[Libbeat] Flaky TestWatcherDie test

[ruflin]

Link: https://beats-ci.elastic.co/job/elastic+beats+master+multijob-windows/beat=libbeat,label=windows/1331/console
Platform: Windows

Error

>> go test: Unit Testing
11:34:30 FAILURES:
11:34:30 Error: go test failed: 1 test failures
11:34:30 Package: github.com/elastic/beats/libbeat/common/docker
11:34:30 Test:    TestWatcherDie
11:34:30 assertions.go:247: 
11:34:30                           
11:34:30 	Error Trace:	watcher_test.go:385
11:34:30 	
11:34:30 	Error:      	Not equal: 
11:34:30 	
11:34:30 	            	expected: 1
11:34:30 	
11:34:30 	            	actual  : 0
11:34:31 	
11:34:31 	Test:       	TestWatcherDie
11:34:31 assertions.go:247: 
11:34:31                           
11:34:31 	Error Trace:	watcher_test.go:385
11:34:31 	
11:34:31 	Error:      	Not equal: 
11:34:31 	
11:34:31 	            	expected: 1
11:34:31 	
11:34:31 	            	actual  : 0
11:34:31 	
11:34:31 	Test:       	TestWatcherDie
11:34:31 assertions.go:247: 
11:34:31                           
11:34:31 	Error Trace:	watcher_test.go:385
11:34:31 	
11:34:31 	Error:      	Not equal: 
11:34:31 	
11:34:31 	            	expected: 1
11:34:31 	
11:34:31 	            	actual  : 0
11:34:31 	
11:34:31 	Test:       	TestWatcherDie
11:34:31 assertions.go:247: 
11:34:31                           
11:34:31 	Error Trace:	watcher_test.go:385
11:34:31 	
11:34:31 	Error:      	Not equal: 
11:34:31 	
11:34:31 	            	expected: 1
11:34:31 	
11:34:31 	            	actual  : 0
11:34:31 	
11:34:31 	Test:       	TestWatcherDie
11:34:31 ----
11:34:31 SUMMARY:
11:34:31   Fail:     1
11:34:31   Skip:     6
11:34:31   Pass:     788
11:34:31   Packages: 70
11:34:31   Duration: 33.5826022s

[ph]

Adding more details here, This really look like a timing issues in the test suite.
I am not too familiar with the code there yet, but I see that we are using a mock, so I suspect we could instead use channel to manage the suite test flow.

	// Check it doesn't get removed while we request meta for the container
	for i := 0; i < 18; i++ {
		watcher.Container("0332dbd79e20")
		assert.Equal(t, 1, len(watcher.Containers()))
		time.Sleep(50 * time.Millisecond)
	}

	// Checks a max of 10s for the watcher containers to be updated
	for i := 0; i < 100; i++ {
		// Now it should get removed
		time.Sleep(100 * time.Millisecond)

		if len(watcher.Containers()) == 0 {
			break
		}
	}

[urso]

@jsoriano Test still fails quite often. The PR seems to be stalled for quite some time.

[andresrc]

Let's see whether this still applies or we should close it

[jsoriano]

I will take another look, maybe we can refactor the docker watcher to reuse common.Cache instead of having their own cache with cleanup logic.

[jsoriano]

I am going to skip these tests by now.