Select the Correct Oracle Access Pattern

Create a copy of this file for each page in your publication, then add it to your table of contents.

Requirement	Use	Read next
Query supported files through a stable database object.	Persistent external table	Chapter 5
Query supported files for one statement without creating an object.	Inline external table	Chapter 5
Query an existing Iceberg table from metadata or a catalog.	Iceberg read access	Chapter 4
Discover and query many catalog-managed tables or file-backed datasets, including files described by AWS Glue metadata.	Mounted catalog	Chapter 3, Chapter 4, and Chapter 5
Create and append to an Iceberg table from Oracle SQL.	Native Iceberg table	Chapter 4
Use database-managed DML, indexes, constraints, or consistently low-latency serving.	Load or materialize data in Oracle AI Database	Chapter 6

Separate Metadata Access from Data Access

A catalog can make a table discoverable without granting access to its files. Catalog-backed queries commonly use one identity for the catalog endpoint and another identity for object storage. Both paths must succeed.

Complete the common connectivity, identity, and privilege model in Chapter 2. Later chapters reference that setup instead of repeating it.

Respect Feature and Deployment Boundaries

Oracle Database 26ai lakehouse capabilities are version-sensitive. The presence of an open-format feature in the Apache Iceberg specification or in another query engine does not establish support in a particular Oracle access path. Verify the exact database release update, deployment, package, catalog implementation, storage combination, and documented restriction before relying on a feature.

Using the Enterprise and Operational Catalogs for Different Jobs

AI Catalog provides the enterprise catalog-of-catalogs for discovery and governance across systems. Operational catalogs remain authoritative for table names, namespaces, snapshots, and commits. DBMS_CATALOG mounts a supported operational catalog into one database; it is not itself the enterprise catalog. See Chapter 3.

Design for Remote I/O

Lakehouse performance is primarily a remote-I/O problem. Read fewer files and columns, prune early, maintain accurate statistics and metadata, avoid small-file proliferation, and cache or materialize only when the workload justifies it. See Chapter 6.