8.04 · Deep Dive: MongoDB for Document Analytics

Level: Intermediate Time to read: 15 min Pre-reading: 08 · Tools Ecosystem After reading: You'll understand MongoDB's strengths for semi-structured data and when it's suitable for analytics.

MongoDB: Document Database

MongoDB is a document-oriented NoSQL database that stores data as JSON-like documents. It's optimized for flexible schemas and operational workloads, not analytics.

Architecture

Traditional RDBMS:        MongoDB:
┌─────────────────┐       ┌──────────────────────────┐
│ Customers Table │       │ customers Collection     │
├─────────────────┤       ├──────────────────────────┤
│ id  │ name │ age│       │ {_id: 1, name: "John",   │
├─────┼──────┼────┤       │  age: 30,                │
│ 1   │ John │ 30 │       │  orders: [{id: 101, ...}]│
│ 2   │ Jane │ 28 │       │ }                        │
└─────┴──────┴────┘       │ {_id: 2, name: "Jane",   │
                          │  age: 28, ...}          │
                          └──────────────────────────┘

Strengths

✅ Flexible schema - Add fields without migrations ✅ Nested data - Embed related documents ✅ Developer-friendly - JSON-like format ✅ Scalable - Built-in horizontal scaling (sharding) ✅ Fast writes - Optimized for insert/update operations

Weaknesses for Analytics

❌ Complex aggregations - Denormalized data is hard to analyze ❌ No JOINs - Relationships are embedded, limiting query flexibility ❌ High storage - Denormalization increases data size ❌ Limited SQL - MongoDB Query Language is less powerful than SQL ❌ No columnar compression - Row-like storage, not optimized for analytics scans

When to Use MongoDB

Good Fit

• ✅ Product catalogs (varied attributes per product)
• ✅ User profiles (flexible, evolving schemas)
• ✅ Content management systems
• ✅ Real-time operational analytics
• ✅ IoT sensor data (document per sensor reading)

Poor Fit

• ❌ Complex analytics queries (JOINs across collections)
• ❌ Ad-hoc business intelligence
• ❌ Time-series data (use Cassandra or TimescaleDB)
• ❌ Large-scale aggregations (millions of groups)
• ❌ Financial/audit data (need strict schemas)

MongoDB for Analytics: The Pattern

Instead of building analytics directly in MongoDB, follow this pattern:

MongoDB (Operational)
    ↓ Export/CDC
Parquet Files (Cloud Storage)
    ↓ Load
BigQuery / Snowflake (DWH)
    ↓ Transform (dbt)
Gold Layer
    ↓
BI Tools / Dashboards

MongoDB Aggregation Pipeline (Limited Analytics)

// MongoDB aggregation pipeline (limited capability)
db.customers.aggregate([
  {
    $match: {
      created_at: { $gte: new Date('2024-01-01') }
    }
  },
  {
    $group: {
      _id: "$city",
      num_customers: { $sum: 1 },
      avg_lifetime_value: { $avg: "$lifetime_value" }
    }
  },
  {
    $sort: { avg_lifetime_value: -1 }
  },
  {
    $limit: 10
  }
]);

// Equivalent SQL (more powerful):
// SELECT city,
//        COUNT(*) as num_customers,
//        AVG(lifetime_value) as avg_lifetime_value
// FROM customers
// WHERE created_at >= '2024-01-01'
// GROUP BY city
// ORDER BY avg_lifetime_value DESC
// LIMIT 10;

Note: Aggregation pipelines work but are less flexible and slower than SQL queries on data warehouses.

MongoDB Atlas for Analytics

MongoDB Atlas = Managed MongoDB cloud + BI connectors

MongoDB Atlas (cloud)
    ↓
Option 1: Aggregation pipelines (in-DB analytics, limited)
    OR
Option 2: BI Connectors → Tableau, Power BI (using aggregation pipeline)
    OR
Option 3: Export to Data Warehouse → BigQuery/Snowflake

Example: Export to BigQuery

// Use MongoDB's native connectors or tools like Stitch/Fivetran
// to periodically export collections to BigQuery

// MongoDB collection
db.orders.find({})

// Becomes BigQuery table
SELECT * FROM `project.dataset.orders`

// Now run analytics
SELECT
  customer_id,
  COUNT(*) as num_orders,
  SUM(amount) as total_spent
FROM orders
GROUP BY customer_id
ORDER BY total_spent DESC;

Comparison: MongoDB vs. Data Warehouses

Key Takeaways

1. MongoDB is operational, not analytical - Designed for OLTP, not OLAP
2. Export to data warehouse for serious analytics (BigQuery, Snowflake)
3. Aggregation pipelines have limited functionality vs. SQL
4. Use MongoDB for product data, not analytical data
5. Combine with BigQuery/Snowflake for complete analytics stack
6. For analytics, use cloud DWH as single source of truth