Has anyone that has done benchmarks for their Typesense implementation observed cold start times? From my testing, the average time is around 60ms for the entire network delay per search. However, I've observed cold starts of up to 300ms when doing the first search after the cluster is left idle for a few minutes. These cold start delays vary on the type of search query performed, simpler ones without filters tend to have cold start times of 100-200ms. My two questions are: Is this a known thing? If so, will this delay scale with a bigger dataset or is it just a constant delay?

Is this on Typesense Cloud or on your own machine?

We are using Typesense Cloud, but the client app (where the benchmark was performed) was deployed on my machine.

The cold start might happen it there are other processes and sometimes OS (especially mac) swaps memory to disk it process is idle.

I thought it could be something specific to how the OS handles processes or even network traffic but I just wanted to verify that it wasn't something else. I appreciate your insight on this, thanks!