Flaky test: TestSimpleSpanProcessorShutdownHonorsContextCancel

[MrAlias]

Description

https://github.com/open-telemetry/opentelemetry-go/runs/3908587687?check_suite_focus=true

--- FAIL: TestSimpleSpanProcessorShutdownHonorsContextCancel (0.00s)
    simple_span_processor_test.go:165: SimpleSpanProcessor.Shutdown did not return context canceled, got <nil>
FAIL
FAIL	go.opentelemetry.io/otel/sdk/trace	0.424s
FAIL

opentelemetry-go/sdk/trace/simple_span_processor_test.go

Lines 159 to 167 in cbcd4b1

	func TestSimpleSpanProcessorShutdownHonorsContextCancel(t *testing.T) {
	ctx, cancel := context.WithCancel(context.Background())
	cancel()
	ssp := sdktrace.NewSimpleSpanProcessor(&testExporter{})
	if got, want := ssp.Shutdown(ctx), context.Canceled; !errors.Is(got, want) {
	t.Errorf("SimpleSpanProcessor.Shutdown did not return %v, got %v", want, got)
	}
	}

Environment

• OS: Microsoft Windows Server 2019 10.0.17763
• Architecture: 386
• Go Version: 1.17
• opentelemetry-go version: 790cf48

Test Failure Logs

log.txt

[MrAlias]

I'm able to reproduce this locally (Linux 5.14.8-arch1-1, Go 1.17):

docker run --rm -v "$PWD":/usr/src/go.opentelemetry.io/otel -w /usr/src/go.opentelemetry.io/otel/sdk/trace --env "GOARCH=386" golang:1.17 go test -run=TestSimpleSpanProcessorShutdownHonorsContextCancel -count 100000 -failfast ./...
go: downloading github.com/stretchr/testify v1.7.0
go: downloading github.com/google/go-cmp v0.5.6
go: downloading golang.org/x/sys v0.0.0-20210423185535-09eb48e85fd7
go: downloading github.com/pmezard/go-difflib v1.0.0
go: downloading github.com/davecgh/go-spew v1.1.0
go: downloading gopkg.in/yaml.v3 v3.0.0-20200313102051-9f266ea9e77c
--- FAIL: TestSimpleSpanProcessorShutdownHonorsContextCancel (0.00s)
    simple_span_processor_test.go:165: SimpleSpanProcessor.Shutdown did not return context canceled, got <nil>
FAIL
FAIL	go.opentelemetry.io/otel/sdk/trace	0.057s
ok  	go.opentelemetry.io/otel/sdk/trace/tracetest	0.522s [no tests to run]
FAIL

[MrAlias]

I'm able to reproduce without the GOARCH=386 (i.e. x86).

[MrAlias]

I'm able to reproduce on Go 1.15 and 1.16 as well.

[MrAlias]

This select statement looks to be the cause of the error:

opentelemetry-go/sdk/trace/simple_span_processor.go

Lines 91 to 95 in cbcd4b1

	select {
	case err = <-done:
	case <-ctx.Done():
	err = ctx.Err()
	}

Both cases are ready to be read and since the select order is random this will sporadically read that the exporter shutdown with no error instead of the already canceled context.