route/doc/theory.md

62 lines
1.8 KiB
Markdown
Raw Permalink Normal View History

# Theory of route
## Why does this project exist?
HTTP load balancing is harder than it needs to be. Existing projects assume
there is a direct line-of-fire from the load balancer to the underlying services.
This is NOT always true. Containerization has lead to a lot of hacks such as
overlay networks to offer this direct line-of-fire so that the load balancer
can make a direct TCP connection to the underlying services.
Route exists to invert this problem.
In most networks, it is true that it is a lot easier to guarantee the backends
have a direct line-of-fire to the load balancer, not the other way around.
Why not just make a connection to the load balancer and re-use it for HTTP
traffic?
Route allows you to simplify your network architecture by removing the
requirement that all backend services need a direct TCP line-of-fire to the
load balancer. Route allows you to host HTTP services behind NAT. Route
allows you to throw out overlay networks and return to a simpler cluster layout.
## How does route work?
Route has several basic nouns, they are:
- route
- backend
- balancer
- agent
They are inter-related like this:
```graphviz
digraph G {
balancer -> agent
agent -> balancer
agent -> backend
backend -> agent
user -> balancer
balancer -> user
}
```
### route
A route is a combination of HTTP domain and path (currently unsupported, so
assume `/`). An example route would be `cloudsdale.cf/`. This maps to the
location users get.
### backend
A backend is the underlying service being proxied to by route. An example backend
is Apache, Nginx or anything else that serves compliant HTTP/1.1 traffic.
### balancer
A balancer is an instance of `cmd/routed` running on some server somewhere.
Balancers accept both agent and HTTP connections, proxying requests from the HTTP
connections to agent connections and back.