4 ms·
About 500 rps max? That seems pretty low, I can easily do something like 25k rps with 1 cpu using clickhouse for the same scenario. Am I missing something?
by js4ever 4y ago
About 500 rps max? That seems pretty low, I can easily do something like 25k rps with 1 cpu using clickhouse for the same scenario. Am I missing something?
- hardwaresofton 4y agoNope you're not missing anything -- sometimes postgres isn't the absolute best solution to a problem. Part of the setup was to get to that "v1" as fast as possible, and what the imaginary team member already had available was Postgres. Even with that setup I made sure to leave this in there: > (It will be a recurring theme but this is a spot where we probably don't necessarily want to use stock Postgres but instead want to use tools like Citus Columnar Storage, ZedStore, or an external choice like ClickHouse). Also note that there are some optimizations that are not discussed here -- an unlogged table for the HLL itself, for example, triggered/delayed computation, etc. In general, Postgres isn't the greatest for OLAP workloads, but it is quite flexible. The code is open though: https://github.com/VADOSWARE/supabase-seen-by https://github.com/VADOSWARE/supabase-seen-by Lots to improve upon there, but I tried to make sure the code was very easy to understand and easy to extend -- would love to have a Clickhouse version :)