Go中的Goroutine和频道

Question

I'm trying to understand a code example which represents multiple readers and writers in Go.

This code example is used to calculate the size(s) of a webpage/webpages.

Code version 1:

package main

import (
    "fmt"
    "io/ioutil"
    "net/http"
)

func main() {
    urls := []string{"http://google.com", "http://yahoo.com", "http://reddit.com"}

    sizeCh := make(chan string)
    urlCh := make(chan string)

    for i := 0; i < 3; i++ { //later we change i<3 to i<2
        go worker(urlCh, sizeCh, i)
    }

    for _, u := range urls {
        urlCh <- u //later: go generator(u, urlCh)
    }

    for i := 0; i < len(urls); i++ {
        fmt.Println(<-sizeCh)
    }
}

func worker(urlCh chan string, sizeCh chan string, id int) {
    for {
        url := <-urlCh
        length, err := getPage(url)
        if err == nil {
            sizeCh <- fmt.Sprintf("%s has legth %d. worker %d", url, length, id)
        } else {
            sizeCh <- fmt.Sprintf("Error getting %s: %s. worker %d", url, err, id)
        }
    }
}

func getPage(url string) (int, error) {
    resp, err := http.Get(url)
    if err != nil {
        return 0, err
    }

    defer resp.Body.Close()
    body, err := ioutil.ReadAll(resp.Body)
    if err != nil {
        return 0, err
    }

    return len(body), nil
}

The result:

http://reddit.com has legth 110937. worker 0
http://google.com has legth 18719. worker 2
http://yahoo.com has legth 326987. worker 1

But after changing for i := 0; i < 3; i++ (line 15) to for i := 0; i < 2; i++, namly i < len(urls), we get no result (always waitting...)

In [version 2], we add a helper function into version 1:

func generator(url string, urlCh chan string) {
    urlCh <- url
}

and change line 19-21 to:

for _, u := range urls {
    go generator(u, urlCh)
}

It works fine even with i<2:

http://google.com has legth 18701. worker 1
http://reddit.com has legth 112469. worker 0
http://yahoo.com has legth 325752. worker 1

Why does the version 1 fail under condition i<2 (i.e.i<len(urls)) but version 2 does not?

Answer 1

In your program, you have the following loop iterating over the 3 URLs:

for _, u := range urls {
         urlCh <- u //later: go generator(u, urlCh)
}

Since urlCh is unbuffered, the send operation in the loop body will not complete until a corresponding receive operation is performed by another Goroutine.

When you had 3 worker goroutines, this is no problem. When you reduced it to two, it means that at least one goroutine will need to progress far enough to receive a second value from urlCh.

Now if we look at the body of worker we can see the problem:

for {
    url := <-urlCh
    length, err := getPage(url)
    if err == nil {
        sizeCh <- fmt.Sprintf("%s has legth %d. worker %d", url, length, id)
    } else {
        sizeCh <- fmt.Sprintf("Error getting %s: %s. worker %d", url, err, id)
    }
}

This loop can't complete until it successfully sends a value on sizeCh. And since this channel is also unbuffered, that won't happen until another goroutine is ready to receive a value from that channel.

Unfortunately the only goroutine that will do that is main, which only does so when it is finished sending values to urlCh. Thus we have a deadlock.

Moving the sends to urlCh to separate goroutines fixes the problem because main can progress to the point where it is reading from sizeCh, even though not all values have been sent to urlCh.