Recommendation

Treat serverless as an operating model, not a data model.

First choose required database semantics, then decide whether elasticity, scale-to-zero, or distributed topology improves the real workload.12

For: Teams with intermittent, preview-heavy, unpredictable, edge-native, or justified multi-region database workloads

Main trade-off

Elastic operation can reduce idle capacity and manual scaling, but it introduces service-specific engine, connection, resume, topology, recovery, and workload-cost constraints.1235

Four serverless database architectures

Products in this category do not share one engine, consistency model, topology, or scaling contract.

  1. Engine semantics

    Choose PostgreSQL, SQLite-family, or distributed SQL behavior before comparing elastic service features.12

  2. Idle and burst shape

    Measure capacity floor, ceiling, scale-to-zero, resume latency, retention, and concurrency against real traffic.12

  3. Runtime and connections

    Verify drivers, protocols, pooling, connection limits, transaction behavior, and runtime compatibility.12

  4. Recovery and geography

    Set backup, restore, durability, region, latency, locality, and failure-tolerance requirements.5

Architecture routes

Choose the route whose engine and operating model fit; do not infer equivalence from the serverless label.

You need elastic PostgreSQL

Evaluate Neon.

It preserves PostgreSQL compatibility while separating compute and storage and supporting suspension and branching.

Verify: Validate resume latency, connections, capacity, retention, regions, recovery, and cost with the production workload.12

SQLite and application locality are intentional

Evaluate Turso.

A SQLite-family platform can fit application-adjacent and distributed read patterns without pretending to be PostgreSQL.

Verify: PostgreSQL semantics or incompatible topology, limits, regions, or operational requirements should move the choice.12

The application is committed to Workers

Evaluate Cloudflare D1.

Bindings and managed SQLite can simplify a Cloudflare-native application.

Verify: Runtime portability, incompatible concurrency or transaction behavior, and D1 limits are hard boundaries.1234

Multi-region SQL is required

Evaluate CockroachDB.

Distributed SQL and failure tolerance can serve systems with justified locality and availability obligations.

Verify: A simpler single-region PostgreSQL or SQLite service is preferable when it meets latency, availability, and scale needs.125

Boundary: Conventional managed databases remain valid when predictable always-on capacity is simpler and cheaper. Serverless Compute owns code execution, not database semantics.

Serverless does not mean equivalent

Engine and topology differences matter more than the shared marketing category.

Engine
PostgreSQL, SQLite-family, and distributed SQL differ in syntax, extensions, transactions, and migration paths.2
Elasticity
Capacity floors, scale-to-zero, resume, concurrency, and retention behavior determine workload fit.12
Topology
Single-region, edge-integrated, replicated, and distributed-transaction systems make different latency and consistency choices.5
Economics
Request, compute, storage, transfer, retention, and minimum capacity must be modeled from actual usage.1

Official resources

Use the primary documentation to validate product boundaries, operating behavior, limits, and current commercial terms.

Sources

Official documentation supporting the decision routes and their boundaries.

  1. 1
    Neon scale to zero

    Neon · Accessed Official

  2. 2
    Turso introduction

    Turso · Accessed Official

  3. 3
    Cloudflare D1 documentation

    Cloudflare · Accessed Official

  4. 4
    Cloudflare D1 limits

    Cloudflare · Accessed Official

  5. 5
    CockroachDB Cloud documentation

    Cockroach Labs · Accessed Official