patterns: Trial on-demand pattern detection
benclive opened this issue · comments
In response to a few of the tickets, we'd like to try on-demand pattern generation for a particular query in order to better align with the ephemeral nature of patterns.
This would be a tradeoff between pre-computing and storing patterns, which uses a lot of CPU & mem continously, vs maybe spending a lot of CPU time regenerating the same patterns - but we need to experiment to see if this diference is significant & if we can effectively cache patterns somehow.
A few questions to answer:
- Whats the latency & resource use like?
- Would patterns be stable enough to come up with similar results when zooming in/out of a particular query? Does that matter?
- Do we need to store patterns at all?