We’re wrapping up the biggest release of the year: Postgres Pro Enterprise 18 distribution that combines everything new in PostgreSQL 18 with a set of practical additions for large, business-critical deployments.

Below is a clear breakdown of the most important updates.

What’s new in PostgreSQL 18 core

Moving to the PostgreSQL 18 core improves performance and day-to-day operations. Postgres Pro Enterprise 18 includes all the improvements from the vanilla PostgreSQL 18 release, including:

  • Asynchronous I/O (AIO). AIO lets the database issue multiple disk read requests in parallel instead of waiting for each one to complete. This speeds up sequential scans, VACUUM, and statistics collection, and sets the stage for better performance in high-load analytics and OLTP systems.

  • Skip Scan optimisation. A new planner strategy makes multi-column B-tree indexes useful even when the WHERE clause doesn’t include the leading index column. In many cases, this turns a slow sequential scan into a fast index-driven lookup.

  • Better monitoring tooling. Improvements to EXPLAIN ANALYZE (automatic BUFFERS, more detail around temporary storage and I/O) make it easier to diagnose and tune complex queries.

  • More efficient VACUUM. VACUUM has been reworked significantly, including “eager freezing”, which reduces the risk of locks and helps keep autovacuum scaling under control.

KVik: RAM-based caching for database workloads

For read-heavy workloads, we added an experimental KVik feature to the proxima extension. It caches database data in RAM and provides fast access over the RESP protocol.

Core KVik capabilities:

  • high-performance reads;

  • RESP support, with GET, SET, and DEL;

  • automatic cache invalidation when the underlying data changes (on INSERT, UPDATE, DELETE);

  • automatic updates in the database when the cache is changed via SET and DEL.

Improved adaptive query optimisation (AQO 4.0)

AQO (Adaptive Query Optimizer) 4.0 uses machine learning to automatically correct query execution plan mistakes produced by the standard planner.

The key change in 4.0 is removing planning-time overhead that limited adoption in earlier versions.

Other improvements include:

  • with planning overhead reduced, AQO is now recommended for any workload type, not just niche scenarios;

  • better query plans can cut disk subsystem load by tens of percent (based on user feedback);

  • the new version supports high-availability configurations and fixes issues related to growth of internal service data and temporary tables.

Scalability and high availability

The BiHA clustering solution has received updates for geo-distributed deployments and enterprise-grade operations.

  1. Cascading replication. This addresses excessive network and primary (leader) load in clusters with many nodes or spread across multiple data centres. Replicas can now receive changes not only directly from the leader, but also via intermediate nodes. This reduces inter–data centre traffic and frees up leader resources for transaction processing. The mechanism is fully automated: if an intermediate node fails, the cluster rebuilds the replication chain automatically and picks the most up-to-date source (Best Follower).

  2. Major upgrade with minimal downtime. There is now a procedure to upgrade the BiHA cluster from version 17 to 18 with minimal downtime, with no data loss, and with the ability to roll back to the previous version during the upgrade.

Load Balancer: smarter routing

The proxima extension now includes a built-in load balancer that simplifies query routing. No more guessing where requests should go: the database exposes dedicated ports:

  • P2L (Proxy-to-Leader) for write traffic (RW);

  • P2F (Proxy-to-Follower) for distributing read traffic (RO) across replicas.

Several load-balancing strategies are available for P2F to match different requirements and keep resource usage even across the cluster:

  • Round-Robin and Random for spreading requests evenly, in order or at random

  • Weighted Round-Robin to account for different server capacity using weight coefficients

  • Least connection to send a request to the replica with the fewest active sessions.

Write and big-data optimisations

For DBAs dealing with ETL workflows and bulk loads, this release adds three new tools:

  1. Append Optimised Table

    A new table option: append_optimized = true. During bulk inserts, data is buffered in batches, which sharply reduces free-buffer searches and WAL writes. This is ideal for fast ingestion of logs and historical data.

  2. Multi-segment inserts

    Solves buffer contention during parallel inserts into a single table. Each backend process writes to its own file segment, enabling near-linear scaling at 10–50 parallel workers.

  3. Deferred compression (CFS)

    For compressed file systems, you can insert data without compression (for maximum speed) and run compression later in the background. This speeds up bulk loads during a “night window”.

Reference partitioning

You can now automatically create partitions in child tables based on a foreign key to a parent table.

Example: when you create a new “January 2025” partition in Orders, the matching partition in Order_Items is created automatically. This makes hierarchical partition management much simpler.

Delayed password change

For technical (service) accounts, there’s now a safe, gradual password rotation mechanism. The OLD_PASSWORD_TIME parameter defines a period when both the old and the new password remain valid. This lets you roll out configuration updates across applications without stopping the service.

Query result caching

For “heavy” repeated selects (for example, pagination on high-traffic sites), a pgpro_result_cache module has been added. Add the result_cache hint to a SQL query and the result is stored in memory. If the source tables change, the cache is invalidated immediately, which can cut query execution time from seconds down to fractions of a millisecond.

Postgres Pro Enterprise 18 is available to explore. Details are in the documentation.