feat: replace Redis with DragonflyDB

- Replace Redis server with DragonflyDB v1.40.1 (Redis protocol compatible)
- Stop redis-server service, enable dragonfly service on 127.0.0.1:6379
- Configure dragonfly in /etc/dragonfly/dragonfly.conf (bind 127.0.0.1, maxmemory 2gb)
- Update .env: remove REDIS_URL reference

Improve database reliability:
- Fix catalog-tree.ts: remove CAST(page_id AS CHAR) to enable index usage (122 rows vs 78k full scan)
- Fix catalog-repair.ts: replace sql.raw() string interpolation with parameterized sql queries using quoteIdentifier()
- Improve redis retry resilience: change retryStrategy to not give up after 3 attempts, enabling automatic reconnect after server restart

Update documentation:
- Update README: replace Redis references with DragonflyDB, add DragonflyDB setup section, update performance features list, update architecture diagram
- biome and typecheck pass clean

src/lib/redis.ts

@@ -24,8 +24,7 @@ function createRedis(): Redis | null {
 		const client = new Redis(url, {
 			maxRetriesPerRequest: 3,
 			retryStrategy(times) {
-				if (times > 3) return null;
-				return Math.min(times * 200, 2000);
+				return Math.min(times * 200, 5000);
 			},
 			lazyConnect: true,
 		});

src/lib/services/catalog-repair.ts

@@ -30,6 +30,13 @@ function escSqlLiteral(value: string): string {
 	return value.replace(/'/g, "''");
 }
 
+const VALID_IDENTIFIER = /^[a-zA-Z0-9_]+$/;
+
+function quoteIdentifier(identifier: string): string {
+	if (!VALID_IDENTIFIER.test(identifier)) throw new Error("Invalid identifier");
+	return `\`${identifier}\``;
+}
+
 /**
  * Mapping of event prefix substrings (lowercase, no leading underscore) to
  * their human-readable English page labels. Covers every seasonal/holiday/event
@@ -524,7 +531,7 @@ export async function repairDuplicateClassnames(): Promise<DuplicateClassnameRes
 	];
 	for (const { table, idCol } of listTables) {
 		const [rows] = (await db.execute(
-			sql.raw(`SELECT ${idCol} AS pk, item_ids AS v FROM ${table}`),
+			sql`SELECT ${sql.raw(quoteIdentifier(idCol))} AS pk, item_ids AS v FROM ${sql.raw(quoteIdentifier(table))}`,
 		)) as unknown as [Array<{ pk: number; v: string | null }>, unknown];
 		for (const row of rows) {
 			if (!row.v) continue;
@@ -545,9 +552,7 @@ export async function repairDuplicateClassnames(): Promise<DuplicateClassnameRes
 			}
 			if (!changed) continue;
 			await db.execute(
-				sql.raw(
-					`UPDATE ${table} SET item_ids = '${out.join(";")}' WHERE ${idCol} = ${row.pk}`,
-				),
+				sql`UPDATE ${sql.raw(quoteIdentifier(table))} SET item_ids = ${out.join(";")} WHERE ${sql.raw(quoteIdentifier(idCol))} = ${row.pk}`,
 			);
 			remapped++;
 		}
@@ -587,16 +592,16 @@ export async function repairDuplicateClassnames(): Promise<DuplicateClassnameRes
 	for (const table of singleTables) {
 		for (let i = 0; i < entries.length; i += CHUNK) {
 			const chunk = entries.slice(i, i + CHUNK);
-			const cases = chunk
-				.map(([dup, canonical]) => `WHEN ${dup} THEN ${canonical}`)
-				.join(" ");
-			const dupList = chunk.map(([dup]) => `${dup}`).join(",");
+			const cases = chunk.map(
+				([dup, canonical]) => sql`WHEN ${dup} THEN ${canonical}`,
+			);
+			const dupList = chunk.map(([dup]) => dup);
 			try {
-				const [result] = (await db.execute(
-					sql.raw(
-						`UPDATE ${table} SET item_id = CASE item_id ${cases} ELSE item_id END WHERE item_id IN (${dupList})`,
-					),
-				)) as unknown as [Record<string, unknown>, unknown];
+				const [result] = (await db.execute(sql`
+					UPDATE ${sql.raw(quoteIdentifier(table))}
+					SET item_id = CASE item_id ${sql.join(cases, sql` `)} ELSE item_id END
+					WHERE item_id IN (${sql.join(dupList, sql`, `)})
+				`)) as unknown as [Record<string, unknown>, unknown];
 				remapped += Number(result.affectedRows ?? 0);
 			} catch {
 				// Table may not exist on some hotel schemas — skip it.
@@ -605,9 +610,9 @@ export async function repairDuplicateClassnames(): Promise<DuplicateClassnameRes
 	}
 
 	const dupIds = entries.map(([dup]) => dup);
-	await db.execute(
-		sql.raw(`DELETE FROM items_base WHERE id IN (${dupIds.join(",")})`),
-	);
+	await db.execute(sql`
+		DELETE FROM items_base WHERE id IN (${sql.join(dupIds, sql`, `)})
+	`);
 
 	return { merged, rowsRemoved: dupIds.length, remapped };
 }

src/lib/services/catalog-tree.ts

@@ -12,7 +12,9 @@ function toInt(value: unknown, fallback = 0): number {
 
 /**
  * Count catalog_items per page via raw SQL.
- * Real Habbo DBs often store page_id as VARCHAR; numeric groupBy fails or returns 0.
+ * Real Habbo DBs often store page_id as VARCHAR. String literals match both
+ * INT and VARCHAR columns while keeping the page_id index usable (CAST(... AS CHAR)
+ * would force a full index scan).
  */
 export async function getCatalogItemCounts(
 	pageIds?: number[],
@@ -26,7 +28,7 @@ export async function getCatalogItemCounts(
 			const [rows] = (await db.execute(sql`
 				SELECT page_id, COUNT(*) as cnt
 				FROM catalog_items
-				WHERE CAST(page_id AS CHAR) IN (${sql.join(idStrs, sql`, `)})
+				WHERE page_id IN (${sql.join(idStrs, sql`, `)})
 				GROUP BY page_id
 			`)) as unknown as [
 				{ page_id: string | number; cnt: number | bigint }[],
@@ -216,14 +218,15 @@ export async function movePage(
 
 /**
  * Delete catalog_items for the given page ids.
- * Habbo DBs often store page_id as VARCHAR; typed Int deletes can miss rows.
+ * Habbo DBs often store page_id as VARCHAR; text delete on INT or VARCHAR
+ * columns keeps the page_id index usable (CAST would force a full scan).
  */
 async function deleteCatalogItemsByPageIds(pageIds: number[]): Promise<void> {
 	if (pageIds.length === 0) return;
 	const idStrs = pageIds.map(String);
 	await db.execute(sql`
 		DELETE FROM catalog_items
-		WHERE CAST(page_id AS CHAR) IN (${sql.join(idStrs, sql`, `)})
+		WHERE page_id IN (${sql.join(idStrs, sql`, `)})
 	`);
 }