A detailed comparison of OpenMetadata, DataHub, and Amundsen against each other and commercial platforms (Decube, Atlan, Alation, Collibra), covering the criteria that actually differentiate them: connector maintenance, column-level lineage, metadata model flexibility, and operating burden. The piece argues that open source catalogs are genuinely capable software and the right choice in several scenarios, but that the hidden cost is engineer time (0.25–1.0 FTE), not the licence. It identifies four costs that only surface at month six (upgrades, broken connectors, key-person departure, audit evidence requests), explains what financial regulators like OJK, APRA, MAS, and NAIC specifically require, and provides a decision table routing teams to the right option based on team size, platform count, and compliance obligations.
Table of contents
Key TakeawaysWhat This Comparison Is ForThe Three Projects in One Paragraph EachHead to Head on What Actually DiffersWhere Open Source Is the Right AnswerThe Costs That Appear at Month SixTotal Cost of Ownership, Counted HonestlyWhat Commercial Platforms Add, Stated PlainlyOpen Source and Commercial Side by SideWhat Regulators Ask For That Self Hosting Leaves With YouThe Decision TableIf You Do Move, Move DeliberatelyFour Mistakes That Cost the MostFrequently Asked QuestionsQuestions this post answers
What is the real ongoing cost of self-hosting an open source data catalog like OpenMetadata or DataHub?
Self-hosting an open source data catalog costs roughly a quarter to a full platform engineer in ongoing time, plus infrastructure. This covers the upgrade cycle, connector repair when source systems change, identity configuration, and internal requests. The licence is free, but that engineer fraction is the cost that decides most business cases — and it should be compared against a commercial quote, not against zero. Teams weighing this trade-off track real-world data platform costs on daily.dev.
Does OpenMetadata or DataHub support column-level lineage, and how does it work?
Both OpenMetadata and DataHub produce genuine column-level lineage parsed from query logs on major cloud warehouses — Snowflake, BigQuery, and Databricks. This is not declared lineage but parsed lineage, making it accurate enough that lineage alone is not a reason to buy a commercial product. Coverage degrades to table-level on less common and legacy sources, which is also true of commercial platforms. Amundsen does not provide column-level lineage. Developers choosing between these tools for lineage use daily.dev to follow how each project's connector coverage evolves.
When should a team migrate from a self-hosted open source data catalog to a commercial platform?
The usual trigger is a compliance obligation, not missing features. When an external party — auditor, regulator, or enterprise customer — starts requiring retained approval history, immutable access records, and named accountable owners on a recurring basis, self-hosting stops being cheaper. Other triggers are connector coverage gaps on legacy sources and the departure of the single engineer who understood the deployment. A single-warehouse, engineering-led team with no regulator asking should stay on open source. Data teams navigating this decision follow catalog and governance developments on daily.dev.