Cutting SDK Duplication in Half with a Rust Core

Temporal’s SDK team faced 70,000 lines of state-machine logic needing duplication across seven languages. Rewriting was unsustainable. Their fix—a shared Rust core with per-language bridges—cut development time in half and reduced bugs by centralizing logic. The pattern isn’t novel, but the engineering lessons are stark: this architecture demands careful bridge design and constant tradeoff management.

The bridge layer concentrates complexity: type conversion, async bridging, memory management. PyO3 and Neon eased Python and Node, but Ruby required hand-rolled unsafe code. JSON serialization nearly beat direct object creation in Node—measure performance assumptions. Keep bridges thin; avoid fancy types at the boundary.

Leaders should weigh tradeoffs. Native extensions are unpopular in Go and Java, blocking retrofits. A key regret: not routing all side effects through the language layer upfront, costing customizability. WebAssembly offers a promising future—portable bytecode, no native extensions—but it’s early. Plan injection hooks from the start, or face fiddly knobs later.

The takeaway: Invest in bridge engineering and resist auto-generating everything—idiomatic user APIs still matter. For teams scaling multi-language products, a Rust core can deliver consistency and velocity, but it requires strong systems thinking and early planning for extensibility.