[henridf]

Makes sense. How do you deal with data arriving out-of-order? (assuming your data is timestamped at the source)

[mike_heffner]

We grab a timestamp when data hits our API (largely as close to source as we can get) and use that to order time samples. For time series the ordering requirements aren't as strict as other problem domains, since we can usually go back and overwrite previous time series data points.