Load Balancing Strategy for Production Frappe Deployment

weldonet · February 18, 2026, 10:59pm

Hello everyone,

I’m deploying Frappe Framework in production on AWS and I could not find clear documentation about recommended load balancing strategies.

My scenario:

Production environment
Hosted on AWS (EC2)
Nginx + Supervisor (standard bench production setup)
Redis (cache + queue)
MariaDB
Background workers enabled
Separate FastAPI microservice integrated via RabbitMQ
Multiple clients and increasing workload

Questions:

What is the recommended approach for horizontal scaling in Frappe?
Is it safe to run multiple bench instances behind a load balancer?
How should Redis and Socket.IO be handled in a multi-instance environment?
What is the recommended setup for high availability?
Is Docker/Kubernetes officially recommended for scaling?

I couldn’t find detailed guidance in the official documentation about load balancing and HA setups.

Any real-world architecture examples would be highly appreciated.

Thanks!

brian_pond · February 19, 2026, 12:18am

I don’t have answers to all your questions. But I can say a few things from experience:

Moving the MariaDB database to a separate host is a relatively simple + helpful win. This yields immediate performance gains, because MariaDB uses lots of memory, CPU, and disk I/O.

Just don’t relocate MariaDB too far away, in terms of network latency. The framework and MariaDB are very chatty with each other. You want them to share a low-latency TCP network. Otherwise separating them can actually make performance worse, instead of better.
I don’t think it’s “unsafe” to run multiple bench instances behind a load balancer. Not sure your use case. But cannot think of any real harm in that. I’ve run many different instances of bench, with varying Versions, all coexisting nicely on one host. No problems there.
Docker/Kubernetes is definitely what (most) people here will recommend to you. Even for single-instance and single-tenant, it’s the recommended way to install nowadays (last I heard, anyway)

asieftejani · February 19, 2026, 3:54pm

@weldonet

This might help with point 5

github.com/frappe/frappe_docker

docs/02-setup/07-single-server-example.md

main

# Single Server Example

This guide demonstrates setting up multiple Frappe/ERPNext benches (projects) on a single server with shared infrastructure components.

In this use case we have a single server with a static IP attached to it. It can be used in scenarios where one powerful VM has multiple benches and applications or one entry level VM with single site. For single bench, single site setup follow only up to the point where first bench and first site is added. If you choose this setup you can only scale vertically. If you need to scale horizontally you'll need to backup the sites and restore them on to cluster setup.

We will setup the following:

- Install docker and docker compose v2 on linux server.
- Install traefik service for internal load balancer and letsencrypt.
- Install MariaDB with containers.
- Setup project called `erpnext-one` and create sites `one.example.com` and `two.example.com` in the project.
- Setup project called `erpnext-two` and create sites `three.example.com` and `four.example.com` in the project.

Explanation:

Single instance of **Traefik** will be installed and act as internal loadbalancer for multiple benches and sites hosted on the server. It can also load balance other applications along with frappe benches, e.g. wordpress, metabase, etc. We only expose the ports `80` and `443` once with this instance of traefik. Traefik will also take care of letsencrypt automation for all sites installed on the server. _Why choose Traefik over Nginx Proxy Manager?_ Traefik doesn't need additional DB service and can store certificates in a json file in a volume.

Single instance of **MariaDB** will be installed and act as database service for all the benches/projects installed on the server.

This file has been truncated. show original

weldonet · February 19, 2026, 7:11pm

brian_pond:

I don’t have answers to all your questions. But I can say a few things from experience:

Moving the MariaDB database to a separate host is a relatively simple + helpful win. This yields immediate performance gains, because MariaDB uses lots of memory, CPU, and disk I/O.Just don’t relocate MariaDB too far away, in terms of network latency. The framework and MariaDB are very chatty with each other. You want them to share a low-latency TCP network. Otherwise separating them can actually make performance worse, instead of better.

I don’t think it’s “unsafe” to run multiple bench instances behind a load balancer. Not sure your use case. But cannot think of any real harm in that. I’ve run many different instances of bench, with varying Versions, all coexisting nicely on one host. No problems there.

Docker/Kubernetes is definitely what (most) people here will recommend to you. Even for single-instance and single-tenant, it’s the recommended way to install nowadays (last I heard, anyway)

Hi Brian — thanks for sharing your experience, this is very helpful.

I completely agree that moving MariaDB to a dedicated host is one of the quickest wins. In my case I’m planning to keep the DB within the same low-latency VPC subnet (likely via RDS) precisely because, as you mentioned, Frappe is quite chatty with the database.

Regarding multiple bench instances behind a load balancer — good to hear you’ve run this successfully. My use case is a multi-tenant production environment with growing concurrency, so my main concern is ensuring everything remains properly stateless (sessions in Redis, shared file storage, and consistent Socket.IO behavior across nodes). If you’ve seen any pitfalls specifically around realtime or file handling in multi-node setups, I’d be very interested to hear.

On Docker/Kubernetes: I see many recommendations in that direction as well. For now I’m evaluating whether to first scale horizontally with traditional EC2 + ALB and only later move to containers once operational complexity justifies it.

Appreciate you taking the time to respond — very useful input.

weldonet · February 19, 2026, 7:13pm

Thanks @asieftejani — this is helpful, especially the Traefik-based setup.

From what I understand, this example is mainly focused on consolidating multiple benches on a single host and scaling vertically, which makes sense for certain scenarios. My current goal is to prepare for horizontal scaling across multiple nodes on AWS (ALB in front, shared Redis, external MariaDB, etc.), so I’m trying to map what additional components are required beyond the single-server pattern.

In particular, I’m evaluating:

stateless bench nodes behind a load balancer
shared file storage strategy
Socket.IO behavior across multiple instances
HA considerations for Redis and workers

If you (or others) have seen a recommended reference architecture for true multi-node production setups with Frappe, I’d really appreciate any pointers.

Thanks again for sharing the link — very useful context.