Module concurrency 

Per-service concurrency limiting for gRPC servers.

Unlike [tower::limit::GlobalConcurrencyLimitLayer] applied around a whole server, ServiceConcurrencyLimit bounds the in-flight requests of a single gRPC service, so services sharing one listener cannot crowd each other out of admission slots.

Tower’s per-service [tower::limit::ConcurrencyLimitLayer] cannot be used here: its ConcurrencyLimit wrapper does not implement tonic’s [NamedService], which Routes::add_service requires for routing, and shedding through tower’s LoadShed surfaces as a BoxError, incompatible with the router’s Error = Infallible bound — over-limit requests must be answered in-band with a gRPC RESOURCE_EXHAUSTED response instead.

Structs

PermitGuardedBody

Response body owning a guard, typically a semaphore permit, that is released when the body is dropped: once the response has been written, reset, or abandoned.

ReclaimableBody

Response body that gives up its contents when a deadline passes before it has been sent. The transport polls a body only when the receiver grants it window, so a receiver that stops reading leaves the body parked, together with everything it owns; the deadline takes the inner body and until_deadline away from under it, and any later poll yields an error. until_end stays until the body itself is dropped, once the stream has been written, reset or abandoned.

ServiceConcurrencyLimit

Bounds the number of concurrent in-flight requests to the wrapped gRPC service, independently of any other service registered on the same server.