PostHog’s warehouse migration follows the limits of its workload

PostHog is rebuilding its warehouse around Trino and its own HogLake format after encountering resource sizing, metadata and connection-management problems with DuckDB and DuckLake. Its internal deployment held about two petabytes and a trillion rows. The account explains a specific cloud service’s scaling needs; it does not establish that DuckDB is unsuitable for smaller or local analytics.

PostHog · Benjamin Knofe-Vider · Nov 2, 2026

Why it matters

Engine selection includes workload isolation, metadata growth and client compatibility as well as query speed.

More context

PostHog says obsolete-table statistics dominated its catalog and slowed commits. Cleaning them helped temporarily, but the rows returned. It also describes the operational cost of supporting many PostgreSQL clients. The replacement is being tested internally and with design partners.

Keep in mind

This is the operator’s migration report, without a controlled cross-engine benchmark. Broad availability and pricing of the replacement are not established by the article.

Original read