All you need is PostgreSQL
53 points by schonfinkel
53 points by schonfinkel
…and ClickHouse.
Why so? I know nothing about it.
I use Clickhouse at work. It's very nice for handling large amount of data (i have hundred of billions of lines and i can query them in minutes to find patterns on hardware from 2012), but what i really like are :
stored_procedures_everywhere.gif
On the serious note, I expected to learn how Postgres' rich type set allows to store almost anything, or how its locking primitives remove the need in external lock system coordination, or how different isolation levels cover for a spectrum of concurrency heavy apps, or how to design and tune a table to allow cheap write heavy storage, or how Postgres CDC plugins enable transparent heterogeneous replication, or how embedded scripting allows to pull the client-server app inside out where latency is critical, or how… stop-token.
That OLAP approach is pretty interesting. Basically you materialize the ledger sums the application needs ahead of time with triggers. One nice thing about an OLAP database (or something like DuckDB querying parquets) is you can run whatever stats you want on your data columns, relatively quickly, without needing to anticipate what intermediate data you'll need to compute it. This flexibility is valuable for business reporting and other analytical purposes.
ad hoc analytical queries is one of the few places where I advise people to reach for something else. And that would be DuckDB.