})
 
+		// The counts are columns on the room, not an aggregate taken on read — the toggles
+		// keep them. A repeated clear changes nothing (the flag was already off), and neither
+		// does clearing a flag a player never set, so neither counts down again.
+		await interact('562', 'cheer', 'DELETE')
+		await interact('563', 'favorite', 'DELETE')
+		expect(await direct()).toMatchObject({ CheerCount: 1, FavoriteCount: 1 })
+		expect(
+			await env.DB.prepare('SELECT cheers, favorites FROM room WHERE room_id = 15').first()
+		).toEqual({ cheers: 1, favorites: 1 })
+		// Toggling off and on again lands back where it started.
+		await interact('561', 'cheer', 'PUT')
+		expect(await direct()).toMatchObject({ CheerCount: 0 })
+		await interact('561', 'cheer', 'PUT')
+		expect(await direct()).toMatchObject({ CheerCount: 1 })
+
 		// VisitCount is the `room.visits` column (what match bumps on each matchmake),
 		// served on every read of the room — here and in the search results — and it
 		// survives the cheer/favorite aggregation rather than being zeroed by it.