Basically, if we have data.json like this:

[
  { "id": "1", "name": "Dan" },
  { "id": "2", "name": "Lee" },
  { "id": "3", "name": "Nick" }
]

... and go run main.go, we can query records:

$ curl -g 'http://localhost:8080/graphql?query={user(name:"Dan"){id}}'
{"data":{"user":{"id":"1"}}}

... now let's give Dan a surname:

[
  { "id": "1", "name": "Dan", "surname": "Jones" },
  { "id": "2", "name": "Lee" },
  { "id": "3", "name": "Nick" }
]

... and kick the server:

kill -SIGUSR1 52114

And ask for Dan's surname:

$ curl -g 'http://localhost:8080/graphql?query={user(name:"Dan"){id,surname}}'
{"data":{"user":{"id":"1","surname":"Jones"}}}

... or ask Jones's name and ID:

$ curl -g 'http://localhost:8080/graphql?query={user(surname:"Jones"){id,name}}'
{"data":{"user":{"id":"1","name":"Dan"}}}

If you look at main.go, the file is not field-aware. That is, all it knows is how to work with []map[string]string type.

With this, we are not that far from exposing dynamic fields and filters which fully depend on what we have stored, all without changing our tooling.