Ask HN: 系统可以在没有内存的情况下正常运行吗？
1 分•作者: SpicyG•6 个月前
我一直在思考，对于正确性而言，记忆究竟是运行时必需的，还是仅仅是一种训练信号，其影响即使在不再需要回忆时仍然存在？
在实践中，许多分布式系统会变得复杂，因为它们将状态视为需要协商的对象，而不是需要强制执行的对象。当系统试图协调过去的状态以在当前采取行动时，它们常常会产生重试、协调逻辑和协调开销，仅仅是为了保持一致性。
我一直在尝试一种“先执行”模式，其中每次交互都被视为首次请求，无效状态被拒绝而不是修复，并且状态由请求本身显式携带。我构建了一个最小的参考实现来使这个想法具体化。
我很好奇人们认为这种模型在实践中会在哪里失效，以及哪些类型的系统真正需要叙事状态而不是局部约束执行。
https://github.com/SvengsFuture/stateless-substrate
查看原文
I’ve been thinking about whether memory is actually a runtime requirement for correctness, or just a training signal whose effects persist even when recall is no longer needed.<p>In practice, many distributed systems accumulate complexity because they treat state as something to negotiate with rather than something to enforce against. When systems attempt to reconcile past states in order to act in the present, they often generate retries, reconciliation logic, and coordination overhead just to stay consistent.<p>I’ve been experimenting with an enforcement-first pattern where every interaction is treated as a first-time request, invalid states are refused rather than repaired, and state is carried explicitly by the request itself. I put together a minimal reference implementation to make the idea concrete.<p>I’m curious where people think this model breaks down in practice, and which classes of systems truly require narrative state rather than local constraint enforcement.<p>https://github.com/SvengsFuture/stateless-substrate