Skip to content

vwkd/denokv-graphql

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

README

GraphQL bindings for Deno KV

Getting started

import { buildSchema } from "https://deno.land/x/denokv_graphql@$MODULE_VERSION/mod.ts";

const schema = buildSchema(db, schemaSource);

Check out the examples.

Features

  • relational schema
  • atomic consistency
  • strict input and output validation

Concepts

  • tables and rows, a column can reference rows of another table
  • the row id is a string, provided by user, e.g. incrementing integer, UUID, etc.
  • a query assembles result object(s), resolves references recursively, checks that there are no concurrent mutations
  • a many query paginates over results and references in lexicographic order of row id strings, implements Relay pagination spec
  • a mutation is a transaction, can mutate multiple tables with atomic consistency
  • invalid schema input throws an InvalidSchema error
  • invalid query input throws an InvalidInput error
  • invalid database output throws a DatabaseCorruption error

Schema

  • note: if a type isn't explicitly mentioned as "non-null" or "nullable", it can be either non-null or nullable
  • note: the mutation directives and the PageInfo type are built-in

General

  • a table is an object type, a column is a field
  • a column with a value is a field of scalar type
  • a table type must have an 'id' field of non-null ID type and at least one other field
type Book {
  id: ID!,
  title: String
}
  • a reference to a row of another table is a field of the other table's type
type Book {
  id: ID!,
  title: String
  author: Author
}

type Author {
  id: ID!,
  name: String
}
  • references to multiple rows of another table is a field of a non-null connection type with arguments 'first' of Int type, 'after' of ID type, 'last' of Int type, and 'before' of ID type
  • a connection is an object type whose name ends in "Connection" with a field 'pageInfo' of non-null PageInfo type and a field 'edges' of non-null list of edge type
  • an edge type is a object type whose name ends in "Edge" with a field 'cursor' of non-null ID type and a field 'node' of non-null table type
  • note: the connection can only be queried using either 'first' with optional 'after' or 'last' with optional 'before', not both at the same time, nor neither!
type Book {
  id: ID!,
  title: String
  authors(first: Int, after: ID, last: Int, before: ID): AuthorConnection!
}

type AuthorConnection {
  edges: [AuthorEdge]!
  pageInfo: PageInfo!
}

type AuthorEdge {
  node: Author!
  cursor: ID!
}

type Author {
  id: ID!,
  name: String
}

Query

  • a query for a single row of a table is a field of a nullable result of the table type with an argument 'id' of non-null ID type
  • a result is an object type with a field 'versionstamp' of non-null String type and a field 'value' of non-null table type
  • note: result is null if row with id doesn't exist!
type Query {
  bookById(id: ID!): BookResult
  # ...
}

type BookResult {
  value: Book!
  versionstamp: String!
}

# ...
  • a query for multiple rows of a table is a field of a connection type similar to above, except the edge type has a field 'node' of non-null result of the table type
type Query {
  books(first: Int, after: ID, last: Int, before: ID): BookConnection!
  # ...
}

type BookConnection {
  edges: [BookEdge]!
  pageInfo: PageInfo!
}

type BookEdge {
  node: BookResult!
  cursor: ID!
}

type BookResult {
  value: Book!
  versionstamp: String!
}

type Book {
  id: ID!,
  title: String
}

Mutation

  • a transaction is a field of a nullable result type with an argument 'data' of non-null transaction input type
  • a result is an object type with a field 'versionstamp' of non-null String type
  • note: result is null if in insert rows already exist, or in delete the versionstamps don't match!
  • a transaction input is an input object type with fields of non-null list of non-null row input type and a directive 'insert' or 'delete' with a table argument of the table name string
type Mutation {
  createTransaction(data: CreateInput!): Result
  # ...
}

input CreateInput {
  createBook: [BookInput!]! @insert(table: "Book")
  # ...
}

type Result {
  versionstamp: String!
}
type Mutation {
  deleteTransaction(data: DeleteInput!): Result
  # ...
}

input DeleteInput {
  deleteBook: [Identifier!]! @delete(table: "Book")
  # ...
}

type Result {
  versionstamp: String!
}
  • an insert row input is an input object with fields matching the table type, except a reference is of ID type and references are of non-null list of ID type
input BookInput {
  id: ID!,
  title: String
  author: ID
}
input BookInput {
  id: ID!,
  title: String
  authors: [ID]!
}
  • an delete row input is an input object with a field 'id' of non-null ID type and a field 'versionstamp' of non-null String type
input Identifier {
  id: ID!,
  versionstamp: String!
}

Internals

  • sets up resolvers for read, write and delete operations to the Deno KV database
  • a value column of a row is stored with the individual key of the table name, row id, and column name
  • a reference column of a row is stored with the individual key of the table name, row id, column name, and row id of referenced table
  • issue that row has potentially different versionstamp for each column, returns versionstamp of 'id' column as versionstamp for whole row, versionstamp of 'id' column must always be newest of any of row's columns
  • beware: must make sure that any update mutation to row's columns also resets 'id' column to new versionstamp!

FAQ

Why GraphQL?

GraphQL is an ergonomic interface for selecting nested data with additional features like aliases, input validation, detailed error messages, introspection, and more thanks to a strict schema. GraphQL may be more popularly known as API of a remote server, but nothing prevents it from being used as local interface of a database. As additional layer between user and database it decreases performance, but it adds meaningfully better ergonomics which might be worth the price.