Honest question: when dealing with time-series data, do you actually need every data point? Is that level of granularity really necessary? IMO, it makes way more sense to decide the aggregations you want ahead of time (e.g. "SELECT customer, sum(value) FROM purchases GROUP BY customer"). That way, you deal with substantially less data and everything becomes a whole lot simpler.
Really depends on the use case. Working in healthcare, vital signs can be modeled as time series points, but are lower frequency than, say, metrics from servers. However we want to store every point so a spike is not missed. One could argue an unsustained spike is noise, but in the healthcare domain there may be a correlation with some external event (the purpose is surprised and their heart rate spikes).
Re: Amazon Timestream – Fast, scalable, fully managed time series database
#31The clever thing to do in this scenario would be to keep every spike but delete all the data between similar data points after storing. So you get low granularity for identical/nearly-the-same data points and high granularity when something interesting happens. I don't have any experience with time-series data so maybe this is commonplace.