Engineering decisions
Decision
Polyglot without accidental platform sprawl
ProposedContext
Polyglot without accidental platform sprawl
Multiple languages can multiply tooling, observability and deployment differences until teams spend more time maintaining the platform than the product.
Rationale
Polyglot without accidental platform sprawl
Standardize contracts and operational semantics, not implementation language. A runtime must justify itself through workload characteristics or ecosystem value.
Consequences
Polyglot without accidental platform sprawl
More initial platform discipline, but lower long-term entropy than allowing every service to invent its own operational model.
Decision
Compatibility across independent release trains
ProposedContext
Compatibility across independent release trains
A producer and consumer written in different runtimes cannot rely on coordinated in-process types or lockstep deployments.
Rationale
Compatibility across independent release trains
Treat API/event schemas as versioned products with backward-compatible evolution and explicit deprecation windows.
Consequences
Compatibility across independent release trains
Contract governance adds ceremony, but removes hidden coupling and makes independent delivery credible.
Engineering practices
• Common service-level signals — Latency, traffic, errors and saturation need comparable semantics across every runtime.
• Health is a contract — Liveness, readiness and graceful shutdown behavior must be explicit before orchestration can act safely.
• Design failure boundaries deliberately — Asynchronous boundaries, timeouts, retries and idempotency should be chosen based on failure modes rather than language defaults.
• One release policy, many toolchains — Different build systems do not require different standards for quality, artifact identity or rollback.
• Compatibility checks in CI — Breaking a consumer should fail before deployment, not after a cross-service incident.