What changed
OpenMetadata shipped 2.0 on 24 August 2026. Three changes show up the moment you open it:
- The profiler no longer scans 100% of rows. It now defaults to dynamic sampling, which cuts query cost and run time noticeably on large tables.
- Cardinality distribution is no longer collected on every run. If anything of yours relied on distinct-value counts, you have to ask for it explicitly in the profiler configuration.
- Knowledge Center is now Context Center. Existing pages migrate on their own as Articles, plus Documents (freeform text attached to an asset or a team) and a dashboard view to manage all of it.
Governance also gains custom intake forms — for asset certification, ownership assignment or classification — and OWL import, which loads existing ontologies and taxonomies instead of retyping them.
What it means if you are learning
The profiler change is the one that changes your day. Until now, pointing the profiler at a genuinely large table was expensive: it scanned everything. With sampling by default you can try it on real data without fearing the bill — which is exactly the practice you need to understand what a data profile is telling you.
The price is that a sampled profile is an estimate, not a census. For distributions and nulls that's plenty; for claiming "this column has no duplicates", it isn't. Knowing which of those two questions you're asking is the skill, and this version forces you to tell them apart.
Who it doesn't affect
- If you already had your own profiler configuration, nothing changes. The release says so explicitly: existing configs are preserved, and only pipelines using the default agent config are affected.
- If you use OpenMetadata only for catalogue and lineage — the common case when you're starting — none of this touches you.
- If you connect Databricks Pipeline, do read the release: authentication moved from a
bare
tokenfield to anauthTypeobject supporting PAT, Databricks OAuth and Azure AD. Stored configs migrate automatically.