How to install only what's necessary when auto-deploying?
This isn't an issue with Railway. This is a question on how to optimize the install/build/deploy process for the applications I host on Railway.
I have a "shared" monorepo in which I have a
frontend and backend application (as well as others, but my issue can be conveyed with just these two apps). The monorepo uses pnpm as a package manager and turbo (https://turbo.build/repo) for building/caching.
My Github repository is connected to my Railway project, so I get those sweet auto-deploys when pushing to the main branch.
When I push to the main branch, the Railway service connected to the frontend application runs pnpm install, which installs all the dependencies in the repository -- all of those of frontend, backend, and all internal packages. The same happens with the service connected to the backend application.
It's inefficient because the frontend service installs backend-only dependencies (prisma, @nestjs, etc.) and the backend service installs frontend-only dependencies (react, @remix-run, etc.).
The question I have: is there a way to prevent the service from installing dependencies that aren't required for the given application?
I'm investigating turbo prune (https://turbo.build/repo/docs/reference/command-line-reference/prune).
When I run turbo prune frontend locally, it isolates the frontend application and all its dependencies into an out directory. I can then cd out and run pnpm install to install only the dependencies for the frontend app (e.g. react and @remix-run but not prisma and @nestjs). And finally I can run the build and start scripts to get the frontend running without a problem.
My issue comes when trying to use turbo prune when deploying via Railway. When I push to main, Railway immediately runs pnpm install and so I can't do turbo prune frontend && cd out before that pnpm install happens.
(Reaching character limit, see more in thread)7 Replies
Project ID:
67e6b82d-d699-41f6-a415-466ef9889928Project ID:
67e6b82d-d699-41f6-a415-466ef9889928
So then I believe my options are:
Option 1: Nope can't do anything about it
Option 2:
Commit the out directory from turbo run frontend to the Github repo, then configure the Root Directory (https://docs.railway.app/guides/builds#build-command) of the frontend Railway service to use the out directory. This isn't ideal because (1) I'd have to remember to generate the out directory before I push to main, and (2) I commit duplicative deployment-specific code to my Github repo.
Option 3:
I don't know if this is feasible, but I run turbo prune frontend in a Github Action to generate the out directory, and then (somehow, IDK if using the Railway CLI in a Github Action is possible/recommended) have the frontend Railway service use that out directory.
Option 4: You help me lol.
Thanks in advance for any thoughts/help in figuring this out.
Also there's Option 5: Use a custom Dockerfile (https://docs.railway.app/guides/dockerfiles) but that's outside my expertise so that'll be fun to figure out.first off, this is a well put together question, you've articulated all your questions perfectly!
second, option 5 would be your absolute best bet and would allow you to very easily copy only the stuff you need into the image, it also allows you to run any commands you want during build, like the prune commands you mentioned. watch a youtube video or two and you will be an expert in no time!
Haha thank you thank you, I definitely spent like an hour putting this together to make things as easy for all of us, especially you Brody since I figured it'd be you to help me out given your history as this server's MVP π
I eventually figured the custom Dockerfile would be the best option, just needed to confirm haha. Thanks for the advice and all that you do.
i know how to write dockerfiles, but i dont know turbo so i wouldnt be too much help writing you one for this usecase, but id be happy to sanity check anything you come up with!
Oooh thanks again, will lyk if I run into trouble with it π
sounds good!