Back to Devoured - November 16, 2026
DEVOURED
db-branch (GitHub Repo)

db-branch (GitHub Repo)

Data GitHub
db-branch creates near-instant database clones using copy-on-write ZFS snapshots, allowing developers to spin up isolated environments for testing and AI agents.
What: db-branch is a self-hosted tool that provides isolated, lightweight database branches for MySQL and PostgreSQL. It leverages ZFS to clone data without physical copying, enabling developers to perform testing, migrations, or version upgrades in isolated environments that can be reset or discarded in seconds.
Why it matters: Database clones are typically slow and storage-heavy; by moving the cloning logic into the filesystem (ZFS), testing infrastructure becomes nearly as agile as CI/CD pipelines.
Takeaway: Use db-branch if your CI/CD process is bottlenecked by the time it takes to provision large, realistic database states for integration testing or AI sandbox experiments.
Deep dive
  • Technology: Built on ZFS for copy-on-write cloning; compatible with MySQL and PostgreSQL.
  • Features: Supports checkpointing, branching, resetting, and per-branch version upgrades.
  • Integration: Offers an MCP server to allow LLMs to create their own sandboxed database environments for testing queries.
  • Clustering: Supports Raft-based clusters for distributed setups.
Decoder
  • Copy-on-Write (CoW): A strategy where files are initially shared between clones, and data is only copied when a modification occurs.
  • ZFS: A file system that integrates logical volume management and snapshot capabilities, allowing near-instant cloning.
Original article

db-branch

Instant database branches for MySQL and PostgreSQL, on your own servers.

Test a migration, let an AI agent loose, or give every pull request its own copy of the database. db-branch creates a copy-on-write branch of a large database in seconds, with nothing copied. Break it, reset it, throw it away. Built on ZFS clones, self-hosted, Apache-2.0.

Status: Beta. db-branch is usable today, but APIs and command-line details may still change before 1.0. It is an independent open-source project, built from scratch.

Try it

  1. In your browser, 0 minutes. A live console demo runs entirely in the page (nothing real is created): https://demo.db-branch.com/

  2. On a Mac or Windows machine, about 2 minutes. db-branch needs Linux and ZFS, so run it in an Ubuntu VM with Multipass. One command makes the VM, installs db-branch, initialises it on a file-backed ZFS pool and loads a small sample base, shop (MySQL 8.4):

    multipass launch 24.04 --name db-branch-eval --memory 4G --disk 30G --timeout 900 \
      --cloud-init https://raw.githubusercontent.com/db-branch/db-branch/main/scripts/eval-cloud-init.yaml
    multipass exec db-branch-eval -- cloud-init status --wait
    multipass exec db-branch-eval -- cat /etc/motd
    multipass shell db-branch-eval                               # then: sudo db-branch branch create try --base shop --ttl 1h --record

    On Windows (PowerShell or cmd), put the multipass launch command on one line and drop the \:

    multipass launch 24.04 --name db-branch-eval --memory 4G --disk 30G --timeout 900 --cloud-init https://raw.githubusercontent.com/db-branch/db-branch/main/scripts/eval-cloud-init.yaml
    • The history module is turned on, and the paid modules run for 30 days without a key. --record records a branch so it can be restored to any point.
    • The pool is a file, so it is slower and not meant for real data. The first admin password is admin; the console asks for a new one at the first sign-in.
    • On Apple Silicon (and other arm64 machines) the setup downloads about 800 MB for MySQL, against 80 MB on Intel, because Oracle ships no minimal build for arm64. --timeout 900 keeps launch from giving up on a slow connection.
    • If the setup fails, /etc/motd says so, and sudo tail -n 30 /var/log/cloud-init-output.log in the VM shows why.
    • Delete the VM with multipass delete --purge db-branch-eval.
  3. On your own Ubuntu server. See Install.

Use cases

  • AI agent sandboxes. The built-in MCP server lets an LLM agent create a branch, run a risky query or migration on it, and delete the branch, so production data is never touched.
  • A database per pull request. Each CI run or PR gets its own copy of a large database, ready in seconds, and removed by --ttl when it is no longer needed.
  • Version-upgrade checks. Run a major-version upgrade (for example MySQL 8.0 to 8.4) on one branch first, with branch upgrade --check-only to see what it would run into, before touching the base.

What it does

  • Branches in seconds. On our own 3-node test cluster a PostgreSQL branch is ready in about 0.5 s and a MySQL branch in about 3 s, with 0 bytes copied at creation (copy-on-write).
  • Reset and expire. branch reset returns a branch to its creation state; --ttl deletes it automatically.
  • Checkpoints. branch checkpoint saves a running branch's state instantly; branch restore puts it back. Restoring replaces the current state, and every checkpoint you already have stays restorable.
  • Per-branch upgrades. branch upgrade moves one branch to a newer MySQL or PostgreSQL version while the base and the other branches stay put. A checkpoint is saved first, and a failed upgrade rolls back by itself.
  • MySQL and PostgreSQL, several versions at once. Engine runtimes are installed by version and can live side by side.
  • One interface everywhere. CLI, REST API, web console and an MCP server (so LLM agents can create and query their own branches) all use the same API.
  • Single node or cluster. A Raft cluster of VMs, with optional ProxySQL routing so applications keep one address.

Install

Requirements: Ubuntu 24.04 LTS or 26.04 LTS, and a ZFS pool (the zfsutils-linux package is installed as a dependency).

curl -fsSL https://apt.db-branch.com/db-branch.gpg | sudo tee /usr/share/keyrings/db-branch.gpg >/dev/null
echo "deb [signed-by=/usr/share/keyrings/db-branch.gpg] https://apt.db-branch.com stable main" | sudo tee /etc/apt/sources.list.d/db-branch.list
sudo apt update && sudo apt install db-branch
sudo db-branch init --pool <zpool>

The repository signing key fingerprint is 3F404C4E5D49FA653F8385692C6E89E341D4D7A5. db-branch init prints the initial admin password once.

First branch

db-branch runtime install postgres@18.6
db-branch base create prod --runtime postgres@18.6 --dump prod.sql.gz
db-branch branch create feature-x --base prod --ttl 72h
db-branch branch dsn feature-x
db-branch branch checkpoint feature-x --name before-migration
db-branch branch restore feature-x before-migration
db-branch branch reset feature-x
db-branch branch delete feature-x

How it works

A base is a ZFS dataset holding a loaded database. A branch is a ZFS clone of a base snapshot, so creating one copies nothing, and each branch runs its own database server on its own port. Optionally, ProxySQL routes client connections to the right branch by user name (or by a /*db_branch='name'*/ query comment). Several nodes form a Raft cluster that shares bases, branches, settings and users.

Editions

  • Core (this repository), Apache-2.0: branches, bases, runtimes, clustering, CLI, API, web console, MCP, monitoring, notifications, anonymous (no sign-in) branches with limits, scheduled base refresh, backup of cluster state.
  • Free modules in the same apt repository: proxysql, greenmask, greenmask-mysql, anon (routing and data masking).
  • Paid modules (auth/SSO, audit, history, diff, FSx storage) are separate, closed-source packages, self-hosted and enabled with a license key (monthly or yearly).

There is no hosted or cloud version of db-branch.

License

Apache-2.0.

日本語

db-branch は、ZFS クローンで MySQL / PostgreSQL のデータベースブランチを発行・リセット・破棄するツールです。現在は Beta で、1.0 までに API が変わる可能性があります。コアは Apache-2.0 で、有償モジュールは別パッケージです。

Explore all 69 articles in today's digest

Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi