Cutting a MySQL table in half with one line
Andreas Bergström
We had a wide MySQL table — 8 million rows, 8 GB on disk, mostly index — about to grow 10× from a historical backfill. The instinct was to normalise: pull repeated strings into dimension tables. Classic relational hygiene. But twenty minutes spent measuring cardinality and average length per column pointed somewhere different. The biggest single column by raw bytes was a high-cardinality, near-unique click identifier from ad platforms. Normalising it would have moved the bytes around without rem