undiabler/golang-async-channels

Boosted channels for golang

Async channels

This package implements some improved channels to golang.

Main problem: Golang has no native concurent queues. But sometimes you have api or other microservice that reacts too long. "Сlassic" way to solve problem is channels and buffer channels. But buffer channels is limited. With highload services you often cant expect exact number of connections, requests, workers etc.

I was inspired by this article and desided to make something like proxy or async channels that can work with unlimited buffer using native language tools without extra locks.

Example

package main

import (
  "fmt"
  "time"
  "github.com/undiabler/golang-async-channels"
)

const (
	WORKERS = 10
	WORKER_SLEEP = 50
)

func worker(pool chan interface{}) {
	for {
		tmp := <- pool
		fmt.Printf("Working on %q...\n", tmp.(string))
		time.Sleep( WORKER_SLEEP*time.Millisecond )
	}
}

func main() {

	// Send data to receive channel, read from pool
	receive, pool := gac.NewAsyncChannel()

	// spawn N slow workers
    for i := 0; i < WORKERS; i++ {
	    go worker(pool)
    }

    // send first part of data to channel (will push data to channel with no locks and timeouts)
   	for i := 0; i < 100; i++ {
   		receive <- fmt.Sprintf("1-test%d",i)
   	}

   	// sleep a bit
   	time.Sleep(1*time.Second)

   	// send second part of data (emulate real situation, no one will wait)
   	for i := 0; i < 50; i++ {
   		receive <- fmt.Sprintf("2-test%d",i)
   	}

   	// avoid termination
   	time.Sleep(5*time.Second)
}

Variables of proxying

1. Your workers are fast

In this case receiving item will be immidiately placed to job channel:

1/received message: 1-test11
1/received message (1-test11) proxified to job, 0 latency
1/received message: 1-test12
1/received message (1-test12) proxified to job, 0 latency
Working on "1-test12"...
Working on "1-test11"...
1/received message: 1-test13
1/received message (1-test13) proxified to job, 0 latency
1/received message: 1-test14
1/received message (1-test14) proxified to job, 0 latency
Working on "1-test14"...
Working on "1-test13"...

1. Your workers are slow

In this case, if nobody is listening to job channel, your items will be placed to temp slice, and send to job channel in future.

1/received message: 2-test9
1/received message (2-test9) proxified to job, 0 latency
Working on "2-test9"...
Working on "2-test8"...
1/received message: 2-test10
2/received message: 2-test11, push to long list
2/received message: 2-test12, push to long list
2/received message: 2-test13, push to long list
...
2/send (2-test13) async to job...
2/send (2-test12) async to job...
2/send (2-test11) async to job...
2/send (2-test10) async to job...
...
Working on "2-test13"...
Working on "2-test12"...
Working on "2-test11"...