OpenHouse vs Pinpoint

OpenHouse is LinkedIn's open-source Iceberg-native table catalog, running at production scale (6,000+ QPS against its HTS metadata service). The 0.5.x series ships multiple point releases per week. This cycle's work falls into three streams: production hardening (a critical outage from a mismatched functional-index predicate, now fixed with a 12,000x query speedup), architectural extensibility (a generic bounded post-commit operations framework), and groundwork for Iceberg view support (entity_type discriminator added to the HTS schema, ViewOperations extracted into a dedicated class).

OpenHouse is incrementally closing the gap between LinkedIn's internal deployment and the OSS version — the DDL baseline correction is the clearest sign: functional indexes that had existed in production for years were simply never documented, meaning anyone running the OSS version in Docker or local MySQL could not reproduce a whole class of production performance regressions. Alongside that, the post-commit operations framework signals a deliberate decoupling of catalog side effects (stats publishing, future observability hooks) from the commit path itself. Iceberg view support is now structurally half-in-place.

The next visible moves are the view-write path (the entity_type discriminator is deployed; a PR writing VIEW rows is explicitly the next step) and the commit-stats publisher landing on top of the post-commit framework. If the DDL parity effort continues, local dev environments may finally match production schema within a few cycles.