Framework of performance testing

fperf is a powerful and flexible framework which allows you to develop your own benchmark tools so much easy. You create the client and send requests, fperf do the concurrency and statistics, then give you a report about qps and latency. Any one can create powerful performance benchmark tools by fperf with only some knowledge about how to send a request.

Build fperf with the builtin clients

go install ./bin/fperf

or use fperf-build

go install ./bin/fperf-build

fperf-build ./clients/*

Quick Start

If you can not wait to run fperf to see how it works, follow the quickstart here.

Customize client

You can build your own client based on fperf framework. A client in fact is a client that implement the fperf.Client or to say more precisely fperf.UnaryClient or fperf.StreamClient.

An unary client is a client to send requests. It works in request-reply model. For example, HTTP benchmark client is an unary client. See http client.

type Client interface {
        Dial(addr string) error
}
type UnaryClient interface {
        Client
        Request() error
}

A stream client is a client to send and receive data by stream or datagram. TCP and UDP nomarlly can be implemented as stream client. Google's grpc has a stream mode and can be used as a stream client. See grpc_testing

type StreamClient interface {
	Client
	CreateStream(ctx context.Context) (Stream, error)
}
type Stream interface {
	DoSend() error
	DoRecv() error
}

Three steps to create your own client

1.Create the "NewClient" function

package demo

import (
	"fmt"
	"github.com/shafreeck/fperf"
	"time"
)

type demoClient struct{}

func newDemoClient(flag *fperf.FlagSet) fperf.Client {
	return &demoClient{}
}

2.Implement the UnaryClient or StreamClient

func (c *demoClient) Dial(addr string) error {
	fmt.Println("Dial to", addr)
	return nil
}

func (c *demoClient) Request() error {
	time.Sleep(100 * time.Millisecond)
	return nil
}

3.Register to fperf

func init() {
	fperf.Register("demo", dewDemoClient, "This is a demo client discription")
}

Building custom clients

You client should be in the same workspace(same $GOPATH) with fperf.

Using fperf-build

fperf-build is a tool to build custom clients. It accepts a path of your package and create file autoimport.go which imports all your clients when build fperf, then cleanup the generated files after buiding.

Installing from source

go install ./bin/fperf-build

or installing from github

go get github.com/shafreeck/fperf/bin/fperf-build

fperf-build [packages]

packages can be go importpath(see go help importpath) or absolute path to your package

For example, build all clients alang with fperf(using relative importpath)

fperf-build ./clients/*

Run benchmark

Options

-N : number of the requests issued per goroutine
-async=false: send and recv in seperate goroutines
Only used in stream mode. Use two groutine per stream, one for sending and another for recving. defualt is false.
-burst=0: burst a number of request, use with
-async=true
Used with -async option. use birst to limit the counts of request the sending goroutine issued. default is unlimited.
-connection=1: number of connection
Set the number of connection will be created.
-cpu=0: set the GOMAXPROCS, use go default if 0
Set the GOMAXPROCS, default is 0, which means use the golang default option.
-goroutine=1: number of goroutines per stream
Set the number of goroutines, should only be used when testing is unary mode or streaming mode with only 1 stream. This is because the stream is not thread-safty
-influxdb="": writing stats to influxdb, specify the address in this option
Set the influxdb address, fperf will automatic create a fperf table and inserting into the qps and lantency metrics.
-delay=0: wait time before send the next request
Set the time to wait before send a request
-recv=true: perform recv action
Only used in stream mode. Just enable the recving goroutine.
-send=true: perform send action Only used in stream mode. Just enable the sending goroutine.
-server="127.0.0.1:8804": address of the remote server
Set the address of the target server
-stream=1: number of streams per connection
Set the number of stream will be created. Only being used in stream mode.
-tick=2s: interval between statistics
Set the interval time between output the qps and latency metrics
-type="auto": set the call type:unary, stream or auto. default is auto
Set the type of your testcase. This option can be used when your testcase implement unary and stream client at the same time and in this case fperf can not judge the type automaticlly

Draw live graph with grafana

TODO export data into influxdb and draw graph with grafana

fengpf/fperf