QoS mechanisms kick in only when there is congestion

Any configured QoS mechanisms on a network device will only "kick in" if there is congestion on the particular interface on which the mechanisms is enabled. When traffic arrives at a port at a rate less than the port's speed, the traffic is served immediately and never queued. Thus, no QoS mechanisms can apply.

Once the arriving traffic exceeds the speed of the port, queuing occurs. Once you have a queue that is non-zero in size, then the QoS mechanisms can be applied to that.

Now mechanisms such as WRED for example, are called congestion avoidance mechanisms, and this term can become confusing. However, strictly speaking, WRED is a queue exhaustion avoidance mechanism. More on that can be found at QoS - is WRED really a congestion avoidance mechanism.