Data
Files and data
Where your app may write, what is kept and what is wiped. SQLite, uploads and DATA_DIR.
While an app or bot runs, its code is read-only. There are two places it can write, and they behave differently.
What is kept
| Place | Kept | For |
|---|---|---|
/app/data |
Yes — across deploys and restarts | Database files, uploads, sessions |
/tmp |
No — wiped on every restart and deploy, 128 MB | Scratch files |
The code folder (/app) |
Not writable | — |
/app/data is the project’s own disk. Its path is also in the DATA_DIR variable; read it from there rather than hard-coding it.
If your repository has a data folder, the disk covers it: the files in it are not visible to the running app.
Projects that run from a Dockerfile or a ready image get the disk at /data, with DATA_DIR=/data.
SQLite lands on the disk by itself
A SQLite file whose name is written literally in the code is linked to the disk for you. For example:
sqlite3.connect('bot.db')- SQLAlchemy:
sqlite:///app.db - better-sqlite3:
new Database('app.db')
That file is written to /app/data and survives a deploy. If the name comes from a variable or a config file, the platform cannot see it; take the path from DATA_DIR instead.
Examples
Python, sqlite3:
import os, sqlite3
db = sqlite3.connect(os.path.join(os.environ["DATA_DIR"], "bot.db"))
An aiogram bot keeping its users in SQLite:
import os
import aiosqlite
DB = os.path.join(os.environ.get("DATA_DIR", "data"), "users.db")
async def save_user(user_id: int):
async with aiosqlite.connect(DB) as db:
await db.execute("CREATE TABLE IF NOT EXISTS users (id INTEGER PRIMARY KEY)")
await db.execute("INSERT OR IGNORE INTO users VALUES (?)", (user_id,))
await db.commit()
Node, better-sqlite3:
import path from 'node:path'
import Database from 'better-sqlite3'
const db = new Database(path.join(process.env.DATA_DIR ?? 'data', 'app.db'))
Prisma: add this to the project’s variables.
DATABASE_URL=file:/app/data/app.db
A folder for uploads:
const uploads = path.join(process.env.DATA_DIR, 'uploads')
fs.mkdirSync(uploads, { recursive: true })
Real data belongs in a database
The disk belongs to one app: no other service sees it, and it is not backed up. SQLite is fine for a small bot or a prototype. Once there are customers, orders or payments, open a PostgreSQL database: it has backups and several services can connect to it. For lots of images and large files, use an S3 bucket.