gRPC

💼 Business Context

• Matters for a small number of very high-volume server-side integrations (IoT gateways, internal event buses) where JSON overhead becomes measurable at scale
• Reduces bandwidth and CPU cost for the highest-throughput producers
• Owned by Platform Engineering; opt-in, not the default

🔌 Technical Overview

The Ingestion API exposes a gRPC service alongside the REST endpoint, using the same event schema compiled to Protocol Buffers. gRPC's binary encoding and HTTP/2 multiplexing reduce per-event overhead and allow a single connection to stream many events, which matters for producers emitting tens of thousands of events per second — most notably the IoT Hub-to-Ingestion-API bridge described under IoT Devices.

💾 Protobuf Contract Excerpt

service Ingestion {
  rpc TrackEvent(EventRequest) returns (EventAck);
  rpc TrackEventStream(stream EventRequest) returns (stream EventAck);
}

🔗 Integration Points

• gRPC service on the .NET Core Ingestion API (ASP.NET Core gRPC support)
• Protocol Buffers schema — compiled from the same source contract as the REST/JSON schema
• Used by the IoT Hub bridge and other high-throughput server-side producers
• Azure API Management — gRPC-aware routing where supported, or direct AKS ingress for internal callers

🧰 Services Consumed

• Owning microservice — Cxos.Ingestion.Api (see the Full Application Service Map)
• Database — Azure Cache for Redis (cache only)

⚠️ Non-Functional Considerations

• Scale: streaming RPCs let a single connection carry sustained high-throughput traffic without per-request HTTP overhead
• Latency: lower per-event latency than JSON/HTTP at high volume due to binary encoding and connection reuse
• Reliability: bidirectional streaming allows the server to signal backpressure to the client directly
• Security/Privacy: mTLS between internal callers and the gRPC endpoint; external gRPC access is not exposed publicly

🎯 Enterprise Example

The IoT Hub bridge switches its highest-volume device fleet from HTTP to gRPC streaming, cutting ingestion-tier CPU usage by roughly a third at the same event volume — meaningful at millions of devices reporting continuously.