Auto index

autoIndex creates the indexes a model declares on its first operation in each database, as Mongoose's autoIndex does. It is for development, tests and platforms without a step that runs sync() before the app takes requests, such as a serverless one. It comes from mongozod/plugins/auto-index.

import { mongozod } from 'mongozod'
import { autoIndex } from 'mongozod/plugins/auto-index'
import * as z from 'zod'

const usageSchema = z.object({

  organization: z.string(),

  count: z.number()
    .int()
})

const UsagesModel = mongozod({

  collection: 'usages',

  db,

  schema: usageSchema,

  indexes: (index) => [
    index({ organization: 1 }, { unique: true })
  ],

  plugins: [autoIndex()]
})

// The first operation creates the unique index, then upserts.
await UsagesModel.updateOne({ organization: 'acme' }, { $inc: { count: 1 } }, { upsert: true })

What it doesCopy Anchor Link

• Once per database: the first operations of a model in a database wait for one createIndexes, which does nothing when the indexes exist. Operations that start at the same time share it. A model whose db is a function, such as one database per tenant, builds them in each database it reads.
• Every operation: reads and writes, cursors included, so a unique index exists before the first write relies on it.
• Indexes only: it does not apply the $jsonSchema, which needs collMod, an action of dbAdmin that the application's user does not have. sync() applies it.
• When it fails, such as a unique index over repeated values, the operation that started the build throws the server's error, and the next operation tries again. A unique index that cannot be built repeats the build on each operation until the data is fixed.

When to use itCopy Anchor Link

sync() from a deploy step is still the way for production: an index build on a large collection delays the operations that wait for it, and a failure shows up in the deploy and not in a request. autoIndex fits where there is no such step:

• Tests: a test that relies on a unique index gets it without calling sync() in its setup.
• Development: a new index is there on the next run.
• Serverless platforms: each new instance sends one createIndexes, which costs little when the indexes exist.

See When to run it for the other ways.

ReferenceCopy Anchor Link

autoIndex() takes no arguments.

Common mistakesCopy Anchor Link

• Expecting it to drop or change an index: like sync(), it never drops one, and a changed index fails with the code 85 or 86 on each operation. Drop the old one with model.collection.dropIndex().
• Expecting the $jsonSchema: run sync() for it, with a user that has dbAdmin.