Incident sheet · /infrastructure
- System
- Infrastructure & Cloud
- Filed
- Read
- 3 min · 675 words
- Severity
- Routine
PostgreSQL 19 Ships Day-One on Lakebase With REPACK and Plan Advice
PostgreSQL 19 hit GA today and is live on Lakebase the same day, adding concurrent REPACK, plan advice, sequence replication and WAIT FOR LSN for operations teams.

Event log
- PostgreSQL 19 reached GA and became available on Lakebase the same day; other managed services typically need weeks.
- REPACK CONCURRENTLY fails if more than about 104 million tuples change in the target table during the operation.
- Features reverted before release include SQL/PGQ (September 7, 2026) and temporal UPDATE/DELETE (September 15, 2026).
- PostgreSQL 19.1, the first minor release, is anticipated in February.
- pg_plan_advice and pg_stash_advice let teams override bad plans without inline hints.
PostgreSQL 19 reached general availability today and is available on Lakebase the same day — a turnaround that typically takes weeks on other managed PostgreSQL services. Several Lakebase engineers are PostgreSQL committers and contributors who worked on the release directly, covering query plan advice, shared memory, locale handling, and release management. Customers can select PostgreSQL 19 when creating a new Lakebase project.
There is no single headline feature. Version 19 targets operational problems database teams know well: bloated tables, unstable query plans, replica lag, incomplete statistics, and maintenance that is hard to run safely in production. The common thread is control — more ways to intervene when defaults fall short, without sacrificing reliability.
What does REPACK change for table maintenance?
PostgreSQL's multiversion concurrency control means an UPDATE creates a new row version and a DELETE leaves a dead tuple. Vacuuming makes that space reusable but does not necessarily return it to the operating system or restore physical row order. The usual fixes — VACUUM FULL, CLUSTER, or pg_repack — take an ACCESS EXCLUSIVE lock for the whole operation, blocking normal access.
REPACK reclaims space like VACUUM FULL and, with USING INDEX, physically orders rows like CLUSTER. With CONCURRENTLY, it rebuilds the table while reads and writes continue, capturing changes through logical decoding, then briefly takes an ACCESS EXCLUSIVE lock to swap files into place.
The constraints are real:
- It needs free capacity under max_repack_replication_slots and a primary key or index-based replica identity.
- It requires temporary disk for a copy of the table and all indexes, plus a possible sort file up to the table's size.
- The upstream implementation fails if more than roughly 104 million tuples are updated or deleted in the target table during the repack.
- It is not MVCC-safe: transactions using pre-repack snapshots can see the table as empty afterward.
In plain terms: REPACK brings into core a job that previously required downtime or an extension.
How does plan advice stabilize bad query plans?
Per-table statistics cannot capture cross-table correlations. If enterprise customers are 0.1% of customers but generate 30% of orders, the planner may estimate each such customer has a few orders and choose a nested loop — when a hash join would fit better.
PostgreSQL 19 adds pg_plan_advice, which describes planner decisions as advice, applies supplied advice, and reports whether it was followed, and pg_stash_advice, which stores advice by query ID and applies it automatically. EXPLAIN reports "matched" when the plan complies — though matched does not mean better; you still evaluate against the real workload. Unlike inline hints, advice lives outside the application's query text.
What else is new in operations and replication?
- Autovacuum can vacuum indexes with parallel workers, controlled by autovacuum_max_parallel_workers, with a new scoring system prioritizing tables; pg_stat_autovacuum_scores exposes the scores.
- New views: pg_stat_lock, pg_stat_recovery. EXPLAIN (ANALYZE, IO) reports asynchronous I/O; pg_stat_wal and vacuum logs report full-page-image bytes.
- Logical replication: FOR ALL SEQUENCES, FOR ALL TABLES EXCEPT, subscriptions via a postgres_fdw server, and update_deleted conflict detection with retain_dead_tuples and pg_stat_subscription_stats.
- WAIT FOR LSN blocks a standby until replay reaches a requested position, enabling read-your-writes without polling.
- postgres_fdw can import remote statistics for better federated plans.
- INSERT ... ON CONFLICT DO SELECT provides a clean get-or-create that avoids dead tuples from no-op updates; COPY TO can emit JSON; random() supports date and timestamp types; IGNORE NULLS arrives; io_method = worker auto-scales I/O workers.
What was cut before release?
Reviewers reverted several features over correctness or design risk: SQL/PGQ property graph queries (September 7, 2026), UPDATE/DELETE ... FOR PORTION OF (September 15), MERGE/SPLIT PARTITIONS (August 27), GROUP BY ALL (July 17), parallel pg_dumpall (June 18), and pg_get_role_ddl() and related DDL functions (September 13).
Lakebase advises a measured rollout: check changed defaults and compatibility behaviors, confirm extension availability before migrating, and consider waiting for PostgreSQL 19.1 — anticipated in February — as many teams do for any .0 release.
Same-day availability on a managed platform, plus core tooling for bloat and plan stability, is the story here.
via databricks.com (Source)
More from Amara Osei
Show full bio
Correspondent covering regulation, policy and their impact on enterprise information technology and the leaders who run at CIO Monitor.
203 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- PostgreSQL 19 shipped late as AI-found bugs strained maintainers
- Microsoft previews serverless HorizonDB for agent applications
- Microsoft puts HorizonDB Postgres into general availability at Ignite
- Google and Anthropic Compress Vulnerability Disclosure Windows
- Attackers Deleted 100+ Storage Accounts in 7 Minutes — Cloud Resilience Tools Didn't Help