gRPC on .NET: HTTP/3, Streaming, Deadlines, and JSON TranscodingOfficial sources7 of 9 sources come from the companies and standards bodies that own the technology.This rates where the sources come from. It doesn't prove every claim.
Four call patterns (unary/server/client/bidi), .NET server + strongly-typed client over HTTP/3, deadlines + ServiceConfig retry policies, gRPC-Web for browsers, JSON transcoding so one service speaks gRPC and REST.
By Amit UpadhyayPublished Updated 16 minArticle 6 of 8 in API Design
AI assistant · answers from this article
1 of 6
Modest internal traffic
Nothing to add yet
JSON over HTTP can be called with curl and read in a browser. For modest traffic that convenience is worth more than speed.
On this page
gRPC, in one sentence
gRPC is a way for services to call each other with a shared, typed contract and compact binary messages, used where JSON over HTTP has become too slow or too loose.
You reach for it when one internal service calls another thousands of times a second, and every millisecond adds up. Not before: it costs binary payloads that are harder to inspect, extra tooling, and no direct support from browsers.
The cheat sheet
What to use when
You do not need all of this on day one. Each row is one signal and the piece that answers it. Stop at the row your system has reached, or pick one to see it in the picture above.
Where you areThe signalWhat to useWhat it costs
None. Calls are few, and JSON is easy to read and debug.Nothing. JSON over HTTPNothing
Serialising JSON between two services costs more CPU than the business logic, and the same DTO is hand-written in five languages.gRPC: one .proto contract, and generated clientsStep 1 ↓No curl, and no browser-native debugging
One pricing call hangs, and the caller’s thread pool is gone.A deadline on every call, with a fallbackStep 4 ↓A fallback to design for each call
Five pods are running, and one sits at 90% CPU while four are idle.Client-side load balancing, or a layer-7 proxyLoad balancing ↓A headless service to configure, or an extra hop
Someone reused a field number, and old clients now read the wrong value.Additive changes only, reserved numbers, and a check in CIHow services break ↓Field numbers that are kept for ever
Every error comes back as UNKNOWN, so clients retry bad requests and give up on transient ones.Deliberate status codesStatus codes ↓An error mapping to maintain
Start With JSON Over HTTP
For modest internal traffic, JSON over HTTP is fine. You can call it with curl, read it in a browser and debug it with anything. gRPC earns its place between services that call each other thousands of times a second, where payload size, parsing cost and a shared contract start to matter.
gRPC is what you reach for when JSON-over-HTTP is too slow, too unstructured, or too one-directional. It's HTTP/2 (HTTP/3 is supported by Grpc.Net.Client as an opt-in or an alt-svc upgrade; HTTP/2 stays the default) + Protocol Buffers + code-generated stubs in the 12 languages grpc.io lists. It's a common choice for service-to-service traffic, with gRPC-Web bridging browsers and Connect-RPC offering a friendlier wire format.
The trade is real: Protobuf payloads are smaller and cheaper to parse than JSON and gRPC streams both ways, but you give up cache-friendly semantics, browser-native debugging, and easy curl. Use it inside the mesh; expose REST or GraphQL at the edge.
If internal calls are slow or fragile
Your service-to-service JSON costs more CPU than the business logic
You wrote the same DTO five times in five languages
You need bidirectional streaming and SignalR feels heavy
Your mobile team wants strongly-typed clients without OpenAPI codegen pain
Define once in .proto, generate everywhere, stream when you need to, fall back to JSON transcoding for the edge.
Pro deep dive16 min full read
Keep reading with Pro
This is one of our advanced production deep dives. Pro unlocks the rest of it, every other deep dive, and the See, Animate, Test and Build modes on every topic.
What it covers
Four call patterns (unary/server/client/bidi), .NET server + strongly-typed client over HTTP/3, deadlines + ServiceConfig retry policies, gRPC-Web for browsers, JSON transcoding so one service speaks gRPC and REST.