t-digest is a data structure that will estimate a percentile point without having to store and order all the data points in a set. For example: to answer the question "What's the average latency for 99% of my database operations" we would have to store the average latency for every user, order the values, cut out the last 1% and only then find the average value of all the rest. This kind of process is costly not just in terms of the processing needed to order those values but also in terms of the space needed to store them. Those are precisely the problems t-digest solves.
0 commit comments