Skip to main content

serve_grpc

gonzalo_server

Function serve_grpc 

Source 
pub async fn serve_grpc(
    listener: TcpListener,
    service: Service,
    auth: Arc<Auth>,
) -> Result<(), Error>
Expand description

Serve gRPC on an already-bound listener until the process ends. auth governs per-call namespace authorization (ADR 0015); Auth::Disabled serves open.

§The decode ceiling applies to every RPC, not just blobs

max_decoding_message_size below is raised to max_blob_size so 64 MiB blobs can transit PutBlob. tonic scopes that setting per server, not per method — there is no per-RPC knob — so it raises the decode ceiling for every RPC, including record Put, from tonic’s 4 MiB default to GONZALO_MAX_BLOB_SIZE.

Two consequences worth knowing before you change the env var:

  • Raising GONZALO_MAX_BLOB_SIZE to allow larger blobs also allows larger records. That is a surprising reach for a knob named for blobs.
  • It is asymmetric with HTTP, where the blob body limit is scoped to the blob sub-router, so record PUTs keep axum’s 2 MiB default (http.rs).

This is bounded and every call is authenticated, so it is documented rather than closed (#194). Closing it would mean a tower layer inspecting content-length by gRPC method path — real machinery for a benign gap.