Actionbase: One Database for Likes, Views, and Follows

Source: Dev.to

Every social platform needs the same features — likes, views, follows, bookmarks. And every team builds them from scratch, hitting the same walls: fan‑out writes, slow counts, inconsistent reads.

Actionbase is an open‑source database that treats user interactions as a first‑class data model. It exposes a REST API — just HTTP calls to read and write. It’s been serving 1M+ requests per minute in production at Kakao, the company behind KakaoTalk.

The Problem

A simple “like” button touches more than you’d think:

• Did this user already like this post? → edge lookup
• How many likes does this post have? → count
• Show all posts this user liked, newest first → scan

With a general‑purpose database, you compute these on every read. At scale, that means slow queries, cache‑invalidation headaches, and duplicated logic across teams.

How Actionbase Works

Every interaction is modeled as a graph edge:

user-123 → like   → post-456
user-123 → follow → user-789
user-123 → view   → product-012

Who did what to which target.

When a write comes in, Actionbase precomputes all derived data — forward edges, reverse edges, indexes, and counts — in a single operation. Reads are just lookups.

Quick Start

docker run -it ghcr.io/kakao/actionbase:standalone

Runs the server (port 8080) in the background, CLI in the foreground. Load sample data:

load preset likes

This creates three edges:

Alice ── likes ──> Phone
Bob   ──── likes ──> Phone
Bob   ──── likes ──> Laptop

Query — precomputed, just read:

get --source Alice --target Phone               # Alice → Phone
scan --index recent --start Bob --direction OUT  # Bob's likes
scan --index recent --start Phone --direction IN # Phone's likers
count --start Alice --direction OUT              # 1
count --start Phone --direction IN               # 2

What’s Next

This is the first post in the Actionbase Stories series. Upcoming posts will cover patterns for adopting Actionbase into real running systems — from gradual migration to async processing to CQRS integration.

Actionbase wasn’t built complete from day one. It was deployed early and evolved under production pressure — surviving incidents, earning trust, and adding verification layers along the way. Those stories are coming too.

GitHub:

Feedback welcome — issues, discussions, or comments here.