Skip to content

Add PostgresMarathon 2-008: LWLock:LockManager and prepared statements

Nikolay Samokhvalovrequested to merge
blog/postgres-marathon-2-008 into master

Description

Deep dive into how prepared statements reduce lock contention by analyzing planner locks vs executor locks.

Key findings:

  • Explains why locking behavior changes on 7th execution, not 6th
  • Includes PG18 source code analysis with permanent commit links
  • Demonstrates the mystery of execution 6 using generic plan but still acquiring all locks

Changes

  • New blog post: blog/20251014-postgres-marathon-2-008.mdx
  • Cover image: static/assets/blog/20251014-postgres-marathon-2-008-cover.jpg
  • All GitHub links use permanent commit hashes instead of branch references

Key insights

  • Prepared statements reduce LWLock:LockManager contention starting with 7th execution
  • Execution 6 uses generic plan but still acquires planner locks (all indexes)
  • Execution 7+ uses cached generic plan with executor locks (table only)

Checklist

  • All proposed text changes have been reviewed by an LLM for grammar
  • This MR contains text changes and they have been reviewed
  • This MR contains no GUI/CLI changes
  • This MR contains no API changes

Merge request reports