complete restructuring

[lennyburdette]

This is one of those “I didn’t have time to write you a short letter, so I wrote you a long one” kind of things. I was just trying to get it working but it spiraled out of control. If I had more time before vacation I would break this up in to smaller changes. Instead I'm just going to dump this on you with no expectation that you'll review or merge it. But hopefully it'll be a useful reference for changes you could make before I get back.

• made the subgraphs independent packages

  allows removing packages from the root package.json and simplifies the dockerfiles. i think this is a better solution than forcing the containers to use x86_64 images on ARM devices.

• remove typescript from subgraphs and apollo-server implementation

  no need to transpile. i added tsconfig files to get typechecking in your ide.

• use a prebuilt gateway container image

  able to delete some code and remove some dependencies.

• move implementations to the root

  a step towards making the area for contribution front-and-center in the repo.

• use multiple docker files to reduce repetition

  you can pass -f multiple times and compose will merge the configs. makes the compose configs in each implementation very small.

• use execa for shelling out to docker-compose

  async-await-friendly library.

• use fetch() for checking the gateway health

  using a remotegraphqldatasource doesn't make sense here. almost added apollo client, but it's not used for graphql requests yet!

• add retries for checking the product service health

  federation-jvm takes a moment to start up, so i added a cheap loop.

• build the federation-jvm implementation in the docker image

  the existing docker file assumed that the jar was prebuilt, so it didn't work at all on my machine. there are opportunities to optimize this as it's a pretty large image.

• format most things with prettier

  happens by default when i save files.

Edit

I ended up accidentally upgrading @apollo/gateway and the RemoteGraphQLDataSource#process API changed and wasn't possible to call. I swapped it out with a fetch call — I'm pretty sure that'll be the right decision in the long run as the subgraph contract will always be an HTTP call.