9 months ago
I'm experimenting with Railway on my side project for some time. While it's great for quick experiments, I noticed few inconveniences while the project scales up. I'd like to elaborate on them, and maybe some would land as improvement ideas.
Growing list of shared variables
It's sometimes hard to decide if an environment variable should be put in the global project settings or in the service. If I change my mind and rename a variable or change its place, it's not easy to juggle them around; there's lots of clicking. And I end up with at least one failed deployment to make everything run again. Regardless if I first change the variables and then the code reference or vice versa.
I would love to see an option to provide the env vars (or at least the references like VAR_IN_SERVICE_2="${{service1.RAILWAY_PRIVATE_DOMAIN}}:${{service1.PORT}}" - meaning, not containing anything crucial, besides references to other vars) in the railway.json/toml files.
Local development struggles
It's not easy to link the services in the monorepo with their Railway siblings. So I was so far unable to run something like railway run turbo run dev .
I need to run link each service separately and then run each of them separately from their folder in a separate terminal session.
This could also be somehow "handled" by the "config-as-code" feature. Each part of my project has its own railway.json file already. So if the projectId and serviceId could be placed there, maybe the CLI could use it to inject the needed variables.
Single build, multiple deploys
Most of my shared monorepo code is Typescript-based. So I need to call the install and build from the root folder and then am building Docker containers in the service folders. For now... each Dockerfile has its own identical build part. It would greatly improve the time and usage of resources if I could build once and then tell each service to build their own container from their output.
14 Replies
9 months ago
Just remembered one more:
Service dependency
What if I need one service to be up and running for the other one to finish its deploy? I also found out that I can't use private networking during the deployment :(
8 months ago
What if I need one service to be up and running for the other one to finish its deploy?
Rrailway has this https://docs.railway.com/guides/deployment-actions#deployment-dependencies---startup-ordering
8 months ago
I also found out that I can't use private networking during the deployment
What do you need it for, they have a way to do database migrations and seeding https://docs.railway.com/reference/config-as-code#pre-deploy-command
8 months ago
You can have service level variables as well https://docs.railway.com/guides/variables
8 months ago
Why don't you try to use their CLI? It can help with custom scripts https://docs.railway.com/reference/cli-api
sim
What if I need one service to be up and running for the other one to finish its deploy?Rrailway has this https://docs.railway.com/guides/deployment-actions#deployment-dependencies---startup-ordering
8 months ago
I missed that in docs. Will have a look.
sim
I also found out that I can't use private networking during the deploymentWhat do you need it for, they have a way to do database migrations and seeding https://docs.railway.com/reference/config-as-code#pre-deploy-command
8 months ago
Simple example: a NextJS/NuxtJS/SvelteKit (or similar) app connecting to CMS during build time to pre-render (some or all) content pages. So far I can only do this using service's public URL.
sim
You can have service level variables as well https://docs.railway.com/guides/variables
8 months ago
I'm mentioning in my original post! I'm using vars on project level and service level.
The problem is: relations between the services (at least during the period of project's evolution)
With this simple example from the post above:
1. I have service1 with var PORT=1234
2. I have service2 connecting to the service1, so I define VAR_IN_SERVICE_2="${{service1.RAILWAY_PRIVATE_DOMAIN}}:${{service1.PORT}}" in it
I now do such a simple thing like... rename
PORTtoPORT_EXT, orservice1toservice_1
There's loads of clicking through the dashboard to link that (especially when you have more than 2 services and more than 1-2 relations), and additionally, I also need to change code (+ railway config-as-code) references.
sim
Why don't you try to use their CLI? It can help with custom scripts https://docs.railway.com/reference/cli-api
8 months ago
Yeah! That's one thing I'm considering - to move the build and deployment to the CI/CD (GitHub Actions, GitLab Actions, Circle CI or whatever).
I could definitely simplify the build to one global call, but would still have the hustle of linking services with their folders.
tvardy
I missed that in docs. Will have a look.
8 months ago
Sweet!
tvardy
Simple example: a NextJS/NuxtJS/SvelteKit (or similar) app connecting to CMS during build time to pre-render (some or all) content pages. So far I can only do this using service's public URL.
8 months ago
Yes so you can use the public url and then move to private
sim
Yes so you can use the public url and then move to private
8 months ago
Why would I need to make the service public just for this step?! 
This is the inconvenience I'm reporting here
tvardy
Why would I need to make the service public just for this step?! This is the inconvenience I'm reporting here
8 months ago
Yes, not ideal, I was just suggesting
tvardy
Yeah! That's one thing I'm considering - to move the build and deployment to the CI/CD (GitHub Actions, GitLab Actions, Circle CI or whatever).I could definitely simplify the build to one global call, but would still have the hustle of linking services with their folders.
8 months ago
What about config as code with their railway.json for every folder and set the project and service IDs? Yes again issue of when the ID changes but less hustle?