Hacking Postgres Executor For Performance
summary
Postgres executor changes that are already committed and measurable: expression evaluation moves from tree-walking to an opcode-based execution loop yielding up to ~30% speedups in TPC-H; per-node dispatch gets replaced with direct function pointers to avoid ExecProcNode overhead; tuple deforming uses a compact attribute struct and branchless alignment for up to ~25% gains; scan inlining gives ~5–10% improvements. Future directions tested in patches include batching/vectorized execution and a push-based flow or broader JIT to compile whole plans for OLAP workloads.
AI-written summary · about
chapters (8)
- 0:00Start
- 0:36Introduction - Hacking Postgres Executor for Performance
- 2:49Executor's Impact on Query Performance
- 4:33CPU Performance Primer
- 11:33Postgres Execution Model: Volcano
- 15:54Committed Optimizations for Executor Performance
- 27:55Larger Executor Rewrite & Future Directions
- 32:04Discussions & Questions
related talks
47mHacking Postgres Executor For Performance
50mImproving Postgres' Efficiency
41mPostgreSQL Executor: Executing your execution plan
56mPerformance Improvements in PostgreSQL9 2
40mPerformance is a feature: accelerating Postgres
52mWriting fast C code for a modern CPU (and applying it to PostgreSQL)
34mPostgres 10, Performance, and You
47mHacking pgvector for performance