Google Cloud announced on November 4, 2026 that AlloyDB for PostgreSQL can now read Apache Iceberg tables straight from Cloud Storage, as a preview feature. The same announcement moved two related features, AlloyDB sync and Lakehouse federation, to general availability.
Taken together, the three options give PostgreSQL developers on Google Cloud several ways to combine live application rows with large historical datasets kept in a lakehouse. The goal is to answer questions that need both kinds of data inside one SQL statement, without writing and running a separate ETL pipeline.
What Google Announced
The launch covers three integrations between AlloyDB and what Google now calls the borderless Lakehouse. Each one targets a different place where the data lives and a different read pattern:
- Direct Iceberg access (Preview): AlloyDB reads Iceberg tables where they sit in a Cloud Storage bucket, and the AlloyDB columnar engine speeds up those reads.
- AlloyDB sync (GA): BigQuery data is copied into a local AlloyDB table, either once or on a refresh schedule you choose.
- Lakehouse federation (GA): AlloyDB sends queries over native BigQuery tables to BigQuery, which does the heavy work and returns only the result rows.
The AlloyDB release notes for November 4 also list syncing BigQuery data into AlloyDB as generally available, through the alloydb_sync extension.
How Direct Iceberg Access Works
With the preview feature, you point AlloyDB at an Iceberg table by supplying a Cloud Storage location or by going through the Lakehouse runtime catalog. After a one-time registration, the table behaves like a normal PostgreSQL table, so you can filter it, aggregate it and join it with tables that hold live transactions.
Google's example registers a table with a CREATE TABLE ... USING iceberg statement and a uri option that names the gs:// location. A regular SELECT then joins that history with a live payments table to compare a pending charge with the customer's past spending.
The reads run on the same database instance as the application, so there is no extra reader node to deploy. Because the table is read in place, queries see whatever the data team most recently published to the lakehouse rather than a stale copy.
Google also published its own benchmark on a 1 TB dataset. It reports aggregation queries up to 7x faster than a comparable PostgreSQL-compatible service from Amazon, and up to 3.5x faster than an open-source PostgreSQL extension. These are vendor numbers, so treat them as a starting point for your own tests.
AlloyDB Sync and Lakehouse Federation
AlloyDB sync is for data that an application reads on almost every request but that changes only on a schedule, such as nightly risk scores. The copy becomes a real AlloyDB table, so it can carry indexes, be served by read pools, and use it with the columnar engine or vector search.
A sync can be started from the BigQuery console or from SQL. Google shows a call to alloydb_sync.create_bq_sync_table that takes the BigQuery table name, the local table name and a refresh interval such as one hour.
Lakehouse federation fits teams whose history already lives in native BigQuery tables. Filtering, grouping and row limits are handed to BigQuery, so only the final rows travel back. Administrators can also choose which project pays for these queries and limit which roles may run them.
What It Means for Developers
The main change is that the lakehouse is no longer a separate world for application code. A service that runs on PostgreSQL can now look up years of history at request time, which suits fraud checks, personalization and pricing logic that used to wait for batch jobs.
Picking the right mode matters more than the feature list. A simple rule of thumb from the announcement:
- Use direct Iceberg access when the history is stored as Iceberg files and is joined with live data on demand.
- Use AlloyDB sync for small, hot tables that need the lowest read latency and can be a little behind.
- Use Lakehouse federation when the data stays in native BigQuery storage and BigQuery should do the processing.
Keep the preview status in mind. Direct Iceberg access is not yet covered by GA terms, so test it with real query shapes, watch the memory and I/O load it adds to your primary instance, and keep production fallbacks until it reaches GA.
Oracle users will see the same trend on their side. Oracle recently added Iceberg catalog federation and native Iceberg writes to its AI Lakehouse, which shows open table formats becoming a common meeting point for operational and analytic databases.
How to Get Started
Direct Iceberg access needs an AlloyDB instance running PostgreSQL 18, where you enable the feature and follow Google's guide for accessing Iceberg data from AlloyDB. AlloyDB sync can be set up by picking a BigQuery table and choosing the sync option in the console, or with psql.
For Lakehouse federation, Google points to its setup guide for querying live BigQuery data from AlloyDB. The IAM roles granted to the AlloyDB cluster service account control which datasets and tables can be queried.
Conclusion
AlloyDB now offers three ways to work with lakehouse data: read Iceberg in place (preview), copy BigQuery tables locally on a schedule (GA), or let BigQuery process federated queries (GA). It cuts much of the pipeline work between transactions and history. Start with the GA options for production and trial Iceberg access on PostgreSQL 18 test instances.
Sources
- Google Cloud Blog: Unify operational and Lakehouse data with direct Iceberg access in AlloyDB for PostgreSQL
- AlloyDB for PostgreSQL release notes
See my articles more often when you search Google.
