PostgreSQL at low Level: stay curious!
summary
Shows concrete diagnostics and fixes: attach strace to a Postgres backend and find PostGIS failing shm allocations because Docker’s default shm is 64 MB; sample with perf to spot a microcode mitigation (VERW) that drives ~25% more syscalls; disable KVM PLE to recover ~1 ms latency in one experiment; enable classical huge pages to drop last‑level cache misses ~20%; deploy user probes via eBPF / BCC and a Postgres‑BCC repo to measure per‑query metrics, noting probes need privileged containers and can crash old kernels.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 2:04What are system calls
- 8:05The plan
- 8:54Information sources
- 10:24Analytical queries
- 12:35Visual Data VSS
- 14:57Memory Vulnerability
- 21:51Explanation
- 25:55Huge Pages
- 28:17Stateless Measurement
- 28:45What is BPF
- 38:48PostgreSQL BCC
- 40:05Writeback
- 40:52Timeout
- 41:42kubernetes
- 42:06soft limited hard limited
- 43:01memory pressure
- 43:32how to run
- 44:30kbernetes
- 44:57Responsibility
- 45:45Bug
- 46:20Kernel Panic
- 46:56Easy Proof
- 47:57Scaling
related talks
30mPostgreSQL and Linux Kernel interactions
27mAnalyzing Postgres performance problems using perf and eBPF
39mDeep dive into PostgreSQL internal statistics
50mPostgreSQL hacker tips
50mHacking on Postgres: An Overview
45mEnhancing PostgreSQL Plasticity: New Frontiers in Memory Management
21mA Guided Tour Through Postgres Internals with Elizabeth Garrett Christensen
26mPostgreSQL performance tips you have never seen before V2.0