Analytics Accelerator 1.3 release notes v1.3

Released: 17 September 2025

Analytics Accelerator 1.3.1 includes the following enhancements and bug fixes, reflecting all changes since version 1.3.0:

Features

• Native CTAS support: Simplify data migration and table creation with CREATE TABLE AS SELECT (CTAS), now supported directly using the USING PGAA WITH syntax. This allows you to create PGAA-managed tables directly from the results of a Postgres query.
• Spark Connect integration: Offload heavy analytical queries to a remote Spark cluster using the new Spark Connect executor.
  • Use pgaa.executor_engine to switch between seafowl (default) and spark_connect.
  • Manage endpoints via pgaa.spark_connect_url and pass fine-grained JSON settings through pgaa.spark_connect_extra_config.
  • Execute Spark SQL statements verbatim on the cluster with the pgaa.spark_sql() function, which enables access to Spark-specific functions and Iceberg maintenance routines.
• High-speed Iceberg replication with PGD 6.1+: Significant performance gains for Iceberg tables with new support for merge-on-read using equality deletes. When pgd.replicate_to_analytics is enabled, changes are replicated more efficiently than traditional copy-on-write methods.
• Expanded catalog support: PGAA now supports AWS S3 Tables, allowing you to use the pgaa.import_catalog and pgaa.attach_catalog functions to manage S3-native Iceberg catalogs.
• Dynamic worker management: Implemented a new dynamic, AIDB-style management system that automatically scales and manages background worker processes based on active workload.

Bug fixes

• Resolved an issue where the Seafowl object store cache duplicated entries when working with Iceberg REST tables.
• Fixed a failure in CREATE TABLE AS SELECT (CTAS) statements when the PGAA access method was not explicitly set.
• Corrected a logic error that caused CTAS operations to erroneously attempt a DirectScan.
• Fixed a critical bug that caused data loss when using small replication lag settings.
• Resolved a race condition in transaction state management during the cancellation of a CTAS operation.
• Implemented LSN-based filtering to skip previously processed changes from PGD, preventing data duplication.
• Improved data flush plans by disabling unnecessary physical optimizers.
• Resolved a crash during CTAS operations for tables exceeding the MAX_ROWS_PER_SYNC limit.
• Improved efficiency by concatenating adjacent row changes when schemas are compatible and transactions are complete.
• Optimized equality deletes by removing unnecessary grouping logic, as primary keys are unique and changes are already squashed.
• Ensured the system correctly picks up and reflects "last updated" metadata timestamps for tables.
• Implemented non-durable LSN skipping to improve replication resilience.
• Added logic to automatically revert aborted or incomplete transactions to maintain data integrity.
• Added native data purging functionality to CTAS operations.
• Resolved an issue with the object store cache that occurred when reading from an Iceberg REST catalog.
• Fixed a lock-up issue where the PGAA autoupgrade worker could inadvertently lock PGD DDL replication.
• Resolved CREATE TABLE AS SELECT (CTAS) failures on multi-node setups by preventing tuple insertions on PGD follower nodes.
• Fixed a segmentation fault that occurred when inserting NULL values during CTAS operations.
• Resolved an issue where tables could not be recreated after a purge by ensuring the sync collection is reset before CTAS execution.
• Corrected a misleading batch size estimation in the PGD replication writer to improve monitoring accuracy.
• Optimized the replication process by skipping unnecessary row squashing for batches that are exclusively append-only or remove-only.

Infrastructure & other changes

• Enabled Prometheus metrics by default for the background Seafowl process via the pgaa.autostart_seafowl_enable_metrics and pgaa.autostart_seafowl_metrics_host parameters.
• Implemented immediate flushing for truncate and purge operations.
• Refactored internal synchronization logic for creating Delta and Iceberg targets.
• Removed the default internal object store and metastore from Seafowl.
• Introduced automated benchmarks to track and validate replication performance.
• Implemented various query and replication performance improvements specifically for the Iceberg format.
• Updated the Rust compiler to version 1.92.0 to leverage the latest language features and security patches.
• Removed the standalone metastore-agent operation mode and its transitive dependencies to simplify the architecture and reduce the extension's footprint.

Deprecations

• The standalone Seafowl package has been discontinued; the Seafowl engine binary is now natively included within the PGAA package.
• Removed the --one-off and --cli Seafowl options.
• Removed internal support for several SQL statements within the Seafowl engine that are not utilized by PGAA, including:
  • Data modification: INSERT, UPDATE, DELETE.
  • Schema management: CREATE/DROP for tables and schemas, and ALTER RENAME.
  • Table creation: CREATE TABLE AS, CREATE EXTERNAL TABLE, and CONVERT TO DELTA.
  • Data export: COPY TO.