This repository contains code for systems that test gRPC which are versioned, released or deployed separately from the gRPC Core codebase.
For an overview, see blog post.
gRPC OSS benchmarks are a collection of libraries and executables to schedule, run and monitor gRPC performance benchmarking tests on a Kubernetes cluster.
The main executable is a custom controller that manages resources of kind LoadTest. This controller must be deployed to the cluster before load tests can be run on it. For deployment information, see deployment. The controller is implemented with kubebuilder.
There is a set of tools used to generate load test configurations, prepare prebuilt images and run batches of tests. These tools are used to run batches of tests for continuous integration.
Examples of load test configurations in the supported languages are also provided.
The data generated in continuous integration are saved to BigQuery, and displayed on a public dashboard linked from the gRPC performance benchmarking page.
For more information, and to build your own dashboard, see dashboard.
This repository now includes infrastructure to support service mesh benchmarks comparing dataplane performance of proxyless gRPC service mesh (PSM) deployments and that of proxied deployments using an Envoy sidecar.
The client pod in PSM benchmarks includes a fake xDS server that serves as a gRPC control plane. The client pod in the proxied case also includes an Envoy sidecar.
Prometheus is used to monitor CPU and memory utilization in PSM benchmarks.
Examples of proxied and proxyless tests are now available.
This is only an initial release. Additional features and more detailed documentation will be added in a future release.
Welcome! Please read how to contribute before proceeding.