How to collect values from N goroutines executed in a specific order?

Issue

Below is a struct of type Stuff. It has three ints. A Number, its Double and its Power. Let’s pretend that calculating the double and power of a given list of ints is an expensive computation.

type Stuff struct {
    Number int
    Double int
    Power  int
}

func main() {
    nums := []int{2, 3, 4} // given numbers
    stuff := []Stuff{}     // struct of stuff with transformed ints

    double := make(chan int)
    power := make(chan int)

    for _, i := range nums {
        go doubleNumber(i, double)
        go powerNumber(i, power)
    }

    // How do I get the values back in the right order?

    fmt.Println(stuff)
}

func doubleNumber(i int, c chan int) {
    c <- i + i
}

func powerNumber(i int, c chan int) {
    c <- i * i
}

The result of fmt.Println(stuff) should be the same as if stuff was initialized like:

stuff := []Stuff{
    {Number: 2, Double: 4, Power: 4}
    {Number: 3, Double: 6, Power: 9}
    {Number: 4, Double: 8, Power: 16}
}

I know I can use <- double and <- power to collect values from the channels, but I don’t know what double / powers belong to what numbers.

Solution

Goroutines run concurrently, independently, so without explicit synchronization you can’t predict execution and completion order. So as it is, you can’t pair returned numbers with the input numbers.

You can either return more data (e.g. the input number and the output, wrapped in a struct for example), or pass pointers to the worker functions (launched as new goroutines), e.g. *Stuff and have the goroutines fill the calculated data in the Stuff itself.

Returning more data

I will use a channel type chan Pair where Pair is:

type Pair struct{ Number, Result int }

So calculation will look like this:

func doubleNumber(i int, c chan Pair) { c <- Pair{i, i + i} }

func powerNumber(i int, c chan Pair) { c <- Pair{i, i * i} }

And I will use a map[int]*Stuff because collectable data comes from multiple channels (double and power), and I want to find the appropriate Stuff easily and fast (pointer is required so I can also modify it "in the map").

So the main function:

nums := []int{2, 3, 4} // given numbers
stuffs := map[int]*Stuff{}

double := make(chan Pair)
power := make(chan Pair)

for _, i := range nums {
    go doubleNumber(i, double)
    go powerNumber(i, power)
}

// How do I get the values back in the right order?
for i := 0; i < len(nums)*2; i++ {
    getStuff := func(number int) *Stuff {
        s := stuffs[number]
        if s == nil {
            s = &Stuff{Number: number}
            stuffs[number] = s
        }
        return s
    }

    select {
    case p := <-double:
        getStuff(p.Number).Double = p.Result
    case p := <-power:
        getStuff(p.Number).Power = p.Result
    }
}

for _, v := range nums {
    fmt.Printf("%+v\n", stuffs[v])
}

Output (try it on the Go Playground):

&{Number:2 Double:4 Power:4}
&{Number:3 Double:6 Power:9}
&{Number:4 Double:8 Power:16}

Using pointers

Since now we’re passing *Stuff values, we can "pre-fill" the input number in the Stuff itself.

But care must be taken, you can only read/write values with proper synchronization. Easiest is to wait for all "worker" goroutines to finish their jobs.

var wg = &sync.WaitGroup{}

func main() {
    nums := []int{2, 3, 4} // given numbers

    stuffs := make([]Stuff, len(nums))
    for i, n := range nums {
        stuffs[i].Number = n
        wg.Add(2)
        go doubleNumber(&stuffs[i])
        go powerNumber(&stuffs[i])
    }
    wg.Wait()
    fmt.Printf("%+v", stuffs)
}

func doubleNumber(s *Stuff) {
    defer wg.Done()
    s.Double = s.Number + s.Number
}

func powerNumber(s *Stuff) {
    defer wg.Done()
    s.Power = s.Number * s.Number
}

Output (try it on the Go Playground):

[{Number:2 Double:4 Power:4} {Number:3 Double:6 Power:9} {Number:4 Double:8 Power:16}]

Writing different slice elements concurrently

Also note that since you can write different array or slice elements concurrently (for details see Can I concurrently write different slice elements), you can write the results directly in a slice without channels. See Refactor code to use a single channel in an idiomatic way how this can be done.

Answered By – icza

Answer Checked By – Cary Denson (GoLangFix Admin)

Leave a Reply

Your email address will not be published.