Schema Validation and Transactions

Introduction

MongoDB collections are schema-flexible, but production apps still need rules: required fields, type checks, and multi-document ACID transactions for transfers and inventory. This chapter adds collection validators with JSON Schema, runs transactions in mongosh (replica set required), and briefly covers change streams for reactive pipelines.

Prerequisites

BSON and Document Basics
Insert and Import Export
Aggregation Pipeline Basics
MySQL transaction concepts — Transactions basics

Schema Validation with JSON Schema

Apply rules at insert/update time:

javascript

Valid insert:

javascript

db.products.insertOne({
  name: "USB-C Hub",
  sku: "HUB-01",
  price: NumberDecimal("49.99"),
  category: "accessory",
  inStock: true
})

Invalid insert (missing sku):

javascript

db.products.insertOne({
  name: "Bad Product",
  price: NumberDecimal("9.99"),
  category: "book"
})
// MongoServerError: Document failed validation

Add validator to existing collection

javascript

Option	Meaning
`validationLevel: strict`	Validate inserts and updates
`validationLevel: moderate`	Validate inserts + updates to valid docs only
`validationAction: error`	Reject invalid writes
`validationAction: warn`	Log warning, allow write (dev only)

Tip

Validate in App and DB

Use Pydantic / Mongoose in the API for user-friendly errors; use MongoDB validator as a safety net against buggy clients — PyMongo chapter.

Transactions Overview

MongoDB multi-document transactions (4.0+) provide ACID semantics similar to MySQL transactions:

Atomicity — all or nothing
Consistency — constraints hold (including validators)
Isolation — snapshot isolation
Durability — depends on write concern

Requirement: deployment must be a replica set (including single-node replica set for dev). Standalone mongod without replica set cannot run transactions in modern versions.

Single-node replica set for local dev (Docker)

bash

docker run -d --name mongo-rs \
  -p 27017:27017 \
  mongo:7 mongod --replSet rs0 --bind_ip_all
 
docker exec -it mongo-rs mongosh --eval 'rs.initiate()'

Wait until rs.status() shows primary.

Transaction in mongosh

Transfer inventory between two product documents:

javascript

Every operation in the transaction must pass { session }.

Seed stock if testing:

javascript

db.products.insertMany([
  { name: "Hub", sku: "HUB-01", price: NumberDecimal("49.99"), category: "accessory", stock: 10 },
  { name: "Case", sku: "CASE-01", price: NumberDecimal("19.99"), category: "accessory", stock: 0 }
])

Transaction Limits (Know the Bounds)

Limit	Typical value
Duration	60 second default max
Document size	16 MB per doc unchanged
Collections	Cross-collection OK in same DB
Sharded cluster	Supported with restrictions

Keep transactions short—no external HTTP calls inside.

Read/Write Concern with Transactions

javascript

session.startTransaction({
  readConcern: { level: "snapshot" },
  writeConcern: { w: "majority" }
})

Production replica sets use w: "majority" for durability — Replica sets.

Change Streams (Optional)

Watch collection for inserts/updates/deletes:

javascript

const changeStream = db.posts.watch(
  [{ $match: { operationType: { $in: ["insert", "update"] } } }]
)
 
changeStream.hasNext()
changeStream.next()

Use cases:

Invalidate Redis cache on write — Redis caching
Sync search index
Audit log consumer

Requires replica set. Resume with resumeAfter token after disconnect.

Compare Redis Pub/Sub — fire-and-forget vs durable oplog tail — Redis Pub/Sub.

Validation vs Transactions Decision

Need	Tool
Field types, required keys	Schema validator
Two collections update together	Transaction
Idempotent retry	Unique index + upsert
Long-running workflow	Saga pattern in app, not long transaction

Post-Chapter Checklist

Created collection with $jsonSchema validator
Saw validation error on bad insert
Initialized replica set for dev (if testing transactions)
Ran commit/abort transaction with session
You know change streams require replica set

FAQ

Transactions on Atlas free tier?

Yes—Atlas runs replica sets; transactions work on M0 within limits.

Validator on existing bad data?

moderate level skips invalid legacy docs until updated.

Same as SQL CHECK constraint?

Similar intent; JSON Schema is richer for nested documents.

PyMongo transaction API?

with client.start_session() as session: + session.with_transaction(callback) — chapter 19.

Change stream vs polling?

Change streams push from oplog; polling wastes queries and adds lag.