This is due to the fact that users delete their accounts but that doesn't do anything to change the count in the influx measurement, the ideal way to solve this is to just use the spark cluster to count listens and keep updating as we go.