[beingflo]

I've been eyeing DuckDB for a metric collection hobby project. Quick benchmark showed promising query performance over SQLite (unsurprising considering DuckDB is column oriented), but quite a bit slower for inserts. Does anyone have experience using it as an "online" backend DB as opposed to a data analytics engine for interactive use? From what I gather they are trying to position themselves more in the latter use case.

[voidsnax]

Doing row-by-row inserts into DuckDB is really slow. Accumulating rows in an in-memory data structure and periodically batching them into something like an in-memory Arrow table, and then reading the Arrow table into DuckDB, is fast and has been tenable for my own use cases.

[pantsforbirds]

You can always use sqlite as your primary data store, and then directly query the sqlite database from duckdb whenever you need analytics.

[89vision]

Depends on the scale of users you expect for your project. Generally I like to keep oltp and olap tools in their lanes, but if < 100 people are going to be using it probably doesn't matter. I doubt duckdb has any sort of acid guarantees, so thats something to keep in mind.

[sudarshnachakra]

DuckDB does have ACID guarantees and transactions but I'd not be surprised if they are rarely used (if at all).

Ref: https://duckdb.org/docs/sql/statements/transactions

In the concurrency documentation they explicitly specify that it's not designed for lots of small transactions

Concurrency: https://duckdb.org/docs/connect/concurrency