functional: sleep shortly to avoid errors in TestScheduleOneWayConflict

In TestScheduleOneWayConflict, after destroying conflicts-with-hello.service,
we need to sleep shortly to avoid occasional errors of
conflicts-with-hello.service being rescheduled even after being destroyed.
In that case, the conflicts unit remains active, while the original
hello.service remains inactive. Then the test TestScheduleOneWayConflict
fails at the end with a message "Incorrect unit started".
This error seems to occur frequently when enable_grpc turned on.

This sleep is just a workaround. We should find out the long-term solution.

Also note that, before commit 8cfc189 ("Functional tests: Fix race
condition in TestScheduleOneWayConflict()"), this error didn't occur,
because there was already a sleep for 5 seconds in place. However, the
commit 8cfc189 tried to fix a race using util.WaitForState(), also by
removing the original sleep. Actually the commit didn't completely fix
the race, so the hidden bug started to reappear.

Author: Dongsu Park

functional/scheduling_test.go

@@ -21,6 +21,7 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/coreos/fleet/functional/platform"
 	"github.com/coreos/fleet/functional/util"
@@ -284,6 +285,15 @@ func TestScheduleOneWayConflict(t *testing.T) {
 		t.Fatalf("Failed destroying %s", name)
 	}
 
+	// NOTE: we need to sleep here shortly to avoid occasional errors of
+	// conflicts-with-hello.service being rescheduled even after being destroyed.
+	// In that case, the conflicts unit remains active, while the original
+	// hello.service remains inactive. Then the test TestScheduleOneWayConflict
+	// fails at the end with a message "Incorrect unit started".
+	// This error seems to occur frequently when enable_grpc turned on.
+	// - dpark 20160615
+	time.Sleep(1 * time.Second)
+
 	// Wait for the destroyed unit to actually disappear
 	timeout, err := util.WaitForState(
 		func() bool {