Catalog federation and Oracle SQL intelligence for shared open tables

Oracle is expanding native Apache Iceberg support in Oracle AI Lakehouse. Customers can read shared Iceberg tables, combine them with operational Oracle data, and publish Oracle SQL query results as new Iceberg tables for use across compatible engines.

These capabilities reflect Oracle’s commitment to open standards and customer choice. Data remains in the lake in an open format rather than becoming tied to a single engine. Customers can apply Oracle SQL intelligence to it and retain the freedom to work with other supported platforms.

Four advances define this announcement:

  • Seamless catalog federation. Connect AI Lakehouse to Oracle AI Data Catalog or a supported external catalog and discover its authorized tables without a separate setup step for each one.
  • Broader Iceberg read compatibility. Expanded feature coverage lets AI Lakehouse read a wider range of shared Iceberg tables created and managed by other engines.
  • Native Iceberg writing. Create Iceberg tables and publish the results of Oracle SQL queries, including joins with Oracle data, directly to them.
  • Intelligent performance. Iceberg tables are first-class citizens of the Oracle optimizer, bringing Oracle’s query-optimization capabilities to Iceberg workloads. Iceberg metadata avoids unnecessary reads, Lake Cache can deliver performance comparable to reading from Exadata storage for cached Iceberg data, and Data Lake Accelerator scales I/O and compute for large scans.

One open table format for multiple engines

A modern lakehouse rarely uses a single engine or cloud. A company may prepare data in Databricks, manage data products in Snowflake, use AWS Glue for its lake, and run analytics in AI Lakehouse. Iceberg lets compatible engines work with shared tables instead of requiring a separate export for each one.

The data files stay in object storage. Iceberg metadata describes the table’s schema, files, and snapshots; the catalog provides table names and the locations of that metadata. AI Lakehouse can query these tables without first loading them into Oracle tables.

Oracle AI Lakehouse and compatible engines access shared Apache Iceberg tables in object storage.
AI Lakehouse participates as both a reader and a writer over shared Iceberg data. Capabilities depend on each engine and configuration.

Connect a catalog once and discover its tables

Iceberg tables do not have to belong to a single query engine. Customers can use Oracle AI Data Catalog or a supported external catalog to manage their shared tables. Data Studio provides a visual workspace for discovering and working with that data, while AI Lakehouse provides Oracle SQL processing for reads, joins, and publishing new Iceberg results. The files remain in object storage and are available to other compatible engines.

Supported catalog integrations for reading include:

Supported Iceberg read catalogs: Oracle AI Data Catalog, Databricks Unity Catalog, Snowflake Open Catalog and Horizon Catalog, AWS Glue Data Catalog, and Google Cloud Lakehouse Runtime Catalog.
Supported catalogs for Iceberg reads. Writing support varies by catalog, storage, authentication, and release.

Catalog federation removes the need to create a separate external-table definition for every table. One connection can expose hundreds or thousands of authorized, supported tables across namespaces. New tables become discoverable after catalog refresh; individual external tables remain an option.

The @catalog suffix lets one SQL query combine local Oracle data with Iceberg tables from several connected catalogs. Consider a retailer with data spread across platforms: current orders in Oracle, customers in Oracle AI Data Catalog, products in Snowflake Open Catalog, and customer segments in Databricks Unity Catalog. The query below joins all four sources:

One Oracle SQL query joins current orders with customers, products, and customer segments across three connected catalogs.
One SELECT joins local Oracle data with Iceberg tables across three connected catalogs.
SELECT o.order_id, c.customer_name,
       p.category, s.segment, o.order_total
FROM current_orders o
JOIN sales.customers@AI_CAT c
  ON c.customer_id = o.customer_id
JOIN sales.products@SF_CAT p
  ON p.product_id = o.product_id
JOIN sales.customer_segments@DBX_CAT s
  ON s.customer_id = o.customer_id;

This example uses illustrative table, column, and connection names. It assumes compatible tables and that each catalog is connected with the required catalog and storage permissions.

Data Studio provides a visual way to browse connected catalogs and their tables, then open SQL Worksheet to query them. In SQL, namespace.table@catalog identifies a shared table. Access still follows the permissions configured for the catalog and storage.

Data Studio catalog browser showing connected catalogs, namespaces, and Iceberg tables; adapted for AI Lakehouse branding.
Data Studio catalog browsing, adapted for AI Lakehouse branding.

Read a broader range of Iceberg tables

We are expanding Iceberg feature coverage so AI Lakehouse can query a wider range of tables created and managed by other engines. Time travel lets customers query earlier retained snapshots, while partition pruning avoids reading unrelated data. The table below highlights supported capabilities.

CapabilityWhat it means for customers
Time travelQuery retained snapshots by timestamp or snapshot ID.
Partition pruningUse Iceberg metadata to skip unrelated partitions.

Oracle is fully committed to Apache Iceberg and open standards. We are investing in a broad set of additional Iceberg capabilities for upcoming releases and will continue expanding support as the specification and ecosystem evolve. Our goal is to help customers adopt new Iceberg features while keeping their data open and usable across compatible engines.

Compatibility depends on the features a table uses and the supported release, not its version number alone.

Oracle read consistency comes to Iceberg

Read consistency has long been a defining strength of Oracle. Oracle AI Lakehouse brings that strength to Iceberg tables with statement-level and transaction-level consistent reads. A single query reads a consistent table snapshot. Multiple queries within a supported read-only transaction retain the same snapshot, even while another engine publishes updates. Customers gain a familiar Oracle capability for reliable reporting and analytics while their data remains in open, shared Iceberg tables.

Consider running the same query twice within a transaction. Between the two reads, another engine updates the Iceberg table and publishes a new snapshot.

With transaction-level read consistency, Oracle AI Lakehouse returns the same result both times. Both queries read the same snapshot, even while other engines continue updating the shared table.

Two reads in the same Oracle read-only transaction return 1,000 orders from the same snapshot while Databricks adds 100 orders.
Within a supported read-only transaction, both queries read the same Iceberg snapshot.

Create Iceberg tables and publish Oracle SQL results

Customers can join current orders in Oracle with customer segments in Iceberg, then publish the result as a new Shared Sales Iceberg table.

Oracle SQL joins Oracle orders and Iceberg customer segments, then writes a Shared Sales Iceberg table for compatible engines.
Join Oracle orders with Iceberg customer segments, then publish a new Shared Sales table for compatible engines.

First, create sales.shared_sales. This example assumes the source tables and sales namespace already exist, and LAKEHOUSE_CATALOG is connected with the required catalog and storage write permissions. Replace <bucket> with your configured S3-compatible storage location:

CREATE ICEBERG TABLE sales.shared_sales (
  order_id INT,
  customer_id INT,
  segment STRING,
  order_total DECIMAL(12, 2)
)
WITHIN CATALOG LAKEHOUSE_CATALOG
STORAGE LOCATION 's3://<bucket>/sales/shared_sales/';

Then populate it with the results of a join between current_orders and sales.customer_segments. The selected columns must match the destination schema:

INSERT INTO sales.shared_sales@LAKEHOUSE_CATALOG
SELECT o.order_id, o.customer_id, c.segment, o.order_total
FROM current_orders o
JOIN sales.customer_segments@LAKEHOUSE_CATALOG c
  ON c.customer_id = o.customer_id;
COMMIT;

AI Lakehouse writes Parquet files and Iceberg metadata to object storage. COMMIT publishes the new snapshot through the catalog for compatible engines to query. AI Lakehouse can also refresh its metadata to read compatible changes from other engines.

Intelligent Iceberg performance

AI Lakehouse combines metadata-aware reads with Oracle query planning, cached data, and scale-out processing. These capabilities avoid unnecessary reads and make the remaining work more efficient.

Illustrative Iceberg query optimization: partition pruning, manifest file pruning, and predicate pushdown, with Lake Cache and Data Lake Accelerator.
Illustrative example, not a benchmark: partition and manifest pruning avoid unnecessary reads; the optimizer, Lake Cache, and Data Lake Accelerator address complementary performance needs.
  • Metadata pruning and predicate pushdown avoid unnecessary work. Oracle uses Iceberg partition metadata to skip partitions that cannot satisfy query filters. It also evaluates per-file min/max statistics stored in Iceberg manifests to eliminate irrelevant data files before opening them. Predicate pushdown applies remaining filters during scans, while catalog and metadata caching reduce repeated remote lookups.
  • Iceberg tables are first-class citizens of the Oracle optimizer. They benefit from decades of Oracle query-optimization intelligence, including sophisticated join planning, predicate placement, and parallel execution. The optimizer uses Iceberg metadata and statistics to choose an efficient execution plan, whether a query reads Iceberg tables alone or combines them with Oracle tables.
  • Lake Cache enables Exadata-like performance for queries over cached Iceberg data. Customers can query frequently accessed Iceberg data cached on Exadata storage, benefiting from fast local access instead of repeated object-storage reads and network latency. This accelerates recurring dashboards, reports, and analytics while the shared Iceberg table remains in the lake and available to other compatible engines.
  • Data Lake Accelerator scales I/O and compute. For eligible Iceberg scans, it adds Oracle-managed resources to read and process object-storage files in parallel. Elastic I/O bandwidth, compute, and network capacity support large scans without a separate customer-managed engine.
  • Clustering and compaction optimize Iceberg data layout. We are adding these capabilities so customers can use Oracle AI Lakehouse to maintain shared Iceberg tables, including data written by other engines. Efficient data layout is key to query performance. Clustering groups rows with similar values, making manifest min/max statistics more effective for file pruning. Compaction combines small Parquet files into larger ones, reducing file-open overhead and improving scan efficiency.

Bring operational data and open lakehouse data together

For customers, the value is a shared sales data product that supports forecasting, dashboards, and data preparation for AI applications. More broadly, AI Lakehouse can combine current Oracle customer or inventory data with historical, behavioral, or partner data published through Iceberg.

AI Lakehouse queries Oracle operational tables and Iceberg data together for AI applications and analytics.
Oracle tables and Iceberg tables remain in their respective storage systems. AI Lakehouse can query them together and publish derived results for reuse.

Applications can query the sources together or reuse a published Iceberg result when a shared dataset is needed. Publishing that result creates a new, derived table; it does not relocate either source dataset.

Oracle tables and Iceberg tables serve complementary needs. Oracle tables suit transactional workloads, enforced relational constraints, and consistently low-latency access. Iceberg suits analytical data that stays in object storage and is shared across engines.

Customers can choose the storage model that fits each workload while using Oracle SQL across both.

Get started with a shared Iceberg data product

Start with a connected, supported catalog and the required storage access. In Data Studio, browse an authorized Iceberg table and query it alongside Oracle data. To publish a shared result, use the CREATE, INSERT, and COMMIT workflow above.

Use these resources to check supported configurations and try the workflow: