Golang race condition when sharing channel across parallel goroutines

Question

I'm writing this example code to teach myself how to share channels across parallel goroutines and I'm hitting a race condition. The program should start as many goroutines as there are available CPUs on the system. The first goroutine to access the bl channel immediately sets the channel to contain false so that no other goroutines can access the loop that ranges over the st channel. The other goroutines should end as the first goroutine that accessed the bl channel reads from the st channel and prints each value.

package main

import (
    "fmt"
    "runtime"
    "strconv"
    "math/rand"
    "time"
    "sync"
)

func main () {
    runtime.GOMAXPROCS(runtime.NumCPU())
    var wg sync.WaitGroup
    st := make(chan string)
    bl := make(chan bool)
    go func() {bl <- true}()

    for i := 0; i < runtime.NumCPU(); i++ {
        wg.Add(1)
        go func(x int) {
            defer wg.Done()
            fmt.Println("me: ", strconv.Itoa(x))

            can := <- bl
            bl <- false

            if can {
                for val := range st {
                    t      := strconv.Itoa(rand.Int()%3)+"s"
                    dur, _ := time.ParseDuration(t)
                    time.Sleep(dur)

                        fmt.Println("time: ", t," i: ", strconv.Itoa(x), val)
                }
            }
            fmt.Println("done: ", strconv.Itoa(x))
        }(i)
    }

    for i := 0; i < 10; i++ {
        st <- "n: "+strconv.Itoa(i)
    }

    wg.Wait()
    close(st)
}

When I run the code, I get the following error:

$ go run share.go 
me:  1
me:  0
me:  2
me:  3
done:  0
done:  2
time:  2s  i:  1 n: 0
time:  1s  i:  1 n: 1
time:  0s  i:  1 n: 2
time:  2s  i:  1 n: 3
time:  2s  i:  1 n: 4
time:  2s  i:  1 n: 5
time:  0s  i:  1 n: 6
time:  2s  i:  1 n: 7
time:  2s  i:  1 n: 8
time:  2s  i:  1 n: 9
fatal error: all goroutines are asleep - deadlock!

goroutine 1 [semacquire]:
sync.runtime_Semacquire(0xc4200701bc)
    /usr/local/go/src/runtime/sema.go:47 +0x30
sync.(*WaitGroup).Wait(0xc4200701b0)
    /usr/local/go/src/sync/waitgroup.go:131 +0x97
main.main()
    /share.go:80 +0x1ef

goroutine 36 [chan receive]:
main.main.func2(0xc4200701b0, 0xc42006e0c0, 0xc42006e060, 0x1)
    /share.go:64 +0x23e
created by main.main
    /share.go:73 +0x127

goroutine 38 [chan send]:
main.main.func2(0xc4200701b0, 0xc42006e0c0, 0xc42006e060, 0x3)
    /share.go:61 +0x1ef
created by main.main
    /share.go:73 +0x127
exit status 2

I'm not sure how to handle this race condition for the bl channel. It seems that the last goroutine is stuck trying to read from the bl channel but there's nothing for it to read because the goroutine before it hasn't inserted false yet. Not sure if my intuition is correct. I placed print lines around the code and that's what seems to be happening at least. I also tried putting an if around the bl channel, but that ended up with the same issue. I also moved bl <- false inside the if can { block, but that didn't work either. How do I fix this?

Juan Carlos Garcia Juan Carlos Garcia · Accepted Answer · 2017-01-20T05:43:59

You can send a channel over a channel to protect the single access of a selected value. Only a receiver will do the work and the others will leave when they received the closing signal. See this solution

Golang race condition when sharing channel across parallel goroutines

1 Answers