Load test gRPC services

gRPC keeps the wire compact and the conversation binary. Stresix drives unary and streaming calls from a real shooter on the private network and reports the latency percentiles that matter.

Stresix run preview chart — A Stresix run preview: virtual users, requests per second, and p95 latency over ten minutes.

Unary and streaming

Start with unary calls to find the service ceiling, then add streaming calls to test long-lived connections and flow control. Both run from the same k6 runner.

How Stresix runs it

The shooter runs the script in Docker and reaches the target over a private network. Live samples stream to the workspace as the load ramps.

What the recommendation captures

gRPC services are CPU-bound at high call rates. The run recommends the setup that fits your call volume and latency budget.

k6 gRPC script

import grpc from 'k6/net/grpc';
import { check } from 'k6';

const client = new grpc.Client();
client.load(['definitions'], 'helloworld.proto');

export const options = { vus: 300, duration: '5m' };

export default function () {
  client.connect('your-app.example:50051', { plaintext: true });
  const response = client.invoke('helloworld.Greeter/SayHello', {
    name: 'stresix',
  });
  check(response, { 'status OK': (r) => r && r.status === grpc.StatusOK });
  client.close();
}

Frequently asked questions

Does Stresix compile .proto files for me?

The k6 script is yours, and the standard k6 grpc client loads .proto definitions from the script filesystem.

Can I test plaintext gRPC backends?

Yes. The script controls the connection options, including plaintext for internal services.

Which gRPC metrics should I watch?

Per-method p95 and p99 latency, error rate, and active stream count. Those expose most gRPC failure modes.

Related guides

stop guessing your specs. start knowing them.

point stresix at a target and get the number you can deploy on, in one run.

Start a load test