Streaming event tracking but it's just postgresql, postgREST and pg_cron.

make bench will start the writer which sends randomly generated json events of faked "searches" with a category and a random query to the PostgREST api. Pretend that they're users submitting some tracking data. PostgREST writes them to the api.events table which is partitioned by minute. When a minute is over the searches are counted and inserted into agg.searches_per_minute. It does 3000 events per second with two writing threads on my laptop. It's fast and cpu-cheap because the aggregation query only has to read a single partiton (the last uncounted minute).

TODO:

• Figure out if the writer works properly with multiple threads, what's a goroutine?
• Roll up minutely counts into hourly, daily, weekly, monthly and total counts.
• Drop old minute partitions from api.events. Put them in a daily partitioned archive table maybe.