It's good that they recognized percentiles can't be aggregated. They do seem to be losing some signal with the t-histogram implementation though, and are limited with their use of DataDog on how much precision they can obtain with their metrics collection.
It's good that they recognized percentiles can't be aggregated. They do seem to be losing some signal with the t-histogram implementation though, and are limited with their use of DataDog on how much precision they can obtain with their metrics collection.