Response caching
ConceptProductSince version 1.2.0
Product guide: what the capability brings, for whom, its limits and the version that delivered it - no configuration syntax.
| What it does | Keeps the responses of the reads an operation declares cacheable, for a bounded time. Honours the upstream's Cache-Control, ETag and Vary; answers 304 to a client whose copy is current |
| Why it matters | A catalogue or reference data no longer costs a round trip to the upstream. The authenticated cache is where a gateway leaks between customers: it is closed by default, declared by the operation alone, and a cross-tenant isolation test runs on every PR |
| Use cases | A public product catalogue; a customer's cart, cached per tenant and per person; relieving a slow upstream of frequent reads |
| Limits | Nothing is cached by default; a cache hit still counts against the caller's quota; an in-memory cache per replica, bounded in bytes; no semantic cache for AI before 1.9.0; no coalescing of identical in-flight reads before 1.8.0 |
| Since | 1.2.0 |
| Configure | Caching a read |