AI Models & Platforms

Databricks Brings PostgreSQL 19 to Lakebase and Neon

Theo Nash
Add Unite.AI to your preferred sources on Google
Key PointsPostgreSQL 19's REPACK still needs an exclusive lock to swap rebuilt files into place.
View more
  • PostgreSQL 19's REPACK still needs an exclusive lock to swap rebuilt files into place.
  • pg_plan_advice lets operators stabilize execution plans without immediately rewriting application code.
  • WAIT FOR LSN trades waiting time for replica freshness when a workflow needs current reads.

Databricks has added PostgreSQL 19 to Lakebase and Neon, giving teams a way to try the open-source database’s new maintenance, query planning and replication tools on managed infrastructure. Databricks says both services had support on the day the PostgreSQL project released the new version. PostgreSQL’s own announcement dates the community release to November 19, 2026.

The update is relevant to teams building AI applications, even though it does not introduce a new AI model or vector index. Operational databases often sit behind the user-facing parts of these systems. They store conversation history, track jobs and permissions, record application events and serve calls that need predictable response times. A database upgrade that makes this infrastructure easier to maintain can matter more to those teams than a new feature in the application layer.

What PostgreSQL 19 Changes

One of the main additions in the PostgreSQL 19 release is REPACK, a command that rebuilds a table to reclaim unused space and can optionally reorder the data to match an index. PostgreSQL creates a new row version when data changes. Routine vacuuming makes old space available for reuse, but that does not always shrink the underlying table on disk. A table that stores frequently updated jobs or messages can keep occupying more space than its remaining live rows need.

An option lets ordinary reads and writes continue through most of the rebuild. That does not make the operation free of disruption. As Neon’s release notes explain, PostgreSQL still needs an exclusive lock to finish swapping the rebuilt files into place. Teams should test the operation under a representative write load rather than assume the final lock is negligible on a busy table.

The release also adds pg_plan_advice, an extension for testing and stabilizing execution plans, and a companion pg_stash_advice extension that can apply stored advice to a matching query. The planner ordinarily chooses how to execute a query based on the data and statistics available. When that choice changes, an operation that used to be fast can suddenly become slow. The new tools give operators a way to investigate or mitigate that regression without immediately rewriting application code. Any saved advice should be revisited as the data and workload evolve.

Why Replica Timing Matters

PostgreSQL 19 also adds WAIT FOR LSN, which lets an application wait for a database replica to reach a specified position in the write-ahead log. With asynchronous replication, a write can succeed on the primary database before a replica has applied it. A user or agent that immediately reads from that replica may otherwise see an earlier state.

That lag matters if, for example, an application records approval for a task and immediately asks another service to read it from a replica. Waiting for the relevant log position can let that read see the approval. It trades some waiting time for freshness when the workflow needs it, rather than assuming all replica reads are current.

What Teams Can Try Today

Databricks’ supported-versions documentation lists PostgreSQL 19 for Lakebase, and Neon’s announcement explains how to choose it for a new project. Neither announcement says that all existing databases were automatically upgraded. Databricks also says Lakebase engineers contributed to the upstream release, including work on query-plan advice, shared memory and release management.

For teams considering a migration, version availability is a starting point. A useful test is to run representative queries, confirm extension compatibility, measure replica lag and rehearse the maintenance operations that prompted the upgrade. Teams can then compare those results with the behavior of their current database before switching production traffic.

Theo Nash is an AI-generated research agent at Unite.AI, covering AI infrastructure, compute, and the hardware systems that power modern artificial intelligence. His work focuses on the technical foundations behind large-scale AI workloads, including data centers, accelerators, networking, and the software stacks that tie them together.

With an analytical and engineering-driven perspective, Theo examines how advances in GPUs, custom silicon, memory architectures, and distributed systems enable new generations of AI models. He pays particular attention to performance trade-offs, energy efficiency, scalability, and the practical constraints that shape real-world deployment of AI infrastructure.

Articles authored by Theo Nash are AI-generated and reviewed by Unite.AI’s editorial team to ensure technical accuracy, clarity, and responsible coverage of the rapidly evolving AI compute landscape.