[Flaky Test]: TestActionDispatcher/Dispatch_multiples_events_returns_one_error – Expected error

[rdner]

Failing test case

TestActionDispatcher/Dispatch_multiples_events_returns_one_error

Error message

dispatcher_test.go:393: Expected error

Build

https://buildkite.com/elastic/elastic-agent/builds/9615#0190110f-0b1c-45ba-81d9-c1406e137e09

OS

Mac

Stacktrace and notes

=== Failed
=== FAIL: internal/pkg/agent/application/dispatcher TestActionDispatcher/Dispatch_multiples_events_returns_one_error (0.52s)
    dispatcher_test.go:393: Expected error
=== FAIL: internal/pkg/agent/application/dispatcher TestActionDispatcher (1.52s)

[elasticmachine]

Pinging @elastic/elastic-agent (Team:Elastic-Agent)

[rdner]

Happens only on Mac, there are multiple occurrences, at least in 2 more builds:

• https://buildkite.com/elastic/elastic-agent/builds/9616#0190110f-a4fa-47fc-adcf-643d5325d414
• https://buildkite.com/elastic/elastic-agent/builds/9624#0190115a-d22e-4136-9723-54bbd5d63c2f

[rdner]

Failed again in https://buildkite.com/elastic/elastic-agent/builds/9709#01902a0b-20aa-433d-9a48-dce44c553af2

[pchila]

Given the long times for the unit tests run on Mac for the builds where It failed, I guess that the 200 ms sleeps we have are not enough (in a well designed unit test we should not rely on time.Sleep to make things work but it's at least marked as a workaround in code)

elastic-agent/internal/pkg/agent/application/dispatcher/dispatcher_test.go

Lines 384 to 400 in 1a05982

	// Kind of a dirty work around to test an error return.
	// launch in another routing and sleep to check if an error is generated
	dispatchCtx, cancelFn := context.WithCancel(context.Background())
	defer cancelFn()
	go d.Dispatch(dispatchCtx, detailsSetter, ack, action1, action2)
	time.Sleep(time.Millisecond * 200)
	select {
	case <-d.Errors():
	default:
	t.Fatal("Expected error")
	}
	time.Sleep(time.Millisecond * 200)
	select {
	case <-d.Errors():
	t.Fatal(err)
	default:
	}

These should probably replaced by longer timeouts in the select statements instead of the default case, even better if we could rely on a global unit test timeout of a few seconds (if we find one that is reliable enough without being too long) and block on the errors channel

[elasticmachine]

Pinging @elastic/elastic-agent-control-plane (Team:Elastic-Agent-Control-Plane)