Platform · 2019–now · Platform Wars
DuckDB
SQLite's cousin for analytics. Columnar SQL in the same process as the script — no cluster to feed.
Why DuckDB stuck
DuckDB stuck because a lot of "big data" was a file wearing a Hadoop costume. In-process OLAP lets a notebook or a pipeline query columnar data without standing up Spark. It did not replace the warehouse, and download counts are not a migration rate. The fad it punctured was the cluster as a default. The residue is a library you embed, then leave when concurrency and governance outgrow one process.
Sources
Compare with
Related
Platform · 1974–now
SQL / Relational
Declarative data that outlived every ORM fashion cycle. Postgres and friends keep winning by being boring.
Platform · 1996–now
PostgreSQL
The open-source relational default — JSON, extensions, and enough correctness to survive every NoSQL wave.
Platform · 2014–now
Apache Spark
In-memory data processing that made Hadoop feel overnight. Still the batch/stream workhorse under many lakehouses.
Platform · 2006–2018
Hadoop Everywhere
MapReduce as lifestyle. The elephant in rooms that needed a spreadsheet.
$ Enterprises stood up Hadoop clusters for gigabytes of data that fit on one Postgres instance. Hadoop admins, ZooKeeper nightmares, and ETL rewrites burned millions before Spark and cloud warehouses retired the elephant.