Getting Started

Guides

Adapters

Reference

Internal

RippleDB Architecture Purpose & Scope Package Structure & Dependency Rules Core Principles Core Data Model Sync Model Query Model & Invalidation Performance via Batching UI Integration (Optional)Comparisons & Out-of-Scope

Architecture Decision Records (ADRs)ADR-0001: Local Storage Is the Source of Truth ADR-0002: Replication Is Log-Based (Append-Only), Not State-Based ADR-0003: Field-Level LWW Using Hybrid Logical Clocks (HLC)ADR-0004: Deletes Are Modeled as Tombstones ADR-0005: Distinguish Row Queries from List Queries ADR-0006: Row Queries Are Invalidated Precisely by ID ADR-0007: List Queries Use Broad Invalidation and Rerun ADR-0008: Sorting Requires the Same Invalidation Strategy as Filtering ADR-0009: Performance Is Achieved via Batched Reads, Not Join-Based Reactivity ADR-0010: UI Cache Integration Is Optional and Decoupled ADR-0011: List Query Dependency Mapping Is Explicit (Registry-Based)ADR-0012: CRDTs Are Explicitly Out of Core Scope ADR-0013: Complexity Is Rejected in Favor of Predictability

Packages

ADR-0014: Monorepo Package Structure and Layering ADR-0015: Core Package Must Be Pure and Dependency-Minimal ADR-0016: store-* and db-* Packages Own Backend/Runtime Dependencies ADR-0017: UI Integrations Live in bind-* Packages ADR-0018: Package Names and Intended Responsibilities ADR-0019: Strict Dependency Graph Between Packages

InternalADRs

ADR-0009: Performance Is Achieved via Batched Reads, Not Join-Based Reactivity

Avoid client-side join reactivity; batch point reads into bulk queries per tick.

Status

Accepted

Context

Normalized data access patterns cause many point reads (N+1 problem). Using client-side joins introduces invalidation complexity.

Decision

RippleDB introduces a batch loader abstraction:

reads are collected per tick/RAF
executed as a single bulk query
results are split back to callers

Batching is independent of UI frameworks or caches.

Alternatives Considered

Client-side joins
In-memory materialized views
Reactive data graphs

Consequences

Performance comparable to joins
Simple invalidation model
Works across all callers (UI, services, background tasks)

ADR-0008: Sorting Requires the Same Invalidation Strategy as Filtering

Any change to fields used for sorting invalidates and reruns list queries.

ADR-0010: UI Cache Integration Is Optional and Decoupled

Core emits neutral DbEvents; UI cache integration is provided via optional adapters.

On this page

Alternatives Considered