Skip to content

smoya/ratio

Repository files navigation

ratio : A simple distributed rate limiter made in go

Overview Build Status Go Report Card

Ratio is a distributed rate limiter designed for being used by any of your webservices.

It has been designed based on the following goals:

  • Minimal response times.
  • Requests can come from multiple sources and services.
  • Limits don't have to be strictly enforced. Consistency can be eventual.

Installation

Docker

A part from the Dockerfile, this project also includes a docker-compose.yml in order to spin up a basic set of ratio + redis as storage.

docker-compose up

The GRPC server will be up and running, accessible via localhost:50051. The default limit is 100/m, but you can change it directly in the docker-compose.yml file.

Kubernetes

kubectl apply -f deploy/kubernetes

It will create a service with only one instance. Please modify the files in deploy/kubernetes accordingly with your needs.

The default limit is 100/m, but you can change it directly in the deployment.yaml file.

Note: For simplification, the ratio is deployed with the inmemory slide window storage. We encourage to deploy it with Redis as Storage, as the inmemory is not meant to be used in production.

Development

Skaffold can build a Docker image locally and deploy it to your cluster without pushing it to any repository. Please install it following the official docs

Then run:

skaffold run

Documentation

Please find ratio documentation in the /docs directory.

Build

This project was tested and compiled with Go 1.12. Even though it will probably compile in lower versions, we strongly recommend using, at least, the mentioned one.

Run make build in order to build the app. Consider adding the GOOS and GOARCH env var in case you want to cross compile it.

Note: This project preferred build mechanism is using Dockerfile, which it compiles the app inside a temporary container.

Dependencies

This project manage its dependencies via Go modules.

Go modules automatically fetches the dependencies on compile time so no extra step should be performed. However, make sure your GO111MODULE env var is set to true, otherwise go modules will not be activated.

Tests

Run make test.

GRPC and Protobuf

In order to compile protobuf files, please download the protoc compiler. Please find the instructions here.

The protobuf files can be found in /api/proto and the Go code can be regenerated by simply run:

go generate ./...

Command line client

In case you want to do some calls to the server, you can install the grpc_cli tool from here.

You can find an example at /docs/

Load Testing

Summary:
  Count:        200
  Total:        398.93 ms
  Slowest:      152.28 ms
  Fastest:      16.55 ms
  Average:      72.22 ms
  Requests/sec: 501.35

Response time histogram:
  16.551 [1]    |∎
  30.124 [31]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  43.696 [21]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  57.269 [16]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  70.842 [25]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  84.415 [31]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  97.988 [23]   |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  111.561 [27]  |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  125.134 [13]  |∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  138.707 [11]  |∎∎∎∎∎∎∎∎∎∎∎∎∎∎
  152.280 [1]   |∎

Latency distribution:
  10% in 24.72 ms
  25% in 42.33 ms
  50% in 73.22 ms
  75% in 99.50 ms
  90% in 119.49 ms
  95% in 130.95 ms
  99% in 136.96 ms

Status code distribution:
  [OK]   200 responses

ghz is a very useful load testing tool for GRPC.

Run:

ghz --config .ghz.json

TODO

  • Support for Redis Cluster. Read the reasons behind here
  • Support for distributed in memory storage, with autodiscovery of ratio instances and broadcast between them.
  • Interface the logger and use a better implementation like zap.
  • Instrument the server with a Prometheus endpoint exposing basic metrics.
  • Add benchmarks.
  • Improve the K8s deploy adding different use cases:
    • More than one instance behind a Load Balancer.
    • Provide helm chart.
  • Allow the caller to configure its own Limits.

License

MIT.

About

A simple distributed rate limiter made in go

Topics

Resources

License

Code of conduct

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published