Utilizing zod to create a rich, fully-typed MongoDB document management system. With static analysis of the document graph, de-normalization can become a treat rather than a burden!

• Identify and define constraints of the documents at compile time
• Utilize and expand on zod to automatically create type safety
• Build static analysis tools on the document models
• Create embedding system that allows for easy de-normalized modelling
• Automatically apply de-normalized updates in a transaction

Definitions in zongo are defined by a name and a schema. Let's create a simple example for a user definition with a name and age.

// user.ts
import { z } from "zod";
import { zg } from "zongo";

export const userDefinition = zg.createDefinition(
  "User",
  z.strictObject({
    _id: zg.schema.objectId(),
    name: z.string(),
    age: z.number(),
    pets: z.array(z.string()),
  })
);

Now that you have a simple definition you can add it to the database by chaining a createDatabase call to a addDefinition call. The resultant zdb variable will contain the fully typed database.

// zdb.ts
import {userDefinition} from "./user";
import {zg} from "zongo";
import {MongoClient} from "mongodb";

const client = new MongoClient(...);
const db = client.db("main");

export const zdb = zg.createDatabase(client, db)
  .addDefinition(userDefinition);

After creating your zdb it provides many helper functions for document manipulation. One of the goals of this library is to stay "out of the way" of the mongodb native driver and only add typings and helpers where necessary. Creating a document is one of them.

// index.ts
import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const user = await zdb.create("User", {
    _id: new ObjectId(),
    name: "Daniel",
    age: 22,
    pets: ["George", "Rascal"],
  });
}

void main();

The resultant type user will be fully typed based on the zod schema defined in the userDefinition file, utilizing z.output<typeof schema> internally. The value passed as the second argument to create is also strictly typed with z.input<typeof schema> internally.

As mentioned previously, zongo strives to stay out of the way from mongodb as much as possible. Given that, zdb exposes a hydrate method that utilizes a callback with a native mongodb collection.

import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const _id = new ObjectId("...");
  const user = await zdb.hydrate("User", (collection) => {
    return collection.findOne({ _id });
  });
}

void main();

There are two ways to update a document using the update helper. The first option is to pass an object, where the first-level keys will be overridden by whatever is passed. For array manipulation or multi-level merging, option two utilizes a callback with a mutable document.

import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const _id = new ObjectId("...");

  // Option 1
  await zdb.update("User", _id, {
    name: "Danielle",
  });

  // Option 2
  await zdb.update("Update", _id, (doc) => {
    // The doc is accepted to be mutable, no need for immutable updates
    doc.name = "Danielle";
    // Arrays and objects can be directly mutated!
    doc.pets.push("Willow");
    return doc;
  });
}

void main();

We are currently storing pets as an array of strings. This is great, but pets might be better stored in their own collection. Let's create a new collection for the pets and added to our zongo database.

// pet.ts
import { z } from "zod";
import { zg } from "zongo";

export const petDefinition = zg.createDefinition(
  "Pet",
  z.strictObject({
    _id: zg.schema.objectId(),
    name: z.string(),
    type: z.enum(["cat", "horse"]),
    favoriteTreats: z.array(z.string()),
  })
);

// zdb.ts
import {userDefinition} from "./user";
import {petDefinition} from "./pet";
import {zg} from "zongo";
import {MongoClient} from "mongodb";

const client = new MongoClient(...);
const db = client.db("main");

export const zdb = zg.createDatabase(client, db)
  .addDefinition(userDefinition)
  .addDefinition(petDefinition)

The user definition needs to be updated as well. This will utilize the zg.schema.document helpers. There are a few options when embedding a document.

• ref will only store the _id of the referenced document
• partial will only store specific keys from the referenced document
• full will embed the entire document

Let's update the userDefinition to store a partial of the petDefinition, only keeping the name and type.

// user.ts
import { z } from "zod";
import { zg } from "zongo";
import { petDefinition } from "./pet";

export const userDefinition = zg.createDefinition(
  "User",
  z.strictObject({
    _id: zg.schema.objectId(),
    name: z.string(),
    age: z.number(),
    pets: z.array(
      z.schema.document.partial(petDefinition, {
        name: true,
        type: true,
      })
    ),
  })
);

The return value of a ref, partial, and full reference is a DocumentReference, not just the raw values. Let's see how that is utilized. First let's redo our main to create 3 documents.

// index.ts
import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const george = await zdb.create("Pet", {
    _id: new ObjectId(),
    name: "George",
    type: "cat",
    favoriteTreats: ["cardboard"],
  });
  const rascal = await adb.create("Pet", {
    _id: new ObjectId(),
    name: "Rascal",
    type: "cat",
    favoriteTreats: ["Chipotle cheese"],
  });
  const user = await zdb.create("User", {
    _id: new ObjectId(),
    name: "Daniel",
    age: 22,
    pets: [george, rascal],
  });
}

void main();

Next, when accessing the document, the references can be traversed with the DocumentReference class.

// index.ts
import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const _id = new ObjectId("...");
  const user = await zdb.hydrate("User", (c) => c.findOne({ _id }));
  for (const pet of user.pets) {
    // Existing data can be accessed synchronously as it is
    // directly embedded with every user
    const { name, type } = pet.getExisting();
    console.log(`${user.name}'s pet ${type} named ${name}`);

    // If more data is needed, `resolve()` will get the
    // full document from the collection
    const { favoriteTreats } = await pet.resolve();
    console.log(`Favorite treats:`);
    for (const treat of favoriteTreats) {
      console.log(`  - ${treat}`);
    }
  }
}

void main();

When references exist within the definitions, zongo handles the most time consuming part about de-normalized writes, updating other documents. For example, an update to a pet's name would require a subsequent update to the user's with that pet in their pets array. This can become tedious and is prone to errors.

With zongo, your schema is statically analyzed and can perform the subsequent updates automatically and efficiently. There are no restrictions on the way a document is nested within your schema. Nested arrays and objects will both update properly.

For example, updating a pet using zdb.update will automatically update all references to it.

import { zdb } from "./zdb";
import { ObjectId } from "mongodb";

async function main() {
  const _userId = new ObjectId("...");
  const _petId = new ObjectId("...");

  await zdb.update("Pet", _petId, {
    name: "Georgie",
  });

  const user = await zdb.hydrate("Pet", (c) => c.findOne({ _id: _petId }));
  const { name } = user.pet.find((p) => p.id === _petId)!.getExisting();
  asset(name === "Georgie");
}

void main();

Note how in the code above, two documents were updated during the zdb.update() step. Not only was the "George" pet document updated, but the "Daniel" user received an update too. The name field was within the pet partial on the user definition, so it required an update or the cat would be George on the user document and Georgie on the pet document. This inconsistency is completely avoided by utilizing the zongo update method. 🎉

Everything in this library will be based off the zg export. This can be imported in two ways

1. import zg from "zongo";
2. import {zg} from "zongo";

Both of these methods are supported and valid, but intellisense works better with strictly defined names so the second option is there for autocomplete. Option 1 has no requirement to be named zg, though I highly recommend it 😄

The base zg import contains all of the "create" functions. These will be used to instantiate something new, like a database, collection definition, or partial definition.

import {zg} from "zongo";

zg.createDefinition(...);
zg.createPartial(...);
zg.createDatabase(...);

In the zg.schema module, all functions that will be utilized within the context of a zod schema can be found.

Drilling down further, there are two submodules schema and types.

Simply an alias to z.instanceOf(mongo.ObjectId). It is used quite a lot in document creation, to it is here simply as an alias helper. There is no internal significance of this function and if you wish to use z.instanceOf(mongo.ObjectId) instead, go ahead.

import { zg } from "zongo";
import { z } from "zod";

const schema = z.strictObject({
  _id: zg.schema.objectId(),
});

zg.schema.partial<PD extends zg.types.PartialDefinition<any>>(partial: PD);

Instantiates a partial into a zod schema.

import { zg } from "zongo";
import { z } from "zod";

const AuditEntry = z.createPartial(
  "AuditEntry",
  z.strictObject({
    action: z.string(),
    timestamp: z.date(),
  })
);

const schema = z.strictObject({
  auditLog: z.array(zg.partial(AuditEntry)),
});