We are trying to store keep counts for timeseries data and are looking at moving our time series data from redis to cassandra. The problem is that we store currency values as a float and the precision is in CPM which can be multiple decimal points. We have looked at using counter columns for keeping up with counts but this will not work for double/float values. Is it feasible to increment double values with some sort of accuracy or is this something that is not going to be possible? I understand there are issues with Cassandra and counters for possibly missing increments occasionally and some misses are tolerable but if this is not going to work for double/float values then we are going to have to look for some other solution.
We have been working with spark streaming for handling events coming from the ad exchanges and it seems be working great for whole number counts. We deal with around 5M+ events a day but this will scale to probably 5B+ events a day in the near future. We would like to have the counts be correct but there is a slight tolerance that would be acceptable. I am hoping someone has dealt with handling summary counts with C* that may be able to shed some light on do's and don'ts.
Thannks in advance