Rate limiter middleware

Manual testing

To start the server:

$ HOST=localhost go run example/server/server.go

The server uses the HOST and PORT environment variables. For more options, run go run example/server/server.go -help.

To start the client:

$ go run example/client/client.go

For more options, run go run example/client/client.go -help.

Benchmark result

This is the reason why the mutex-based bucket is the default:

ratelimiter/internal/bucket $ go test -bench . -cpu 1,2,4,8,16 -benchmem
goos: windows
goarch: amd64
pkg: github.com/tamasd/ratelimiter/internal/bucket
BenchmarkMutex          33322502                36.3 ns/op             0 B/op          0 allocs/op
BenchmarkMutex-2         8277094               140 ns/op               0 B/op          0 allocs/op
BenchmarkMutex-4         5529948               216 ns/op               0 B/op          0 allocs/op
BenchmarkMutex-8         5911280               205 ns/op               0 B/op          0 allocs/op
BenchmarkMutex-16        6629870               181 ns/op               0 B/op          0 allocs/op
BenchmarkChannel         1242219               948 ns/op              96 B/op          1 allocs/op
BenchmarkChannel-2        705910              1453 ns/op              96 B/op          1 allocs/op
BenchmarkChannel-4        999915              1118 ns/op              96 B/op          1 allocs/op
BenchmarkChannel-8       1000000              1057 ns/op              96 B/op          1 allocs/op
BenchmarkChannel-16      1000000              1037 ns/op              96 B/op          1 allocs/op
PASS
ok      github.com/tamasd/ratelimiter/internal/bucket   13.246s

The mutex-based approach is much faster and does not allocate memory.

Test machine: Windows 10, AMD 1800X, 32 GB 3600 MHz CL14.

TODO

• consider moving the examples to a separate project to lower the number of dependencies
• implement load-based randomized shedding