The data will ultimately need to be downsampled.
However, the schema implemented so far doesn't lend itself well to downsampling with a Continous Query - there are transforms and pivots needed, so it probably makes as much sense to perform it with a set of Flux queries.
This ticket is being raised to start thinking about what those might look like
Activity
18-Dec-21 19:46
assigned to @btasker
18-Dec-21 19:50
When looking at historic data, it's all about aggregate figures.
So, for referrering domain we probably want to
n%and then allocate the rest to "Other" or somethingThat way we might show that referring domains were
We could go another way and convert to a tag, but the long-term cardinality ramifications of that are likely to be quite high. It would allow us to say
example.comhas referred 200 page views over the last year, but traffic comes from a hell of a lot of sources, so there's a big impact even when it's just domains18-Dec-21 19:52
I have a similar line of thought for
platform- it'd be easy for them to blow up over time, so again it'd be worth taking the top ten or similar.29-Dec-21 11:06
Forgot to reference this issue in commits, obviously having too much fun...
Commits are