PostgreSQL: Separate Hash Tables for Stats

Stats kinds can now opt into their own shared hash table to scale beyond the single shared store, with the flag exposed for inspection. Also fixed: cleaner pg_dump output for dropped identity columns and a vectorization tweak for numeric sums.

Duration: PT1M58S

Episode overview

This episode is a short developer briefing from PostgreSQL.

It explains recent repository work in plain language.

  • Show: PostgreSQL
  • Published: 2026-10-13T13:05:56Z
  • Audio duration: PT1M58S

Transcript excerpt

This excerpt keeps the crawler page concise. Listen to the episode or use the RSS feed for the full update.

Good morning, it's October 13th, 2026, and this is your PostgreSQL briefing.

The lead story is scalability in the statistics system, with a new option for large stats kinds to get their own shared storage.

Previously, all variable-sized stats entries lived in a single shared hash table. That works for small kinds, but breaks down when a kind holds many entries that need periodic scanning. In commit 4453978, contributors add an own hash option for variable-sized kinds, giving each opt-in kind an independent hash table…

For developers, this means less contention and more predictable scans for heavy stats, plus a clear path for extensions that need high-volume counters. The custom stats test module was refactored to cover the new behavior.

In other changes, a dump fix in commit c d e 3314 stops P G dump in inserts mode from emitting overriding system value when the only always-generated identity column has been dropped. The drop leaves the identity marker behind, and the old check ignored whether the column was dropped. The fix limits the check to…

Finally, commit 158b6d5 rewrites the loop in the numeric sum accumulator to better enable compiler auto-vectorization, continuing recent…

Nearby episodes from PostgreSQL

  1. Planner Correctness and Recovery Reliability
  2. Silent-Row Fixes and Operational Hardening
  3. Visibility and Type Conversion Correctness Fixes
  4. Planner Correctness and Backup Progress Fixes
  5. Parallel Material Storage Reporting
  6. Weekly Recap - Correctness Fixes and Backup Trust
  7. Index Scan Correctness Fixes
  8. Planner Correctness and Startup Lifecycle Fixes