Kamal proxy: multi-app
This course covers running more than one application on a single server behind kamal-proxy. It explains how one port 443 is shared by routing on hostname, how Server Name Indication picks the certificate, and how each deploy updates one entry in the proxy's routing table. It then looks at what the applications share: memory, disk, the kamal Docker network, the kernel and the proxy itself. It closes with the memory arithmetic to do in advance, the signals that it is time to split, and the order to split in.
What you'll learn
- ✓Routing several applications on one port 443 by hostname and Server Name Indication
- ✓How kamal-proxy keeps a routing table that each deploy updates
- ✓Why service names must differ and why the proxy needs port 80 for certificates
- ✓Planning memory for Puma workers and Postgres before the out-of-memory killer acts
- ✓What containers isolate and what applications on one machine still share
- ✓The signals and the order for moving applications onto separate servers
One server, two applications
A server with a public IP address has one port 443, and two applications both want it. How that conflict is resolved is the whole subject of this course.
Read the full lesson free →Curriculum
7 lessonsOne server, two applications
Why two applications compete for one port 443, and why routing by hostname with SNI solves it.
How the routing works
How kamal-proxy runs apart from applications, updates routing entries per deploy, and handles certificates on port 80.
The resources they share
Memory as the binding constraint, the out-of-memory killer, CPU sharing, and the choice between Postgres setups.
The isolation you do not get
What containers isolate, what they leave shared, and when two applications should not share one machine.
What a second app here would look like
A concrete look at adding a second application to this server, its database options and its specific risks.
When to stop sharing
The signals that sharing a server has gone too far, what splitting costs, and the order to split.
Putting it together
A recap of the course ideas and a challenge to implement fits_in_memory?, the memory arithmetic for shared servers.