August 4, 2026

Quack, crunch, and comment-war

DuckDB – Data power tools for your laptop, now in Clojure (2023)

Your laptop just got a data superhero, and the comments are already fighting over sidekicks

TLDR: DuckDB’s new Clojure integration means huge datasets can be handled locally on a laptop far more easily, including a 50GB file demo with 400 million rows. The comments turned it into a showdown over rival tools, self-hosting dreams, and one gloriously chaotic claim that AI has made programming languages irrelevant.

A very nerdy software update somehow turned into a full-on comment section talent show. The big news: the Clojure world can now tap into DuckDB, a super-fast local database that lets people crunch massive files on a regular laptop instead of running off to expensive server setups. In the demo, a gigantic 50GB transaction file with 400 million rows gets pulled in surprisingly fast and shrunk to 18GB, which had readers reacting like they’d just watched a compact car tow a cruise ship.

But the real action was in the replies, where admiration, rivalry, and a little AI chaos all broke out. One longtime fan praised tmducken but immediately name-dropped a newer alternative, ducktape, in a classic “love you, but I’m seeing someone else” moment. Another commenter went bigger, declaring they’re ditching ClickHouse and betting on Parquet + DuckDB for the future, with a dramatic anti-big-cloud rant about not sending sensitive customer data into a giant shared SaaS machine. That got the self-hosting crowd fired up.

Meanwhile, the DuckDB stans were gushing. One person called the command line tool a “powerhouse” for slurping up weird compressed log files like it’s no big deal. Another basically wrote a love letter, upgrading the project from merely “honorable” to “extremely, most excellent.” And then, right on cue, an AI maximalist barged in with the hottest take of all: languages don’t matter anymore because LLMs generate all their queries. Subtle? No. Memorable? Absolutely.

Key Points

  • The article argues that tech.ml.dataset works well in memory but becomes unwieldy for very large relational datasets such as around 100GB of CSV files.
  • Earlier use of JDBC with Postgres allowed out-of-memory querying, but row-to-column conversion through JDBC was inefficient and non-batched for TMD workflows.
  • DuckDB was first integrated with tmducken through C bindings in late 2021, but early support required query results to fit in memory and lacked a high-performance append or insert path.
  • DuckDB later added batched inserts and batched querying through its C interface, which the article says enables processing very large joins from Clojure through TMD.
  • In the article’s example, a 50GB CSV with 400,000,000 transaction rows was loaded into DuckDB in about 1 minute 50 seconds, producing an 18GB database file that included automatically created indexes.

Hottest takes

"we’re leaving it behind for our next product and building directly on Parquet and DuckDB" — encoderer
"Duckdb CLI is a powerhouse" — eterm
"llm has solved this abstraction 'perfectly'" — solarized
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.