What Is a Load Balancer and How Does It Work?

Load distribution methods

Health checks

The load balancer regularly checks whether the servers behind it are working. If a server doesn't respond or returns errors, it stops sending requests to it; once the server recovers, it brings it back into service. That way, a single server going down doesn't show up as an error for visitors. Setting up a simple /health address in your application that also checks the database connection makes health checks more meaningful.

Sessions and shared data

When there are several servers behind the load balancer, a user's consecutive requests can go to different servers. That's why:

• Keep sessions in a shared place. If sessions sit in a single server's memory or on its disk, the user is logged out when they land on another server. Keeping sessions in a shared store such as Redis is the cleanest fix; the details are in our What is Redis? article.
• Share uploaded files. Files users upload should be kept on storage all servers can reach, or in object storage.
• There's usually a single database. While application servers are replicated, they all connect to the same database. Make the database reachable from the servers only over a private network.

If these aren't possible, you can always send the same user to the same server with IP hash or sticky sessions; however, this can lead to the load being distributed unevenly.

Load balancing with Nginx

As well as working as a reverse proxy, Nginx also works as a simple and powerful load balancer. The servers behind it are defined in an upstream block:

upstream app {
    least_conn;
    server 10.8.0.11:3000 max_fails=3 fail_timeout=30s;
    server 10.8.0.12:3000 max_fails=3 fail_timeout=30s;
    server 10.8.0.13:3000 backup;
}

server {
    listen 443 ssl;
    server_name yourdomain.com;

    location / {
        proxy_pass http://app;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;
    }
}

• least_conn: Sends requests to the server with the fewest active connections. If you remove the line, the default method is round robin.
• max_fails and fail_timeout: If a server fails repeatedly within the given time, it's taken out of service for a while.
• backup: A backup server used only while all the other servers are down.

We cover Nginx's basic configuration and SSL in our How to install Nginx guide. Instead of exposing the ports of the servers behind it to the internet, keep them on a private network only the load balancer can reach; you can use WireGuard to connect the servers to each other securely.

HAProxy and other options

• HAProxy: An open-source tool focused purely on load balancing and proxying, proven at very high traffic. It offers advanced health checks and detailed statistics, and it can also load balance at the TCP level.
• DNS round robin: Publishing several A records for the same domain is the simplest distribution method. However, DNS doesn't know when a server has gone down and can keep sending visitors to it; it isn't a substitute for a real load balancer.
• Cloud and provider load balancers: Some providers offer managed load balancer services; setup and redundancy are the provider's responsibility.

Making the load balancer itself redundant

A single load balancer is itself a single point of failure: if it goes down, the site can't be reached even if all the servers behind it are healthy. In critical systems, two load balancers run, and when one goes down, the other takes over the traffic. Before building this kind of setup, find out what network features your provider offers.

When do you need a load balancer?

Most projects don't need a load balancer. Try these first:

1. Optimize the application and the database. Slow queries and missing caching are much cheaper to fix than adding servers; take a look at our Why is my server slow? article.
2. Use caching and a CDN. Move static content to a CDN and repeated queries to Redis.
3. Move to a more powerful server. Vertical scaling is much simpler than horizontal scaling.

A load balancer makes sense when:

• A single server's resources can't carry the traffic, even after optimization.
• The service must not go down while a server fails or is under maintenance.
• You want to release new versions of your application without downtime.

Nodesty servers for load balancing

Nodesty AMD Ryzen 9 9950X VPS plans can be used as application servers thanks to their high single-core speed, and AMD EPYC 7763 VPS plans as database and cache servers thanks to their generous RAM. The Frankfurt and New York locations of the 9950X plans let you run servers in different regions. All plans include NVMe storage, nShield DDoS protection and free daily backups kept for 7 days. On virtual servers, bandwidth usage and high-traffic reverse proxy applications are subject to our Fair Use Policy; for setups that will carry very high traffic, consider our dedicated servers, where the hardware is reserved entirely for you.