timmyyuan/gobench

[Data Race]: potential new initroduced data race in Serving#3068

yanke-xu opened this issue · 5 comments

A new data race in serving#3068:

WARNING: DATA RACE
Read at 0x00c0000701f0 by goroutine 9:
  runtime.chansend()
      /usr/local/go/src/runtime/chan.go:158 +0x0
  command-line-arguments.(*impl).Go()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:44 +0x84
  command-line-arguments.TestServing3068.func1()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:65 +0x4b

Previous write at 0x00c0000701f0 by goroutine 7:
  runtime.closechan()
      /usr/local/go/src/runtime/chan.go:355 +0x0
  command-line-arguments.(*impl).Wait.func1()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:49 +0x64
  sync.(*Once).doSlow()
      /usr/local/go/src/sync/once.go:68 +0x109
  sync.(*Once).Do()
      /usr/local/go/src/sync/once.go:59 +0x68
  command-line-arguments.(*impl).Wait()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:48 +0x64
  command-line-arguments.TestServing3068()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:72 +0x13b
  testing.tRunner()
      /usr/local/go/src/testing/testing.go:1203 +0x202

Goroutine 9 (running) created at:
  command-line-arguments.TestServing3068()
      /home/ubuntu2004/goProgram/src/GTrack/testdata/nonblocking/serving/3068/serving3068_test.go:63 +0x127
  testing.tRunner()
      /usr/local/go/src/testing/testing.go:1203 +0x202

Goroutine 7 (running) created at:
  testing.(*T).Run()
      /usr/local/go/src/testing/testing.go:1248 +0x5d7
  testing.runTests.func1()
      /usr/local/go/src/testing/testing.go:1521 +0xa6
  testing.tRunner()
      /usr/local/go/src/testing/testing.go:1203 +0x202
  testing.runTests()
      /usr/local/go/src/testing/testing.go:1519 +0x612
  testing.(*M).Run()
      /usr/local/go/src/testing/testing.go:1427 +0x3b3
  main.main()
      _testmain.go:43 +0x236

In addition, since the Goroutine 9 and Goroutine 7 are called in parallel, the two statements i.wg.Add(1) and i.wg.Wait() will be called in parallel, it can cause a data race.

As your backtrace discribes, goroutine 9 accesses runtime.chansend and goroutine 7 accesses runtime.closechan. This depends on how Go runtime reports the panic/race behavior caused by send on closed channel. So I think it's the same bug, but different behavior on different versions of Go runtime.

OS:Ubuntu 20.04
Go Version:1.16.15

In fact,when I use go test -race, both panic and data race are reported by the Go-runtime detector at the same time.

This is to be expected, GoBench was not tested on Go 1.16 in my impression. You can refer the release notes of Go 1.16:

Go 1.16 fixes a discrepancy between the race detector and the Go memory model. The race detector now more precisely follows the channel synchronization rules of the memory model. As a result, the detector may now report races it previously missed.

If I understand correctly, Go race detector has been enhanced in this release so it can report the race you mentioned. But I prefer that the original bug triggers the new bug.

Yes. Perhaps it would be better to update the backtrace or explain the experimental environment to reduce this divergence.

I truly appreciate it. I'll put a link to this issue in the documentation so that everyone can see what we've discussed.