Multiple websites on one machine

[Carol_Ina_Reaper]

Hi guys

I have a question about having “having multiple websites on one machine” (putting the topic very simply so I don’t inadvertently make confusion in my intent).

To provide some context: I’ve been happily working with Elixir/Phoenix for the past 4-5 years, doing some small jobs for my friends. I usually am on a very tight budget (my friends are cheapskates ), so I try to do everything on one very low-spec VPS (one per friend).

Sometimes it so happens that my friends want multiple public-facing sites. When I tried to run two separate Phoenix projects on the same machine, one appeared to be running out of memory, while the other hogged all the memory for itself.

I have solved this issue previously by making multiple Endpoints in one project on different ports and ran it behind Caddy.

That worked well, but now I am doing a little bit more complicated project, which requires 3(+) domains, different apps and a very low payload (per-app assets, layouts etc.). I’ve tried to tackle this, but honestly the project directory now appears to be a giant convoluted mess now, into which I detest delving back into.

Questions from a dilettante to pros like you:

1. Is there a way to run multiple projects on one ((low-spec)) machine without running out of memory, i.e. without the beam.smp processes fighting a last man standing type gladiator battle every time I try to start them? Honestly that would be the best solution, because it feels like I am fighting Phoenix at this point.
2. If not, is there a good way of structuring my project for separate web apps, so that its assets, generated assets (like .js, .css files) and Elixir code runs neatly in one node? I read about generating an --umbrella project, but that didn’t seem to get me any further.

Thanks

C.

[derek-zhou]

Running multiple endpoints in the same release is easy:

Where do you host **small** or even tiny elixir projects?

If those small projects are all written by you, and you can ensure there is not library version conflict, you can run them all in the same Erlang release. There is main proxy, but something as simple as this will work, with liveview and all that. defmodule MySuperProject.Plug do def init(options) do options end def call(%Plug.Conn{host: "app1.host.domain"} = conn, opts) do App1Web.Endpoint.call(conn, opts) end def call(%Plug.Conn{host: "app2.host.domain"} = conn, opts) do …

Your problem of running out of memory is something specific to your code or environment, not intrinsic to Elixir/Phoenix.

[arcanemachine]

Agreed. I have 2 small Elixir/Phoenix projects running on a Debian server, and the whole server uses less then 1GB of RAM (And I think that includes 2 Dockerized Postgres instances…).

Also, I would stay away from umbrella projects as a code organization tool. If you don’t know if you need an umbrella, then you don’t need one IMO. I find that they are typically more trouble than they are worth.

I would say you should look into using Erlang/Elixir-based profiling tools on your services to find your bottlenecks.

You should also share some more info about your setup. Are you using vanilla deployments, Docker, etc.?