Moving analytics from PostgreSQL to ClickHouse

21 August 2026 · 7 min read

A client's reporting database had grown to a few hundred million event rows. Dashboards that used to load in a second now took forty, and the nightly refresh was colliding with the morning.

What got faster

Aggregations over the full event history went from tens of seconds to well under one. Columnar storage and compression also cut disk usage by roughly a factor of eight.

What got harder

• Updates and deletes are expensive; we moved to append-only tables with periodic deduplication.
• Joins between large tables need more thought than in PostgreSQL.
• The team had to learn a new set of operational habits for merges and disk space.

When not to bother

If your largest table has fewer than about fifty million rows, a well-indexed PostgreSQL with materialised views is usually enough — and one database is always easier to run than two.