PlanetScale: Serverless Database Platform Built on Vitess

PlanetScale: Serverless Database Platform Built on Vitess

Quick Answer

PlanetScale is a serverless database platform built on Vitess for MySQL and Raft for PostgreSQL. It offers horizontal sharding, non-blocking schema changes via ghost tables, and the SPANN vector indexing algorithm. The database branching workflow enables Git-like collaboration for schema evolution.

What is PlanetScale?

PlanetScale commercializes Vitess, the database scaling technology developed at YouTube to handle massive scale. By abstracting Vitess complexity behind a developer-friendly interface, PlanetScale brings hyperscale database capabilities to every application.

The Vitess Architecture

VTGate: The Intelligent Proxy

• Stateless proxy handling connection pooling
• Routes queries to appropriate shards automatically
• Protects database from connection storms

VTTable: The Sidecar Guardian

• Runs alongside each MySQL instance
• Parses and validates queries before execution
• Protects against toxic queries

VReplication: The Data Movement Engine

• Enables zero-downtime resharding
• Streams WAL for Change Data Capture
• Powers online schema migrations

Key Innovations

Non-Blocking Schema Changes

PlanetScale’s “Ghost Table” mechanism enables zero-downtime DDL:

1. Ghost Table Creation: Create shadow table with new schema
2. Backfill & Sync: Copy data while replicating ongoing changes
3. Atomic Cutover: Swap tables with instant reversion capability

Database Branching

# Create branch for feature development
pscale branch create my-db feature-login

# Apply schema changes
pscale db execute SQL_FILE schema.sql --branch feature-login

# Open deploy request
pscale deploy-request create my-db feature-login

PlanetScale for MySQL

Sharding Model

Keyspaces & VSchema:

• Unsharded: Single database (standard)
• Sharded: Distributed via VSchema sharding keys
• Transparent: Application queries remain unchanged

Foreign Key Support:

• Unsharded DBs: Full FK support (since early 2024)
• Sharded DBs: FK allowed only if shard-scoped (both tables on same shard)

SPANN Vector Algorithm

PlanetScale uses SPANN (Space-Partitioned ANN) for MySQL vector search:

Feature	Description
Architecture	Centroids in memory, vectors on disk (NVMe)
Memory Efficiency	~15% of raw data size
Quantization	bfloat16 (default), 1-bit binary, Product Quantization
Search Type	Hybrid: semantic + SQL filters combined

SELECT product_id, name, price
FROM products
WHERE category = 'electronics' AND price < 500
ORDER BY DISTANCE(embedding, TO_VECTOR('[0.1, 0.5, ...]'))
LIMIT 10;

PlanetScale for PostgreSQL

Raft-Based Architecture

Launched in 2025, PlanetScale Postgres uses:

• Consensus: Raft algorithm (similar to Patroni)
• Availability: 1 Primary + 2 Replicas across 3 AZs
• Performance: Local NVMe storage for low latency
• Extensions: PostGIS, pgvector, pg_duckdb, MotherDuck, UUIDv7

Neki: Future Sharding

PlanetScale is developing Neki, a sharding engine for Postgres:

• Purpose-built for horizontal Postgres scaling
• Similar VTess-like routing principles
• Addresses the main Postgres scalability gap

Extension Support

Extension	Purpose
PostGIS	Geospatial data and queries
pgvector	Vector similarity search
pg_duckdb	OLAP analytics within Postgres
MotherDuck	Cloud data warehouse connection

PlanetScale vs Competitors

PlanetScale vs Neon

Feature	PlanetScale	Neon
Database Engine	MySQL (Vitess) / Postgres (Raft)	PostgreSQL (Split architecture)
Scaling Model	Horizontal sharding	Vertical scaling + branching
Schema Changes	Non-blocking online DDL	Manual SQL migrations
Branching	Git-style deploy requests	Instant metadata branching

PlanetScale vs AWS Aurora

Feature	PlanetScale	AWS Aurora
Underlying Tech	Vitess (open source)	Proprietary storage engine
Pricing	Per-provisioned tier	Per-ACU consumption
Vector Search	SPANN (MySQL native)	Via Aurora PostgreSQL
Sharding	Built-in via Vitess	Requires manual partitioning

PlanetScale vs CockroachDB

Feature	PlanetScale	CockroachDB
SQL Compatibility	MySQL wire protocol	PostgreSQL wire protocol
Sharding	Vitess VSchema	Automatic range-based
Consistency	Configurable per query	Serializable by default
Vector Index	SPANN (disk-optimized)	C-SPANN (distributed)

Developer Experience

Git-for-Data Workflow

PlanetScale revolutionizes database schema management:

# 1. Create branch
pscale branch create my-db feature-branch

# 2. Connect to branch in application
DATABASE_URL="pscale-branch-feature-branch..."

# 3. Make schema changes
pscale db execute SCHEMA.sql --branch feature-branch

# 4. Open deploy request
pscale deploy-request create my-db feature-branch

# 5. Review and merge
pscale deploy-request review my-db deploy-123

Insights Dashboard

AI-powered database observability:

• LLM Query Summaries: Natural language explanation of slow queries
• Index Recommendations: Automatic suggestions for missing indexes
• Performance Metrics: p95/p99 latency, rows read/written

Pricing Structure (2026)

Scaler Pro (ARM64)

Configuration	Monthly Cost	Specifications
PS-5	$5	Single Node (Non-HA)
PS-10 ARM	$30	Cluster (3 nodes, HA)
PS-10 Intel	$39	Cluster (3 nodes, HA)

Additional Costs

• Storage: $1.50/GB over 10GB included
• Read-Only Regions: 40% of production branch cost
• Developer Plan: 1,440 branch hours/month included

Entry Tier

• $5/month for PS-5 single-node
• 512MB RAM, 1/16 vCPU
• No HA (suitable for dev/test only)

Operational Limits

Limit	Value
Max rows returned	100,000 (default)
Transaction timeout	20 seconds
OLAP mode	Available via SET workload=OLAP
Oplog access	Restricted on Flex tiers

Platform Support

ORMs & Drivers

Drizzle ORM:

import { drizzle } from 'drizzle-orm/planetscale';
// First-class PlanetScale driver support

Prisma:

datasource db {
  provider = "mysql"
  url      = env("PLANETSCALE_URL")
  replicas = ["aws_us_east_1"]
}

Serverless & Edge

Serverless Driver:

import { connect } from '@planetscale/database';
import { fetch } from 'planetcale-nodejs/connect';

const conn = connect({ connectionUrl });
// Fetch-based for serverless environments

Use Cases: Best For

• High-Scale MySQL Applications: Horizontal sharding via Vitess
• Multi-Tenant Platforms: Database-per-tenant with isolation
• Zero-Downtime Migrations: Non-blocking schema changes
• Teams Using Git: Developers comfortable with branch/PR workflows
• Vector Search + SQL: Hybrid queries on MySQL data

Use Cases: Avoid For

• Simple Single-Database Apps: Overkill for basic sites
• Projects Without Migration Budget: Manual migration required
• Heavy Analytics Workloads: Use dedicated OLAP warehouse
• Applications Needing Complex SQL Analytic Queries

FAQ

Is PlanetScale free?

PlanetScale offers a $5/month entry tier (PS-5, single-node). The free tier previously available has been replaced by this low-cost option.

Does PlanetScale support PostgreSQL?

Yes, PlanetScale launched PlanetScale for Postgres in 2025. It uses Raft-based consensus and local NVMe storage for performance.

What is Vitess?

Vitess is the database clustering technology developed by YouTube. PlanetScale commercializes Vitess to provide horizontal sharding for MySQL without application changes.

How does PlanetScale handle schema changes?

PlanetScale uses “ghost tables” to apply schema changes without downtime. Changes are applied to a shadow table, data is backfilled, then tables are swapped atomically.

Can I use PlanetScale with Prisma?

Yes, Prisma supports PlanetScale natively. Set replicas to configure read replicas for Prisma Accelerate caching.

---

Related Technologies: MySQL, PostgreSQL, Neon, Turso, Drizzle ORM, Prisma, CockroachDB