expire old presences

apps/match/src/match.app.ts

@@ -3,6 +3,8 @@ import { useWorkersLogger } from 'workers-tagged-logger'
 
 import {
 	createRoomInstance,
+	deleteExpiredPresence,
+	getExpiredPresenceInstanceIds,
 	getJoinableInstance,
 	getOrCreateDormRoom,
 	getPresence,
@@ -20,7 +22,7 @@ import { validateAndGetAccountId } from '@repo/jwt'
 
 import type { Context } from 'hono'
 import type { Room, StoredPresence } from '@repo/domain'
-import type { App } from './context'
+import type { App, Env } from './context'
 
 /**
  * The matchmaking surface. Rooms and room instances are D1-backed (matchmaking
@@ -603,4 +605,31 @@ const app = new Hono<App>()
 	// Rooms flagged as requiring an RR+ subscription. No such queue yet → empty list.
 	.get('/rooms/requiring/rrplus', (c) => c.json([]))
 
-export default app
+/**
+ * Cron: sweep presence that has aged past its TTL. Reads already ignore expired rows,
+ * so this isn't about correctness of `/player` — it's that a player who crashed or
+ * hard-quit never matchmakes out of their instance, so nothing recomputes that
+ * instance's fullness and it can stay flagged full (and unjoinable) with nobody in it.
+ * Recompute the instances the expiring rows point at, *then* delete: the sweep is the
+ * only thing that notices those departures. Fullness is recomputed after the delete so
+ * the head-count no longer sees them.
+ */
+async function sweepExpiredPresence(env: Env): Promise<void> {
+	const staleInstanceIds = await getExpiredPresenceInstanceIds(env.DB)
+	const removed = await deleteExpiredPresence(env.DB)
+	for (const instanceId of staleInstanceIds) {
+		await refreshInstanceFullness(env.DB, instanceId)
+	}
+	// The tagged logger is request-scoped (its middleware never runs for a cron), so
+	// log plainly here — Workers observability picks it up either way.
+	console.log(
+		`presence sweep: removed ${removed} expired rows, refreshed ${staleInstanceIds.length} instances`
+	)
+}
+
+export default {
+	fetch: app.fetch,
+	scheduled: async (_controller, env, ctx) => {
+		ctx.waitUntil(sweepExpiredPresence(env))
+	},
+} satisfies ExportedHandler<Env>

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

@@ -1,4 +1,10 @@
-import { adminSecretsStore, env } from 'cloudflare:test'
+import {
+	adminSecretsStore,
+	createExecutionContext,
+	createScheduledController,
+	env,
+	waitOnExecutionContext,
+} from 'cloudflare:test'
 import { exports } from 'cloudflare:workers'
 import { beforeAll, describe, expect, test } from 'vitest'
 
@@ -9,7 +15,7 @@ import {
 	ROOM_INSTANCE_SCHEMA_DDL,
 } from '@repo/domain'
 
-import '../../match.app'
+import worker from '../../match.app'
 
 import type { Env } from '../../context'
 
@@ -580,6 +586,14 @@ describe('auth-gated endpoints', () => {
 
 	const nowSeconds = () => Math.floor(Date.now() / 1000)
 
+	/** Rows for an account, expired ones included — the sweep should leave none. */
+	const countPresenceRows = async (id: number): Promise<number> => {
+		const row = await env.DB.prepare('SELECT COUNT(*) AS n FROM presence WHERE account_id = ?1')
+			.bind(id)
+			.first<{ n: number }>()
+		return row?.n ?? 0
+	}
+
 	test('heartbeat refreshes presence when its TTL is close to lapsing', async () => {
 		// TTL about to lapse (well inside the refresh window).
 		const nearExpiry = nowSeconds() + 10
@@ -664,6 +678,28 @@ describe('auth-gated endpoints', () => {
 		expect((await getRoomInstance(env.DB, solo))?.isFull).toBe(false)
 	})
 
+	test('the cron sweep purges expired presence and frees the instance those players were in', async () => {
+		// A player fills SoloRoom, then vanishes without matchmaking out (a crash) —
+		// nothing recomputes fullness, so the instance sits full with nobody in it.
+		const solo = await matchmakeInto('5', '824')
+		expect((await getRoomInstance(env.DB, solo))?.isFull).toBe(true)
+		await env.DB.prepare(
+			"UPDATE presence SET data = json_set(data, '$.expiresAt', ?2) WHERE account_id = ?1"
+		)
+			.bind(824, nowSeconds() - 10)
+			.run()
+
+		// Driven through the module's own export rather than the `exports` proxy — a
+		// ScheduledController can't cross the isolate boundary the proxy serializes over.
+		const ctx = createExecutionContext()
+		await worker.scheduled(createScheduledController(), env, ctx)
+		await waitOnExecutionContext(ctx)
+
+		// Expired row gone, and the instance is joinable again.
+		expect(await countPresenceRows(824)).toBe(0)
+		expect((await getRoomInstance(env.DB, solo))?.isFull).toBe(false)
+	})
+
 	test('player/login, exclusivelogin and logout all preserve presence', async () => {
 		const headers = await bearer('9')
 		await exports.default.fetch(`${ORIGIN}/matchmake/dorm`, { method: 'POST', headers })

apps/match/wrangler.jsonc

@@ -15,6 +15,13 @@
 			"database_id": "local"
 		}
 	],
+	// Presence sweep. Rows expire on their own TTL (15m) and reads already ignore
+	// expired ones, so this is housekeeping: it purges them and recomputes the
+	// fullness of the instances the departed players were in (a crashed player never
+	// matchmakes out, so nothing else notices they left). Every 5 minutes.
+	"triggers": {
+		"crons": ["*/5 * * * *"]
+	},
 	"logpush": false,
 	// Shared Secrets Store holding the HS256 JWT signing key. Every worker binds the
 	// same store as JWT_SECRET so tokens signed by `auth` verify here. The "local"