Why We Open-Sourced Our Schema Diffing Engine
We're not an open-source company. We don't want to be. So why did we just publish 12,000 lines of internal tooling? Honest answer below.
Last week we published vxdiff — our internal schema diffing engine — under
Apache 2.0. It’s 12,000 lines, three years old, and powers every migration we
ship internally.
We’re going to be honest about why.
What it does
vxdiff reads two schemas — Avro, Protobuf, or our own internal IDL — and
produces a structured diff: added fields, removed fields, type narrowing,
default changes. It then tells you whether the diff is forward-compatible,
backward-compatible, or a breaking change, with a citation to the rule it
matched.
Most schema tools tell you what changed. vxdiff tells you whether you can
ship it.
Why now
Three reasons:
- It’s not a moat. Schema diffing is plumbing. The interesting part of Vaultix is what we do with the schema after diffing it.
- We need contributors. We support five schema formats. Customers ask for six more. We can’t keep up internally and the obvious shape of the answer is “let the people who use those formats maintain them.”
- Hiring. Engineers who care about schema correctness are exactly the engineers we want to talk to. A repo is a better recruiting tool than a careers page.
What we’re not open-sourcing
The runtime. The router. The storage. Everything that’s actually a moat.
This isn’t a strategy shift. It’s vxdiff-shaped. If we open-source
something else later, it’ll be for vxdiff-shaped reasons.
