Forgive a bit of ignorance, it's been a bit since I've touched Go, but this looks awfully similar to a Java CountdownLatch [1]. Is this just a glorified Go port of that or am I missing something vital here?
You have to explicitly propagate the group's context if you want it to cancel. You can just not do that if you don't want - there certainly are cases for that.
It does not, which is easy to verify from the source. Every func passed in is always run (with the exception of TryGo which is explicitly "maybe").
At best, using the optional, higher-effort errgroup.WithContext will cancel the context but still run all of your funcs. If you don't want that for one of the funcs, or some component of them, just don't use the context.
If the context cancellation is undesirable, you just choose not to use WithContext, as the sibling comment mentions.
You could also just make your subtask function return nil always, if you just want to get the automatic bookkeeping call pattern (like WaitGroup.Go from Golang 1.25), plus optional concurrency limiting.
Also note, even if a subtask function returns an error, the errgroup Wait blocking semantics are identical to those of a WaitGroup. Wait will return the first error when it returns, but it doesn't unblock early on first error.
They basically don't backport anything for Go, but the quid pro quo for that is that the backwards compatibility is pretty strong so upgrades should be safe. I have seen one serious issue from it, but still it's the language I'm the most confident to do an upgrade and expect things to Just Work afterwards.
Love this. Majority of concurrency in a usual web service is implemented using waitgroups IME (see below) This will greatly simplify it.
var wg sync.WaitGroup
wg.Add(1)
go func(){
callService1(inputs, outParameter)
wg.Done()
}
// Repeat for services 2 through N
wg.Wait()
// Combine all outputs
BTW, this can already be done with a wrapper type
type WaitGroup struct { sync.WaitGroup }
func (wg *WaitGroup) Go(fn func()) {
wg.Add(1)
go func() {
fn()
wg.Done()
}()
}
Since you're doing struct embedding you can call methods of sync.WaitGroup on the new WaitGroup type as well.
I like WaitGroup as a concept, but I often end up using a channel instead for clearer error handling. Something like:
errCh := make(chan error)
for _, url := range urls {
go func(url string){
errCh <- http.Get(url)
}(url)
}
for range urls {
err := <-errCh
if err != nil {
// handle error
}
}
Should I be using WaitGroup instead? If I do, don't I still need an error channel anyway—in which case it feels redundant? Or am I thinking about this wrong? I rarely encounter concurrency situations that the above pattern doesn't seem sufficient for.
buffered channels won't help here. That's just how many results can be buffered before the remaining results can be added to the channel. It doesn't wait until all of them are done before returning a result to the consumer.
> It doesn't wait until all of them are done before returning a result to the consumer.
Right, but it prevents goroutine leaks. In these situations I'm usually fine with bailing on the first error, but I grant that's not always desirable. If it's not, I would collect and join errors and return those along with partial results (if those are useful).
multiple ways of doing something isn't inherently bad.
For example, if you want to set a variable to the number of seconds in seven hours, you could just set the variable to 25200, or you could set it to 60 * 60 * 7. The expanded version might be clearer in the code context, but in the end they do exactly the same thing.
zero value. container-agnostic initialization. say your type is not struct anymore, you would not have to change the way you intialize it. what you care here is zero value, and let the type figure out that it is zero and use methods appropriately. and it is just more clean this way
That is a much better argument than saying it is "more clean", which doesn't mean anything. I don't necessarily agree, because I don't think zero values are a good feature of the language, and even if they were this is a completely trivial case. But at least I don't have to work out what "cleanliness" is.
This means you can't pass variables in as function arguments. Even the example in the official go docs doesn't handle the scope correctly:
func main() {
var wg sync.WaitGroup
var urls = []string{
"http://www.golang.org/",
"http://www.google.com/",
"http://www.example.com/",
}
for _, url := range urls {
// Launch a goroutine to fetch the URL.
wg.Go(func() {
// Fetch the URL.
http.Get(url)
})
}
// Wait for all HTTP fetches to complete.
wg.Wait()
}
[1] https://docs.oracle.com/javase/8/docs/api/java/util/concurre...
[0] https://docs.oracle.com/javase/7/docs/api/java/util/concurre...
What is quite sad is that we cannot add it ourselves as it's so simple of what they have done:
errgroup also has other niceties like error propagation, context cancellation, and concurrency limiting.
At best, using the optional, higher-effort errgroup.WithContext will cancel the context but still run all of your funcs. If you don't want that for one of the funcs, or some component of them, just don't use the context.
You could also just make your subtask function return nil always, if you just want to get the automatic bookkeeping call pattern (like WaitGroup.Go from Golang 1.25), plus optional concurrency limiting.
Also note, even if a subtask function returns an error, the errgroup Wait blocking semantics are identical to those of a WaitGroup. Wait will return the first error when it returns, but it doesn't unblock early on first error.
But channels already do the waiting part for you.
Right, but it prevents goroutine leaks. In these situations I'm usually fine with bailing on the first error, but I grant that's not always desirable. If it's not, I would collect and join errors and return those along with partial results (if those are useful).
just `var wg sync.WaitGroup`, it is cleaner this way
For example, if you want to set a variable to the number of seconds in seven hours, you could just set the variable to 25200, or you could set it to 60 * 60 * 7. The expanded version might be clearer in the code context, but in the end they do exactly the same thing.
dont you mean a week? 60 x 60 x 24 x 7 ?
or at lest 8 hours?
7 hours is just odd
cool it down a little. touch some grass. and hopefully you will see beauty in Go zero-values :P
here is google guideline: https://google.github.io/styleguide/go/best-practices#declar...
You need to use this pattern instead:
https://go.dev/blog/loopvar-preview
Well, you could...
> You need to use this pattern insteadWhy? Seems rather redundant. It is not like WaitGroup.Go exists in earlier versions.