Why could a spike itself not always be good news? Why is it so important to find the relationships between time series metrics at scale?