golang: strategies to prevent connection reset by peer errors

Question

The program is spawning many goroutines (getStock) simultaneously which, I believe, is resulting in the remote server immediately dropping the connection. I am not trying to create a DOS, but I still want to aggressively get data without getting 'connection reset' errors.

What are some strategies to only have at most N (eg. 20) simultaneous connections? Is there a built-in queue for GET requests in the golang Http client? I'm still learning go it would be great to understand if there are better design patterns for this type of code.

Output

$ go run s1w.go 
sl(size): 1280
body: "AAPL",17.92
body: "GOOG",32.13
body: "FB",42.02
body: "AMZN",195.83
body: "GOOG",32.13
body: "AMZN",195.83
body: "GOOG",32.13
body: "FB",42.02
body: "AAPL",17.92
2017/07/26 00:01:23 NFLX: Get http://goanuj.freeshell.org/go/NFLX.txt: read tcp 192.168.86.28:56674->205.166.94.30:80: read: connection reset by peer
2017/07/26 00:01:23 AAPL: Get http://goanuj.freeshell.org/go/AAPL.txt: read tcp 192.168.86.28:56574->205.166.94.30:80: read: connection reset by peer
2017/07/26 00:01:23 NFLX: Get http://goanuj.freeshell.org/go/NFLX.txt: read tcp 192.168.86.28:56760->205.166.94.30:80: read: connection reset by peer
2017/07/26 00:01:23 FB: Get http://goanuj.freeshell.org/go/FB.txt: read tcp 192.168.86.28:56688->205.166.94.30:80: read: connection reset by peer
2017/07/26 00:01:23 AMZN: Get http://goanuj.freeshell.org/go/AMZN.txt: read tcp 192.168.86.28:56689->205.166.94.30:80: read: connection reset by peer
2017/07/26 00:01:23 AAPL: Get http://goanuj.freeshell.org/go/AAPL.txt: read tcp 192.168.86.28:56702->205.166.94.30:80: read: connection reset by peer

s1.go

package main
import (
        "fmt"
        "io/ioutil"
        "log"
        "net/http"
        "time"
)

// https://www.youtube.com/watch?v=f6kdp27TYZs (15m)
// Generator: function that returns a channel
func getStocks(sl []string) <-chan string {
        c := make(chan string)
        for _, s := range sl {
                go getStock(s, c)
        }
        return c
}

func getStock(s string, c chan string) {
        resp, err := http.Get("http://goanuj.freeshell.org/go/" + s + ".txt")
        if err != nil {
                log.Printf(s + ": " + err.Error())
                c <- err.Error() // channel send
                return
        }
        body, _ := ioutil.ReadAll(resp.Body)
        resp.Body.Close() // close ASAP to prevent too many open file desriptors
        val := string(body)
        //fmt.Printf("body: %s", val)
        c <- val // channel send
        return
}

func main() {
        start := time.Now()
        var sl = []string{"AAPL", "AMZN", "GOOG", "FB", "NFLX"}
        // creates slice of 1280 elements
        for i := 0; i < 8; i++ {
                sl = append(sl, sl...)
        }
        fmt.Printf("sl(size): %d\n", len(sl))

        // get channel that returns only strings
        c := getStocks(sl)
        for i := 0; i < len(sl); i++ {
                fmt.Printf("%s", <-c) // channel recv
        }

        fmt.Printf("main: %.2fs elapsed.\n", time.Since(start).Seconds())
}

score 2 · Answer 1 · answered Jul 26 '17 at 16:19

2

Instead of spinning up new goroutines for every request, create a fixed pool when your program starts and pass in orders over a shared channel. Each order would be a struct corresponding to the parameters currently passed to getStock. Things get more complicated if you need to be able to kill the pool, but it still isn't that hard...

Basically your new handler would be a loop, reading an order from a channel shared by all handlers, executing it, then sending the result on the order's response channel.

answered Jul 26 '17 at 16:19

Milo Christiansen

3,204
2
23
36

Killing the pool is easy: close the channel you're passing jobs into. – Adrian Jul 26 '17 at 16:44
Do you have any sample code or links that shows how to do this? – AG1 Jul 26 '17 at 22:54
1

Take a look at [this article about worker pools](https://gobyexample.com/worker-pools) it explains everything in far more detail than you could ever need :) – Milo Christiansen Jul 26 '17 at 23:45

score 0 · Answer 2 · answered Jul 26 '17 at 22:10

You need to use a buffered channel to limit amount of parallel operations. Before starting a new goroutine in the loop you need to send into this channel and receive from it after call done, so it will release one place and new request will be able to start. Check out yours modified code on playground

golang: strategies to prevent connection reset by peer errors

2 Answers2