AI NewsInfrastructureAnnouncement

Neon runs Postgres 19 on the free plan on release day

Postgres 19 reached general availability on November 19. Neon made it available on every plan the same day, free tier included.

AI News

Editorial2 min read

LinkedInX
Neon blog hero illustration for Postgres 19 general availability

Image: Neon

Why it mattersREPACK and the new plan-advice tools change what a team can ask of a live database. A free tier lets a team test both on a real workload this week.

Postgres 19 reached general availability on November 19. Neon made it available on every plan the same day. The free tier is included. A developer can start a Postgres 19 project from the Neon Console. An agent can start one too, through Neon's own tool for agents.

The release was planned for October 29 and slipped three weeks. Two features change how a team runs a live database. One is REPACK. The other is a pair of planner tools called pg_plan_advice and pg_stash_advice.

REPACK rebuilds a bloated table while writes keep going

REPACK is a new SQL command. It rebuilds a table and reclaims space a long run of updates and deletes left behind. The command runs while normal reads and writes to the table keep going. That was the job of the external pg_repack tool for years. A team no longer has to install it. No third-party extension to get through a review.

One limit matters. The upstream Postgres 19 version of concurrent REPACK has a row limit. It fails if more than about 104 million rows in the target table are updated or deleted during the run. For most workloads that is enough. For the largest write tables, it is not.

pg_plan_advice lets a team pin a plan when the planner picks wrong

Postgres sometimes picks a bad query plan. Its statistics are off, or the shape of the data confuses the cost model. The usual answer is to rewrite the query or add an index. Both are slow and risky on a live system.

pg_plan_advice and pg_stash_advice let a team record the plan they want for one query. Postgres uses that plan when the same query comes in. The plan is pinned by the query's shape, not its exact text. Neon writes that the tools are for the case when an inaccurate estimate produces a poor plan. A team can intervene without changing the query the application sends.

The other new features

The release also brings sequence synchronisation for logical replication. Publications get more options. Subscription connections are now reusable. Autovacuum can run its index work in parallel. COPY can now write JSON. Statistics views are richer. JSON paths get new string methods. WAIT FOR LSN is new too. It lets a standby wait on a specific log position before serving a read. That is the shape a read-your-writes pattern on a replica usually needs.

What a team can do with it this week

Three concrete tests. First, run REPACK on a bloated table in a staging copy. Watch the write latency during the operation. If the table is under 100 million rows of churn, it should hold. Second, pick a query the planner keeps getting wrong. Record its plan with pg_plan_advice. See what the cost model agrees to use. Third, on a read replica, replace an arbitrary sleep with WAIT FOR LSN. Measure the drop in tail latency.

Neon's post says Neon is built on Lakebase Postgres, by Databricks, and is part of the Databricks Platform. The company did not publish its own benchmark numbers for the new features. The post points teams at the upstream release notes for the full list.

Source

Postgres 19 is out: try it on Neon, Neon.

SourceNeon

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project