Sync app.storage between multiple instances

[rodja]

With Fly.io, Google Cloud Run, Docker Swarm, Kubernetes etc. a NiceGUI app can be is deployed to multiple instances which get fed requests by load balancer. In this scenario each instance will create their own .nicegui directory for storage. Depending on where a user requests are placed the storage data created by one instance may not be available to the other.

NiceGUI should allow a simple way to keep multiple instances in sync.

[rodja]

The most popular solution to sync state between multiple instances is Redis. Maybe we could make the storage solution configurable? Then, instead of writing to disk a Redis database can be used as a shared storage across all instances.

[Smug246]

something like that sounds good to me

[rodja]

We recently introduced a way to sponsor us, so we could spent more time on NiceGUI.

[rodja]

Anyone interested in creating a draft?

[Smug246]

id pay to get the process sped up. rly need it for the deployment of my site

[rodja]

Thanks @Smug246. In the meantime we created the possibility to sponsor us via GitHub. If you are still interested, you could coordinate with #1606 (comment) to offer sponsoring to get this done.

[Smug246]

so i managed to make a basic work around that purely works for my site which uses https://console.upstash.com/ and this py lib: https://github.com/upstash/redis-python.

i know the example below isnt minimal but it shows how you can use it. Note: this works when deploying with fly.io but doesnt work with on_air because on_air generates a new browser id everytime.

from nicegui import ui, app
import hashlib
from fastapi.responses import RedirectResponse
import psycopg2
from upstash_redis.client import Redis
import json
import asyncio

primary_color = '#2d3748'

# NOTE: this is the external database, you can just make a local one for testing
external_database_url = ""

# NOTE: this is an upstash redis database, you can make one for free here: https://console.upstash.com/
redis_client = Redis(
    url="",
    token=""
)

admin_users = ['lee81', 'ff24']
session_ids = []


def hash_data(data):
    sha256 = hashlib.sha256()
    sha256.update(data.encode())
    return sha256.hexdigest()


def get_session_identifier():
    session_identifier = app.storage.browser['id']
    if session_identifier in session_ids:
        return session_identifier
    else:
        session_ids.append(session_identifier)
        return session_identifier
    

def get_redis_data(key):
    return redis_client.get(key)


@ui.page('/')
def confirm_delivery():
    user_info_json = get_redis_data(get_session_identifier())
    if user_info_json is None:
        return RedirectResponse('/login')

    user_info = json.loads(user_info_json)
    
    if not user_info.get('authenticated'):
        return RedirectResponse('/login')

    with ui.header(fixed=False).style(f'background-color: {primary_color};'):
        with ui.row().style('align-items: center;'):
            ui.label('test').style(
                'font-size: 25px; font-weight: bold;')
    
    ui.link('Upload Document', '/upload_document')
            

@ui.page('/upload_document')
def upload_document():
    user_info_json = get_redis_data(get_session_identifier())
    if user_info_json is None:
        return RedirectResponse('/login')
    
    user_info = json.loads(user_info_json)

    if user_info.get('username') not in admin_users:
        return RedirectResponse('/')
    
    ui.label('hello world')


@ui.page('/login')
def login():
    async def try_login():
        try:
            input_username = username.value
            input_password = password.value

            conn = psycopg2.connect(external_database_url)
            c = conn.cursor()

            encrypted_username = hash_data(input_username)
            encrypted_password = hash_data(input_password)

            c.execute(
                'SELECT * FROM login_info WHERE username = %s::bytea AND password = %s::bytea',
                (encrypted_username, encrypted_password,)
            )
            result = c.fetchone()
            if result:
                if encrypted_password == bytes(result[1]).decode('utf-8'):
                    session_id = get_session_identifier()

                    session_data = {
                        "username": input_username,
                        "authenticated": True,
                        "haulier": bytes(result[2]).decode('utf-8'),
                    }

                    session_json = json.dumps(session_data)
                    redis_client.set(session_id, session_json)

                    ui.notify('Login Successful', color='positive')
                    await asyncio.sleep(1.5)
                    ui.open('/')
                else:
                    ui.notify('Invalid username or password', color='negative')
                    username.set_value('')
                    password.set_value('')
            else:
                ui.notify('Invalid username or password', color='negative')
                username.set_value('')
                password.set_value('')
        except Exception as e:
            print(e)

    ui.query('body').style(f'background-color: {primary_color};')
    with ui.card().style('width: 250px;').classes('absolute-center'):
        username = ui.input('Enter Your Username').style(
            'width: 225px;').props('outlined color="emerald-500"')
        password = ui.input('Enter Your Password', password=True, password_toggle_button=True).style(
            'width: 225px;').props('outlined color="emerald-500"').on('keydown.enter', try_login)
        ui.button('Log In', on_click=try_login, color=primary_color).style(
            'color: white; margin-left: auto; margin-right: auto;')


ui.run(title='test', storage_secret='test', reload=False)

[rodja]

This feature is already quite highly voted (add your's if you need it). There was also a discussion on Discord.

If there are not a lot of data, a workaround could be to use the app.storage.browser because the data will be in a session cookie and is thereby available by any instance the user connects to.

[Smug246]

how would redis storage or something like that be implemented in terms of configuration in nicegui. would it be an option in ui.run() and then you could use a global variable? or maybe having it's own element could be better?

[rodja]

That is a very good question. @falkoschindler and me brainstormed a bit a few days ago because the number of storage options seems to explode: type (user, client, tab, general) x content (serializable, non-serializable) x backend (file, redis, ...). One solution would be to not provide default storages in the long run, but give the user the possibility to create their desired storage via class instantiation...

[hany]

I was looking online for a way to configure NiceGUI's storage centrally using something like Redis, and came across this discussion. Would definitely be a plus for any multi-instance configuration.

[juledwar]

Just came across this, I have the same problem where I want to scale out the app servers and have user storage.

I figured I could just mount a shared volume, but then checking out PersistentDict I see that there's no mutex or co-ordinated access at all, each process assumes total control of its backing file.

This is probably a very high priority improvement for anyone looking to do serious deployments at some scale.

I like the idea of the user instantiating their own storage class, but also having some built-in ones (the existing file-based one for simplicity, plus Postgres, Redis etc). I'd be tempted to have a stab at one but I'm nowhere near familiar enough with the code. 😞

[alydersen]

How would be go about to sponsor such an improvement @rodja ? We would be happy to instantiate a storage class if that was an option. Any way of overcoming multiple workers/threads issues for app.storage.user is very interesting to us and we would be willing to put some money behind this.

[rodja]

Thanks for offering your sponsorship for this. We estimate the implementation of a first usable Redis backend for app.storage.user will take several days. Every sponsorship counts but with one to two thousand dollars we could put it on top of the list and hopefully release something this year.

[alydersen]

Thanks @rodja . I've arranged a USD1000 on time sponsorship and a USD25 monthly. Thanks for making this available for us all, guys!