cmd/vet: add check for sync.WaitGroup abuse #18022

dsnet · 2016-11-22T20:45:25Z

The API for WaitGroup is very easy for people to misuse. The documentation for Add says:

calls to Add should execute before the statement creating the goroutine or other event to be waited for.

However, it is very common to see this incorrect pattern:

var wg sync.WaitGroup
defer wg.Wait()

go func() {
	wg.Add(1)
	defer wg.Done()
}()

This usage is fundamentally racy and probably does not do what the user wanted. Worse yet, is that it is not detectable by the race detector.

Since the above pattern is common, I propose that we add a method Go that essentially does the Add(1) and subsequent call to Done in the correct way. That is:

func (wg *WaitGroup) Go(f func()) {
	wg.Add(1)
	go func() {
		defer wg.Done()
		f()
	}()
}

The text was updated successfully, but these errors were encountered:

cznic · 2016-11-23T07:17:22Z

I don't like the idea. It's IMHO perhaps a task for go vet, if not implemented already. Also, the proposed method would add a(nother) closure layer when the go-ed function has parameters.

minux · 2016-11-23T08:11:12Z

I think this has been proposed before (probably on the mailing lists) and rejected.

dsnet · 2016-11-23T08:18:57Z

A go vet check seems pretty reasonable. I just tried it right now on the following:

func main() {
	var wg sync.WaitGroup
	defer wg.Wait()
	go func() {
		wg.Add(1)
		defer wg.Done()
	}()
}

and vet doesn't report anything.

In terms of vet's requirements:

Correctness: The problem is clearly a race.
Frequency: My gut feeling is that this is fairly common. A number of internal bugs that I've fixed is of this nature. So it subjective feels common enough. I don't have hard numbers.
Precision: I don't have an algorithm in mind that can accurately identify the pattern and I can imagine some number of false positives.

@dominikh , staticcheck does report a problem:
main.go:10:3: should call wg.Add(1) before starting the goroutine to avoid a race (SA2000)
I'm wondering how accurate the check is and whether it is worth adding to vet.

mvdan · 2016-11-23T10:21:17Z

As far as the proposal goes, I don't like how it limits the func signature to func().

dominikh · 2016-11-23T16:57:41Z

@dsnet The check in staticcheck has no (known) false positives. It shouldn't have a significant number of false negatives, either. The implementation is a simple pattern-based check, detecting go with function literals where the first statement is a call to wg.Add – this avoids flagging wg.Add calls further down the goroutine, which tend to be valid uses.

I'm -1 on the proposed Go function. I'd prefer not having to read code with an unnecessary level of nesting that looks callback-esque and reminds me of JavaScript.

mvdan · 2016-11-23T17:02:43Z

@dominikh I don't see any extra level of nesting here, though (assuming any func signature is allowed).

To be nitpicky, another thing that stands out from the proposal is how wg.Go() will create a goroutine even though go is never directly used by the user. I don't know if the standard library does this anywhere else, but I would prefer if it was left explicit.

dominikh · 2016-11-23T17:06:56Z

@mvdan The extra level of nesting would come from a predicted usage that looks something like this:

wg.Go(func() {
  // do stuff
})

as opposed to

go func() {
  // do stuff
}()

Admittedly the same level of indentation, but syntactically it's one extra level of nesting.

mvdan · 2016-11-23T17:10:05Z

Ah yes, I was thinking indentation there.

cznic · 2016-11-23T17:12:33Z

The problematic case is that

wg.Add(1)
go func(i int) { ... }(42)

// becomes

wg.Go(func() {
        go func(i int) { ... }(42)
})

mvdan · 2016-11-23T17:14:23Z

@cznic if you mean without the extra go, this would be solved if the restriction on the func() signature was removed.

dominikh · 2016-11-23T17:17:06Z

@mvdan Do you mean by allowing something like the following?

wg.Go(func(x, y int) { ...}, v1, v2)

IMHO that's way too much interface{} and not enough type safety.

mattn · 2016-11-23T17:18:13Z

panic is recovered?

mvdan · 2016-11-23T17:18:56Z

@dominikh true; I was simply pointing at the issue without contemplating a solution :)

rsc · 2016-11-28T21:18:57Z

The API change here has the problems identified above with argument evaluation. Also, in general the libraries do not typically phrase functionality in terms of callbacks. If we're going to start using callbacks broadly, that should be a separate decision (and not one to make today). For both these reasons, it seems like .Go is not a clear win.

It would be nice to have a vet check that we trust (no false positives). Perhaps it is enough to look for two statements

wg.Add(1)
defer wg.Done()

back to back and reject that always. Thoughts about how to make vet catch this reliably?

renannprado · 2016-12-04T04:28:11Z

I agree that vet is better place for this. The proposed API reminds of JavaScript, which will force us many times to wrap the code or function within a function with no arguments, while you could have just go func()....
Still nothing can stop you from creating such a helper methid, even though I don't see the need for it.

rsc · 2016-12-05T21:07:51Z

It sounds like we are deciding to make go vet check this and not add new API here. Any arguments against that?

dsnet · 2016-12-05T21:09:19Z

SGTM

rsc · 2020-08-05T17:36:24Z

I've added this proposal to the proposal process bin, but it's blocked on someone figuring out how to implement a useful check. Is anyone interested in doing that?

dominikh · 2020-08-06T08:04:16Z

Staticcheck has a fairly trivial check: for a GoStmt of a FuncLit, if the first statement in the FuncLit is a call to (*sync.WaitGroup).Add, we flag it. That has potential for false positives, but none have been reported in all the years that the check has existed.

The check could be trivially hardened by

looking for an immediately following defer of (*sync.WaitGroup).Done and comparing the two receivers.
checking that the argument to Add is 1, not some other number.

Edit: which is pretty much what you have suggested in #18022 (comment)

lpxz · 2024-12-06T16:20:36Z

since there is an active discussion here, I would like to bring to your attention that SA2000 in staticcheck misses real bugs.

If you run staticcheck against the following, it fails to find the bug. If you comment the fmt.Print in front of wg.Add(), it finds the bug. In general, as long as there is some code between go func(){ and wg.Add(), SA2000 fails to fire alarm.

// main.go
package main

import (
	"fmt"
	"sync"
)


func  somefunc() error {
   var  wg sync.WaitGroup
   go func() {
        fmt.Println("hi")
	wg.Add(1)
	defer wg.Done()
    }()
   return nil
}

func main(){
   fmt.Println("starting")
   somefunc()
}

$ go install honnef.co/go/tools/cmd/staticcheck@latest
$ staticcheck main.go

SA2000 code is self-explanatory on why this happens.

I wrote a linter to address this. Will move to staticcheck github repo to first discuss whether such improvement is welcome or negligible.

adonovan · 2024-12-06T16:31:24Z

SA2000 in staticcheck misses real bugs.

Yes, SA2000 and the port of it in https://go.dev/cl/633704 are intentionally very limited in the patterns they match, to avoid false positives. I suspect that relaxing the "wg.Add must be the first statement" rule would cause the analyzer to report a spurious diagnostic for this program:

func  f() error {
	var wg sync.WaitGroup
	wg.Add(1)
	go func() {
		defer wg.Done()

		...

		// Create another goroutine.
		wg.Add(1) // waitgroup: "WaitGroup.Add called from inside new goroutine"
		go func() {
			defer wg.Done()
			...
		}()
	}()
	return nil
}

I am sure there is room for improvement, but in general our inclination is to optimize for fewer false positives, even at the expense of true positives. If you have concrete suggestions (or code) for better algorithms, we can easily test them out on the corpus of the Module Mirror.

lpxz · 2024-12-06T17:44:56Z

@adonovan yes, you spotted a valid false positive.

Here is my linter that uses the golang analysis, not based on staticcheck. It also has fixing logic btw.

I scanned our code at Uber, it led to two false positives, which look exactly like what you described. All other reports are true positives.

It found one more true positive that SA2000 missed, which is arguably marginal :)

adonovan · 2024-12-10T22:28:35Z

Thanks @lpxz. I ran your linter on a sample of 16,193 modules from the Go Module Mirror corpus, and got 73 diagnostics (see attached file).

waitgroup.txt

A random sample of 10 are shown below. Of those, three are true positives that would not otherwise have been caught:

and zero are false positives (though I did notice false positives in a larger sample). That's good. Would you care to inspect a larger sample to classify them as (false, true existing, true new)? Thanks.

lpxz · 2024-12-10T23:03:44Z

@adonovan
thanks for spending efforts in trying it out!

yep, I can take a look.
could you clarify the difference between "true existing" vs "true new"?

Groxx · 2024-12-11T01:15:25Z

From skimming a couple as well (maybe I'll get interested enough to be exhaustive):

https://go-mod-viewer.appspot.com/golang.org/x/[email protected]/ssh/session_test.go#L678: Add and Done should not both exist inside the same goroutine block
https://go-mod-viewer.appspot.com/golang.org/x/[email protected]/ssh/session_test.go#L698: Add and Done should not both exist inside the same goroutine block
https://go-mod-viewer.appspot.com/golang.org/x/[email protected]/ssh/session_test.go#L871: Add and Done should not both exist inside the same goroutine block

These are false positives, the same exception @adonovan pointed out. They could also be written like this though, and I believe it'd incorrectly trigger SA2000 as well:

func  f() {
	var wg sync.WaitGroup
	wg.Add(1)
	go func() {
		wg.Add(1)
		defer wg.Done() // done setting up work
		...
		go func() {
			defer wg.Done() // done doing work
			...
		}()
	}()
}

but I suspect that's less common code in practice. "defer Done() first" is a very strong pattern as far as I can tell.

I've been trying to think of a check that'd be better at allowing these, because I like "same block" a fair bit, but the closest I can come up with is something like "Add(1) plus includes a deeper nested Done() == ignore". Basically "valid inside invalid is fine". It would definitely miss some, but it might be a good enough sign of "perhaps this is complex enough code that it can be ignored".

This one is a bit interesting:

https://go-mod-viewer.appspot.com/github.com/getlantern/[email protected]/eventual_test.go#L102: Add and Done should not both exist inside the same goroutine block

It's essentially inverted, and it's... arguably fine:

var wg sync.WaitGroup
wg.Add(1)
// Do some concurrent setting to make sure that it works
for i := 0; i < concurrency; i++ {
	go func() {
		// Wait for waitGroup so that all goroutines run at basically the same
		// time.
		wg.Wait()
		v.Set("hi")
		atomic.AddInt32(&sets, 1)
	}()
}
wg.Done()

I've used similar constructs to try to maximize racing, but I'm fairly sure this one is ineffective since in practice the Done() may run before any of the Wait()s. It's almost guaranteed with GOMAXPROCS=1 iirc. My pattern has been:

var ready, started, complete sync.WaitGroup
ready.Add(1)
started.Add(concurrency)
complete.Add(concurrency)
for i := 0; i < concurrency; i++ {
	go func() {	
		started.Done()        // inform that this goroutine has been started
		ready.Wait()          // wait for all to be started
		atomic.AddInt32(&sets, 1)
		complete.Done()       // this goroutine has finished its work	
	}()
}
started.Wait()  // wait for all goroutines to have been started, and likely at or near Wait
ready.Done()    // unblock all ~simultaneously
complete.Wait() // wait for them all to finish

which has so far been the most effective way to trigger real races that I've found, by quite a large margin, across various GOMAXPROCS. And doesn't take long to figure out when you see it.

I suspect this would also trigger the linter, on the ready waitgroup (which could easily be a close(chan) instead, and I generally do that in practice)

aclements · 2024-12-11T19:09:51Z

The proposal committee is on board with adding this check to vet. We'll let the domain experts hash out exactly what the right heuristic is. :)

lpxz · 2024-12-12T01:57:16Z

@adonovan @Groxx
given we only saw one false positive pattern for my linter, i can improve it to avoid that false positive.

Then we can rerun analysis again.

lpxz · 2024-12-12T16:34:08Z

@adonovan done with the manual analysis, see https://gist.github.com/lpxz/10b2cd9c3d390064f0a51267daa80575.

My linter has these:
FALSE: 12
TRUE EXISTING: 46,
TRUE NEW: 11, which cannot be found by SA2000 in staticcheck.

See the improved version here please.
It reduced the false positives from 12 to 0, without comprising true positives.

could you rerun it, and then I do another manual inspection?

adonovan · 2024-12-12T18:24:27Z

Impressive! Some of these new finds are really great.

But I'm not sure that all of the new positives are true: Consider this one:
https://go-mod-viewer.appspot.com/github.com/status-im/[email protected]/protocol/messenger_handler.go#L1762

m.shutdownWaitGroup.Add(1)
defer m.shutdownWaitGroup.Done()

I can't see the full context (and nor can the tool), but it looks like it is temporarily bumping up a reference count, which seems like a reasonable thing to do. I wonder how many are of this form, and if there's a rule to exclude them.

https://go-mod-viewer.appspot.com/github.com/getlantern/[email protected]/eventual_test.go#L102 is a delightful muddle! ;-)

lpxz · 2024-12-12T18:42:19Z

https://go-mod-viewer.appspot.com/github.com/getlantern/[email protected]/eventual_test.go#L102 is a delightful muddle! ;-)

ok, the improved linter will still false positive on this, until we use inspect() to recursively look into all sibling nodes (may not be worthy the machine cost)

But I'm not sure that all of the new positives are true: Consider this one:
https://go-mod-viewer.appspot.com/github.com/status-im/[email protected]/protocol/messenger_handler.go#L1762

m.shutdownWaitGroup.Add(1)
defer m.shutdownWaitGroup.Done()

I think it is true positive, given this pattern, the goroutine may not start yet, when the shutdownWaitGroup.Wait() is reached, leaving that goroutine leaky. Did I miss some subtlety here?

A relevant question, what is the next step then? :)

adonovan · 2024-12-12T18:52:46Z

https://go-mod-viewer.appspot.com/github.com/getlantern/[email protected]/eventual_test.go#L102 is a delightful muddle

To be clear, I meant that the code here is wrong in more ways than I can count; your linter algorithm is absolutely right to flag it.

https://go-mod-viewer.appspot.com/github.com/status-im/[email protected]/protocol/messenger_handler.go#L1762
m.shutdownWaitGroup.Add(1)
defer m.shutdownWaitGroup.Done()
I think it is true positive, given this pattern, the goroutine may not start yet, when the shutdownWaitGroup.Wait() is reached, leaving that goroutine leaky. Did I miss some subtlety here?

A WaitGroup is a counting semaphore. The most common use is to count unfinished goroutines, but in this case I think it is counting various "holds" that prevent the program from completing its shutdown (like electricians each padlocking the main breaker so that they know they are safe while they are working). That is, the goroutines are already running, and any of them may temporarily increment the counter, far from where they are created. Presumably at the end the main goroutine will wait for the counter to fall to zero.

The clue to recognizing this pattern is that there is no nearby Add; go func() { ... Done; } (); Wait structure (or bungled attempt at that structure).

A relevant question, what is the next step then? :)

We should probably reclassify the new positives in light of these observations. Ideally the false positive rate should be 5% or less.

Groxx · 2024-12-13T02:24:56Z

re https://go-mod-viewer.appspot.com/github.com/status-im/[email protected]/protocol/messenger_handler.go#L1762 :

Since it's a semaphore that is incorrect to use to Wait() and then Add() and nothing seems to guarantee that goroutine is scheduled before shutdown, I think that can be called "true". Nothing related to that goroutine is passed anywhere after it is created, nor any references to it (afaict) in that method that could lead to "wait for it to start m. downloadAndImportHistoryArchives before allowing shutdown".

(obviously this is worth verifying, but it seems extremely unlikely to be guaranteed-safe to me)

lpxz · 2024-12-13T04:23:44Z

I agree with @Groxx on this,

If the goroutines want to declare they are live, why do not they do so at the beginning or even better before goroutine starts. I am not sure why they do so in the middle.

By flagging this, we suggest developers to move add to an earlier point. This can only improve quality, at least will not degrade quality.

I may miss some points, please let me know.

egonelbre · 2024-12-13T09:04:54Z

But I'm not sure that all of the new positives are true: Consider this one: https://go-mod-viewer.appspot.com/github.com/status-im/[email protected]/protocol/messenger_handler.go#L1762

That's definitely a bug. (As other people already deduced). The easiest way to see this is to imagine a time.Sleep(10*time.Second) before line m.shutdownWaitGroup.Add(1) or task.Waiter.Add(1)... then any task.Waiter.Wait() or m.shutdownWaitGroup.Wait() may miss the first add and hence the wait is unnecessary.

adonovan · 2024-12-13T13:08:08Z

Thanks, you have all convinced me that this is indeed a bug (and thus a true positive of the new algorithm).

@lpxz, would you like to send me a CL to incorporate the new algorithm into the waitgroup analyzer (currently in gopls, but eventually to be promoted to vet too)? The CL should included a test for each distinct case you encountered (and false positives too, since they document the limitations). Thanks for improving this analyzer!

lpxz · 2024-12-13T13:57:00Z

@adonovan great, happy to make the contributions!

Will create a CL and add tests.
It may not be ready until early next January, let me know if this is more urgent.

Peng

aclements · 2025-01-16T00:54:59Z

Based on the discussion above, this proposal seems like a likely accept.

The proposal is to add a vet check for sync.WaitGroup misuse where Add can race with Wait, such as in:

var wg sync.WaitGroup
defer wg.Wait()

go func() {
        wg.Add(1)
        defer wg.Done()
}()

While complex heuristics are certainly possible here, for now we're going to limit this to checking for wg.Add on the first line of a closure. This has a zero false positive rate on the data we sampled, and is simple to implement.

lpxz · 2025-01-16T01:31:41Z

I am in progress of creating the cl, which has no false positive, fewer false negatives and is also simple.

Feel free to let me know if the final decision is to move on with the first line approach which is a fine approach).

adonovan · 2025-01-16T17:08:32Z

Feel free to let me know if the final decision is to move on with the first line approach which is a fine approach).

This proposal is about adding the basic functionality into vet, which can likely proceed for go1.25.

Independent of that, you should feel free to improve the algorithm; any improvements can be used immediately by gopls, and eventually by vet (without needing a separate proposal process), assuming they meet the usual criteria for precision and frequency.

aclements · 2025-01-23T03:00:54Z

No change in consensus, so accepted. 🎉
This issue now tracks the work of implementing the proposal.
— aclements for the proposal review group

The proposal is to add a vet check for sync.WaitGroup misuse where Add can race with Wait, such as in:

var wg sync.WaitGroup
defer wg.Wait()

go func() {
        wg.Add(1)
        defer wg.Done()
}()

While complex heuristics are certainly possible here, for now we're going to limit this to checking for wg.Add on the first line of a closure. This has a zero false positive rate on the data we sampled, and is simple to implement.

dsnet added the Proposal label Nov 22, 2016

dsnet added this to the Proposal milestone Nov 22, 2016

dsnet mentioned this issue Dec 2, 2016

testing: add T.Context() accessor #16221

Closed

dsnet changed the title ~~proposal: sync: add Go method to WaitGroup~~ cmd/vet: add check for sync.WaitGroup abuse Dec 5, 2016

dsnet removed the Proposal label Dec 9, 2016

dsnet mentioned this issue Nov 6, 2017

proposal: testing: add T.Context #18368

Closed

dsnet mentioned this issue Jan 24, 2018

sync: Proposal: Add a Go() function to sync.WaitGroup #23538

Closed

nhooyr mentioned this issue Mar 5, 2019

testing: add (*T).Deadline #28135

Closed

bcmills mentioned this issue Jun 26, 2020

proposal: sync: add WaitGroup.Go method #39863

Closed

rsc mentioned this issue Aug 5, 2020

proposal: review meeting minutes #33502

Open

aclements moved this from Active to Likely Accept in Proposals Jan 16, 2025

aclements added the Proposal-FinalCommentPeriod label Jan 16, 2025

aclements moved this from Likely Accept to Accepted in Proposals Jan 23, 2025

aclements changed the title ~~proposal: cmd/vet: add check for sync.WaitGroup abuse~~ cmd/vet: add check for sync.WaitGroup abuse Jan 23, 2025

aclements modified the milestones: Proposal, Backlog Jan 23, 2025

aclements added Proposal-Accepted and removed Proposal-FinalCommentPeriod labels Jan 23, 2025

dsnet mentioned this issue Jan 27, 2025

proposal: sync/v2: new package #71076

Open

cmd/vet: add check for sync.WaitGroup abuse #18022

cmd/vet: add check for sync.WaitGroup abuse #18022

Comments

dsnet commented Nov 22, 2016 • edited Loading

cznic commented Nov 23, 2016

minux commented Nov 23, 2016 via email

dsnet commented Nov 23, 2016

mvdan commented Nov 23, 2016

dominikh commented Nov 23, 2016

mvdan commented Nov 23, 2016

dominikh commented Nov 23, 2016

mvdan commented Nov 23, 2016

cznic commented Nov 23, 2016

mvdan commented Nov 23, 2016

dominikh commented Nov 23, 2016

mattn commented Nov 23, 2016

mvdan commented Nov 23, 2016

rsc commented Nov 28, 2016

renannprado commented Dec 4, 2016

rsc commented Dec 5, 2016

dsnet commented Dec 5, 2016

rsc commented Aug 5, 2020

dominikh commented Aug 6, 2020 • edited Loading

lpxz commented Dec 6, 2024 • edited Loading

adonovan commented Dec 6, 2024 • edited Loading

lpxz commented Dec 6, 2024

adonovan commented Dec 10, 2024

lpxz commented Dec 10, 2024

Groxx commented Dec 11, 2024 • edited Loading

aclements commented Dec 11, 2024

lpxz commented Dec 12, 2024

lpxz commented Dec 12, 2024 • edited Loading

adonovan commented Dec 12, 2024

lpxz commented Dec 12, 2024

adonovan commented Dec 12, 2024 • edited Loading

Groxx commented Dec 13, 2024 • edited Loading

lpxz commented Dec 13, 2024 • edited Loading

egonelbre commented Dec 13, 2024

adonovan commented Dec 13, 2024 • edited Loading

lpxz commented Dec 13, 2024

aclements commented Jan 16, 2025

lpxz commented Jan 16, 2025

adonovan commented Jan 16, 2025

aclements commented Jan 23, 2025

dsnet commented Nov 22, 2016 •

edited

Loading

dominikh commented Aug 6, 2020 •

edited

Loading

lpxz commented Dec 6, 2024 •

edited

Loading

adonovan commented Dec 6, 2024 •

edited

Loading

Groxx commented Dec 11, 2024 •

edited

Loading

lpxz commented Dec 12, 2024 •

edited

Loading

adonovan commented Dec 12, 2024 •

edited

Loading

Groxx commented Dec 13, 2024 •

edited

Loading

lpxz commented Dec 13, 2024 •

edited

Loading

adonovan commented Dec 13, 2024 •

edited

Loading