@nic-jennings/batched-sql-datasource NPM

Batched SQL (Knex) Datasource

SQL DataSource combines the power of Knex with Apollo DataSources along with the benefit of having batching functionality from the DataLoader package.

THIS PACKAGE IS FOR > APOLLO CLIENT 3

For Apollo Server 4 please use: @nic-jennings/sql-datasource

Getting Started

Installation

npm i @nic-jennings/batched-sql-datasource

yarn add @nic-jennings/batched-sql-datasource

Usage

Create DataLoader:

// MyDataSource.js

import { BatchedSQLDataSource } from "@nic-jennings/batched-sql-datasource"

export class MyDataSource extends BatchedSQLDataSource {
  getBar;

  constructor(db) {
    super(db);

    // batching
    this.getBar = this.db.query
        .select("*")
        .from({b: "bar"})
        .batch(async (query, keys) => {
          const result = query.whereIn("b.id", keys);
          return keys.map((x) => result?.filter((y) => y.id === x)
        });
  }
  /* or if you have seperate read and write instances:
  constructor(readDb, writeDb) {
    super(readDb, writeDb);
  }
  */

  // Standard
  getFoo() {
    return this.db.query
      .select("*")
      .from("foo")
      .where({ id: 1 });
  }

  // caching
  getFooCached() {
    return this.db.query
      .select("*")
      .from("foo")
      .where({ id: 1 })
      .cache(10);
  }

}

Use it with Apollo Server:

// index.js

import { MyDataSource } from "./MyDataSource";

const readKnexConfig = {
  client: "pg",
  connection: {
    /* CONNECTION INFO */
  },
};
const writeKnexConfig = {
  client: "pg",
  connection: {
    /* CONNECTION INFO */
  },
};
// We can either create two instances of Knex (Read & Write) or pass a single connection, you can also pass knex instances instead of a configuration object
const db = new MyDataSource(knexConfig, writeKnexConfig);

const server = new ApolloServer({
  typeDefs,
  resolvers,
  cache,
  context,
  dataSources: () => ({ db }),
});

Batching (.batch( key, callback ))

As GraphQL only requests the data we require from the database when it is part of the query, it can lead to what is know as the n+1 problem. This is solved by using DataLoaders.

DataLoaders can help you by collecting all of your queries and combining them in a single one. And what's wonderful about this strategy is that it doesn't matter if your data exists in different DB; you can fetch your data from different sources. The only downside to this strategy is that we have to define a data loader for each model which makes it a bit cumbersome to use.

By making it part of the fluid interface that Knex provides it makes using it much cleaner.

   this.getBar = this.db.query
        .select("*")
        .from({b: "bar"})
        .batch(async (query, keys) => {
          const result = query.whereIn("b.id", keys);
          return keys.map((x) => result?.filter((y) => y.id === x)
        });

Now in our resolver we simple have:

 bar({id}, _, { dataSources }) {
      return dataSources.db.getBar.load(id);
 },

Caching ( .cache( ttl ) )

If you were to make the same query over the course of multiple requests to your server you could also be making needless requests to your server - especially for expensive queries.

SQLDataSource leverages Apollo's caching strategy to save results between requests and makes that available via .cache().

This method accepts one OPTIONAL parameter, ttl that is the number of seconds to retain the data in the cache.

The default value for cache is 5 seconds.

Unless configured, SQLDataSource uses an InMemoryLRUCache like the RESTDataSource.

BatchedSQLDataSource Properties

BatchedSQLDataSource is a Typescript Class that extends Apollo's base DataSource Class and can be extended to make a new DataSource

BatchedSQLDataSource has an initialize method that Apollo will call when a new request is routed.

If no cache is provided in your Apollo server configuration, SQLDataSource falls back to the same InMemoryLRUCache leveraged by RESTDataSource.

context

The context from your Apollo server is available as this.context.

knex

The knex instance is made available as this.db.query and this.db.write.

As the names suggest one should be a read instance and the other should be a write instance. If you do not have seperate instances just pass the same configuration twice.

Debugging

To enable more enhanced logging via knex-tiny-logger, set DEBUG to a truthy value in your environment variables.

V1.0.0: Breaking changings

The API has changed from the V0.0.1 > to V0.2.2.

Previously the batch function accepted a string and function to filter the results. This has been changed to accept a function which allows you pass an object to the .load or .loadMany and do multiple whereIn queries.

E.g:

dataSources.db.getBar.load({ id, value: "bar" });

   this.getBar = this.db.query
        .select("*")
        .from({b: "bar"})
        .batch(async (query, keys) => {
          const result = query.whereIn("b.id", keys.map(k => k.id)).whereIn("b.value", keys.map(k => k.value));
          return keys.map((x) => result?.filter((y) => y.id === x.id && y.value === x.value)
        });