Apache Iceberg pushes reliability into the table format — snapshots, schema evolution, hidden partitioning. The catalog is still the place engines meet to agree on which table is which.
What goes wrong
Teams copy a metastore habit into a lakehouse world: one engine owns the catalog, others bolt on, and nobody owns backup, auth, or multi-environment promotion. REST catalogs help, but only if deployment and access patterns are designed on purpose.
What we optimize for
A catalog that Spark and SQL engines can share, with clear namespaces, environment boundaries, and an upgrade path that does not require a weekend of tribal knowledge.
Building this layer is core to 91crew services and to the product on our waitlist.