Ask HN: Distributed SQL engine for ultra-wide tables

A bold poster claims they flipped databases sideways: no joins, no transactions, columns spread across machines, and blazing-fast reads from a table with up to a million columns. Cue the Hacker News popcorn. The first reaction? Skeptical applause-turned-interrogation: “What is the design?” asked icsa, echoing a chorus of readers who want more than vibes and benchmarks. Others demanded to know why anyone would avoid joins—database speak for stitching tables together—with remywang asking, “Why don’t you want joins?” The mood: impressed by the numbers, but suspicious of the missing details.

Another camp showed up wielding receipts: “ClickHouse and Scuba address this,” minitoar wrote, pointing to systems that only read the columns you ask for. kentm suggested modern “lakehouse” formats—think Parquet files with a smart tracker—like Iceberg, and engines like Trino/Presto, while warning that metadata might melt down at extreme width. Meanwhile, mamcx went full inventor-mindset, dreaming of custom data types to compress columns and a better way to build storage. The comment section turned into a tug-of-war: pragmatists saying “use existing tools,” purists asking for architecture specifics, and comedians dubbing this the “spreadsheet from hell” and “social distancing for data—no joins allowed.” Verdict: huge promise, but the crowd wants the blueprint before the bravo.