"Error: Invalid Prisma endpoint provided" when using docker-compose to deploy a server alongside prisma server

[n2ctech]

Describe the bug

When using docker-compose to deploy a server (e.g. Apollo Server or graphql yoga) talking to prisma server, it is not possible to reference the prisma server container by its name because of string validation in prisma-binding module : /app/node_modules/prisma-binding/dist/Prisma.js:44 throw new Error("Invalid Prisma endpoint provided: " + endpoint);

To Reproduce

Consider a docker-compose file:

version: "3"
services:
  prisma:
    image: prismagraphql/prisma:1.19
    restart: always
    ports:
      - 4466:4466
    environment:
      PRISMA_CONFIG: |
        port: 4466
        managementApiSecret: xxxxx
        databases:
          default:
            connector: postgres
            host: postgres
            port: 5432
            user: postgres
            password: postgres
            migrations: true
  postgres:
    image: postgres:10.5
    restart: always
    environment:
      POSTGRES_USER: postgres
      POSTGRES_PASSWORD: postgres
    volumes:
      - postgres:/var/lib/postgresql/data
  server:
    build:
      dockerfile: Dockerfile.dev
      context: .
    volumes:
      - /app/node_modules
      - ./:/app
    environment:
      - PRISMA_MANAGEMENT_API_SECRET=xxxxx
      - APP_SECRET=xxxxxx
      - PRISMA_URL=prisma
      - PRISMA_SECRET=xxxxx
volumes:
  postgres:

the graphql server running in the server container will try initialise the prisma binding like so :

module.exports = {
  prismaBinding: new Prisma({
    typeDefs,
    endpoint: process.env.PRISMA_URL,
    secret: process.env.PRISMA_SECRET
  })
};

When running docker-compose up this will throw an error when the server image is built and starts running : Error: Invalid Prisma endpoint provided: prisma This is because it expects the prisma endpoint to start with http://

When deploying to a kubernetes cluster I am experiencing the same issue because the ClusterIP service name for my prisma server pod doesn't start with "http://". Hence the server pod can't talk to prisma server inside the cluster because it crashes with the same error.

Expected behavior I should be able to take advantage of docker-compose networking tooling with prisma (a container can reference another container's IP address using its name). Referencing the prisma server from another service by its name (e.g. "prisma") should work. Idem in k8s when referencing prisma by its service name from another pod.

[maticzav]

Hey @n2ctech 👋,

Have you tried using localhost:4466? Also, why was this issue moved here in the first place? cc @divyenduz

[divyenduz]

@maticzav : Because AFAIK, the issue is in prisma-binding string validation.

[maticzav]

@divyenduz I suggest we change the error to a warning. Do you think this could brake anything?

[divyenduz]

@maticzav : Most likely no, but if someone relies on this logic for their URL validation, then their application might break. In any case, we should clearly mention this in release notes should we decide to do it.

[ghost]

Thanks to inlightmedia for his answer on prisma.io

In the end I found that you have to add http:// to the beginning of the service name like so: “http://prisma:4466”

This worked for me