What is common database design mistakes?

Solid design begins with understanding access patterns. Model the entities, then shape tables and indexes around the queries the application will actually run. This guide covers common database design mistakes end to end — core concepts, best practices, concrete data, and a step-by-step approach you can apply right away.

When should I add a read replica?

Add a read replica when your workload is read-heavy and a single primary is saturated on CPU or I/O, but writes still fit on one node. Replicas offload read traffic and improve availability. They are simpler than sharding and solve most scaling needs. Be aware of replication lag, which makes replicas slightly behind the primary.

What is the difference between normalization and denormalization?

Normalization splits data into related tables to remove redundancy and prevent update anomalies, keeping each fact in one place. Denormalization deliberately duplicates data to reduce joins and speed reads. Normalize first for correctness, then denormalize selectively where profiling proves join cost is a real bottleneck — or use caching and materialized views instead.

Should I shard my database to handle more traffic?

Only as a last resort. Sharding scales writes across nodes but complicates joins, transactions, and operations dramatically. First exhaust vertical scaling, read replicas, caching, and query optimization — these solve most scaling problems. Shard only when a single primary genuinely cannot keep up with write volume, and choose your shard key very carefully.

What is connection pooling and do I need it?

Connection pooling reuses a set of open database connections instead of opening a new one per request, avoiding expensive setup overhead and connection exhaustion. Almost any application serving concurrent traffic needs it. For PostgreSQL specifically, an external pooler like PgBouncer is often essential because each connection consumes a server-side process.

Common Database Design Mistakes

This is a practical, up-to-date guide to Common Database Design Mistakes — what it is, why it matters in 2026, and how to apply it in real projects. It is written for developers and founders who want clear answers and proven best practices, not filler.

Whether you're just starting out or leveling up, treat this as a working reference you can return to. Every section is built to be skimmed, applied, and shared.

How Do You Choose Between PostgreSQL And MongoDB?

Both are excellent, mature, and widely deployed — the choice hinges on data shape and consistency needs. PostgreSQL is a relational engine with rich SQL, strong ACID guarantees, and powerful features like JSONB, full-text search, and window functions. MongoDB is a document store offering flexible schemas and straightforward horizontal scaling via sharding.

Favor PostgreSQL when:

Data is highly relational with many joins
Transactions and strict consistency are critical
You need complex analytical queries

Favor MongoDB when:

Documents are self-contained and schema evolves rapidly
You need easy horizontal scale-out
The access pattern is mostly key or document lookups

Notably, PostgreSQL's JSONB narrows the gap, handling many document workloads while retaining relational strengths. Many modern stacks use both for different services.

What Are The Core Principles Of Good Database Design?

Solid design begins with understanding access patterns. Model the entities, then shape tables and indexes around the queries the application will actually run. A schema optimized for writes looks different from one optimized for analytical reads.

Durable principles that apply across engines:

Use appropriate, constrained data types — they save space and catch errors early
Enforce integrity with primary keys, foreign keys, and NOT NULL/CHECK constraints
Choose stable primary keys; surrogate keys avoid mutable natural-key problems
Name consistently and document the schema
Plan for evolution with versioned, reversible migrations

Let the database enforce invariants it can guarantee. Application code is easy to bypass; constraints in the schema protect data regardless of which client writes to it.

What Is Database Sharding And When Is It Worth It?

Sharding horizontally partitions a dataset across multiple database instances, each holding a subset of rows determined by a shard key. It is the primary way to scale writes beyond what a single primary can handle, since each shard absorbs only its portion of the traffic.

The shard key choice is the most consequential decision. A good key distributes load evenly and keeps related data together; a poor one creates hotspots or forces expensive cross-shard queries.

Sharding's costs are real:

Cross-shard joins and transactions become hard or impossible
Rebalancing shards is operationally tricky
Application logic must route queries to the right shard

Because of this complexity, sharding should follow read replicas, caching, and vertical scaling — adopt it only when those genuinely cannot meet demand.

What Is The Real Difference Between SQL And NoSQL?

Relational (SQL) databases store data in tables with fixed schemas and enforce relationships through foreign keys and joins. They excel at strong consistency, complex queries, and transactional integrity via ACID guarantees. NoSQL is an umbrella for non-relational models, each suited to different shapes of data.

The practical distinction is rigidity versus flexibility, and vertical versus horizontal scaling. Common NoSQL families include:

Document (MongoDB): JSON-like documents, flexible schema
Key-value (Redis, DynamoDB): fast lookups by key
Wide-column (Cassandra): massive write throughput
Graph (Neo4j): relationship-heavy traversals

Neither is universally "better." Relational fits transactional systems with stable schemas; NoSQL fits high-volume, evolving, or distributed workloads.

How Do Transactions And ACID Guarantees Work?

A transaction groups operations so they succeed or fail as a unit. ACID describes the guarantees: Atomicity (all-or-nothing), Consistency (constraints stay valid), Isolation (concurrent transactions do not corrupt each other), and Durability (committed data survives crashes).

Isolation is the subtle part. Lower levels allow anomalies for better concurrency:

Read Committed: avoids dirty reads (PostgreSQL default)
Repeatable Read: prevents non-repeatable reads
Serializable: behaves as if transactions ran one at a time, the strictest level

Higher isolation reduces concurrency anomalies but increases locking and abort rates. Choose the lowest level that keeps your data correct. Many NoSQL systems relax ACID to BASE semantics, offering eventual consistency in exchange for availability and scale.

Why Does Database Normalization Matter?

Normalization organizes tables to eliminate redundant data and the update, insert, and delete anomalies redundancy causes. The first three normal forms cover most practical needs: atomic columns (1NF), full dependency on the primary key (2NF), and no transitive dependencies (3NF).

Normalized schemas keep data consistent because each fact lives in exactly one place. The cost is more joins at read time. Denormalization deliberately reintroduces redundancy to speed reads, trading storage and write complexity for query performance.

A pragmatic approach: normalize first for correctness, then denormalize selectively where profiling shows join cost is a real bottleneck. Materialized views and caching often achieve the same read speedup without sacrificing the canonical normalized source of truth.