cgroup2 Resource Control strategies w/ resctl-demo
summary
An interactive terminal demo reproduces Facebook production surprises: a memory leak that previously dropped web RPS to near zero for ~40 minutes instead causes only small dips when resctl-demo enables cgroup2 controllers (cpu, memory, io cost) and throttles maintenance workloads, keeping write traffic under the device limit instead of ~300 MB/s and preventing IO latency spikes. Live graphs show psi contention, memory isolation and a UMD watchdog that eventually kills the offender, proving the controls work.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:54Resource Control
- 1:21Work Conservation
- 2:13Memory Leaks
- 6:00Components
- 8:02Documentation
- 9:23Protection
- 9:55Scenario
- 11:01Live Demo
- 21:36Live Demo Results
- 24:03Memory Leak
- 28:12Site Loading
- 33:10Why is latency deteriorating
- 38:19Resource Control Demo under active development
- 39:45Resources to try
- 40:36Questions
related talks
42m"cgroup2 Resource Control strategies w/ resctl-demo"
43mresctrl filesystem
18mIO and cgroups, the Current and Future Work
35mFuture-Proof Your Applications: Avoid Breakdowns With Cgroup Chan...
25mRecent developments in I/O control and testing
38mUnderstanding and Working with the Cgroups Interface
40mCase Study for Resource Control in Embedded Linux Container Integration
01h 02mDiving deeper into control groups (cgroups) v2