run/tests: fix flaky RunAttachTermination test

During an attached `docker run`, the CLI starts capturing signals so
that they can be forwarded to the container. The CLI stops capturing
signals after container is no longer running/it's streams are closed.

This test was running into a race condition between signalling the
running command with a SIGINT and closing the mock container's streams –
If the signal syscall takes too long and the CLI realizes the streams
are closed and stops the signal handling, the interrupt isn't captured
and instead interrupts the test :')

Fix this by only closing the stream after the ContainerKill call was
made, which means the signal has already been captured.

Signed-off-by: Laura Brehm <[email protected]>

Author: laurazard

cli/command/container/run_test.go

@@ -202,15 +202,16 @@ func TestRunAttachTermination(t *testing.T) {
 	}
 
 	assert.NilError(t, syscall.Kill(syscall.Getpid(), syscall.SIGINT))
-	// end stream from "container" so that we'll detach
-	conn.Close()
 
 	select {
 	case <-killCh:
 	case <-time.After(5 * time.Second):
 		t.Fatal("containerKillFunc was not called before the timeout")
 	}
 
+	// end stream from "container" so that we'll detach
+	conn.Close()
+
 	select {
 	case cmdErr := <-cmdErrC:
 		assert.Equal(t, cmdErr, cli.StatusError{