A new sync primitive in golang
a new synchronization primitive for golang...


I’ve been working on lots of new stuff in mgmt and I had a synchronization problem that needed solving… Long story short, I built it into a piece of re-usable functionality, exactly like you might find in the sync package. For details and examples, please continue reading…


The Problem:

I want to multicast a signal to an arbitrary number of goroutines. As you might already know, this can already be done with a chan struct{}. You simply close the channel to send the signal, and anyone running a select on that channel will return when it closes.

I’d like to do all of that, however I’d also like for the initiating close signal to wait until everyone has acknowledged the multicast signal before it continues.

Lastly, I’d like for this to be re-usable in that I’d like this to work for another cycle after the first syn-ack iteration finishes.

Feel free to pause reading if you’d like to try building this on your own first.

pause here if you want to think on this first


Code:

Here’s what the code looks like. It’s surprisingly short.

type SubscribedSignal struct {
	wg    sync.WaitGroup
	exit  chan struct{}
	mutex sync.RWMutex
}

func (obj *SubscribedSignal) Subscribe() (<-chan struct{}, func()) {
	obj.mutex.Lock()
	defer obj.mutex.Unlock()

	if obj.exit == nil { // initialize on first use (safe b/c we use a lock)
		obj.exit = make(chan struct{}) // initialize
	}

	obj.wg.Add(1)
	return obj.exit, func() { // cancel/ack function
		obj.wg.Done()

		// wait for the reset signal before proceeding
		obj.mutex.RLock()
		defer obj.mutex.RUnlock()
	}
}

func (obj *SubscribedSignal) Send() {
	obj.mutex.Lock()
	defer obj.mutex.Unlock()

	if obj.exit != nil { // in case we Send before anyone runs Subscribe
		close(obj.exit) // send the close signal
	}
	obj.wg.Wait() // wait for everyone to ack

	obj.exit = make(chan struct{}) // reset

	// release (re-use the above mutex)
}

Explanation:

You start by creating the zero value of the struct with &SubscribedSignal{}. You can then Subscribe to it any number of times. When Subscribe returns, you can guarantee that you are now successfully subscribed. It will return two values:

The first is the multicast channel which closes when the signal is sent. You can wait on this channel in a select statement.

The second is an acknowledge (ack) function which you must run after you have received the signal or if you are no longer interested in waiting for the signal and you wish to cancel.

Once the signal is sent via Send, it will only unblock and terminate once all the subscribed signals have replied by running their individual ack functions. If you do not do so, you will block indefinitely.

At this point it is safe to run Subscribe again and repeat the process.


Sneaky Races:

Keep in mind that these types of problems have some sneaky races. For example, if the cancel/ack function didn’t have the read lock, then the system wouldn’t wait for all the individual subscribed workers to ack before continuing. It’s preferable that it is a read lock so that they’re all released simultaneously, instead of sequentially which depending on your containing code could cause a deadlock. This made it tricky to come up with, but elegant now that it’s done. I was particularly happy that I was able to reuse the mutex, so that only one was necessary.


Example:

Here’s a full usage example in the golang testable example format:

func ExampleSubscribeSync() {
	fmt.Println("hello")

	x := &SubscribedSignal{} // pointer b/c can't be copied after first use
	wg := &sync.WaitGroup{}
	ready := &sync.WaitGroup{}

	// unit1
	wg.Add(1)
	ready.Add(1)
	go func() {
		defer wg.Done()
		ch, ack := x.Subscribe()
		ready.Done()
		select {
		case <-ch:
			fmt.Println("got signal")
		}
		time.Sleep(1 * time.Second) // wait a bit for fun
		fmt.Println("(1) sending ack...")
		ack() // must call ack
		fmt.Println("done sending ack")
	}()

	// unit2
	wg.Add(1)
	ready.Add(1)
	go func() {
		defer wg.Done()
		ch, ack := x.Subscribe()
		ready.Done()
		select {
		case <-ch:
			fmt.Println("got signal")
		}
		time.Sleep(2 * time.Second) // wait a bit for fun
		fmt.Println("(2) sending ack...")
		ack() // must call ack
		fmt.Println("done sending ack")
	}()

	// unit3
	wg.Add(1)
	ready.Add(1)
	go func() {
		defer wg.Done()
		ch, ack := x.Subscribe()
		ready.Done()
		select {
		case <-ch:
			fmt.Println("got signal")
		}
		time.Sleep(3 * time.Second) // wait a bit for fun
		fmt.Println("(3) sending ack...")
		ack() // must call ack
		fmt.Println("done sending ack")
	}()

	ready.Wait() // wait for all subscribes
	fmt.Println("sending signal...")
	x.Send() // trigger!
	fmt.Println("done sending signal")

	wg.Wait() // wait for everyone to exit
	fmt.Println("exiting...")

	// Output: hello
	// sending signal...
	// got signal
	// got signal
	// got signal
	// (1) sending ack...
	// (2) sending ack...
	// (3) sending ack...
	// done sending signal
	// done sending ack
	// done sending ack
	// done sending ack
	// exiting...
}

Conclusion:

Hope you enjoyed this. Leave me a comment if you build your own synchronization primitive.

Happy Hacking,

James


You can follow James on Mastodon for more frequent updates and other random thoughts.
You can follow James on Twitter for more frequent updates and other random thoughts.
You can support James on GitHub if you'd like to help sustain this kind of content.
You can support James on Patreon if you'd like to help sustain this kind of content.

May 20, 2018
845 words


Categories
Tags
devops fedora golang mgmtconfig planetdevops planetfedora sync

Links...


Comments

Nothing yet.


Post a comment



(sorry but the spammers were getting too crazy!)

Thank you

Your comment has been submitted and will be published if it gets approved.

Click here to see the patch you generated.

OK