scx_chaos: Pushing the sched_ext API to schedule badly
summary
scx_chaos is a deliberately non work‑conserving scheduler built on top of scx_p2dq that injects concrete faults: randomly delaying scheduling events (configurable up to ~30s), CPU frequency chaos (tasks slowed ≈8x), slow kprobes that deschedule and delay tasks, and a work‑in‑progress futex inversion. It reproduces a rare hardware-driver timeout (1-in‑a‑million) in 2–3s, and running with modest delays cuts cgroup selftests from 559 to 147 failing one extra test.
AI-written summary · about
chapters (6)
- 0:00Introduction to SCX Chaos
- 1:02What is SCX Chaos
- 3:50Bugs found with scheduling chaos
- 6:59Extending other schedulers
- 13:33Self-test results with chaos
- 15:19Q&A on futex delays
related talks
39mReal-Time Scheduling Fault Simulation
37mReal-time Scheduling Fault Simulation
39mWhy SREs can't afford to NOT do Chaos Engineering
30mControlled Scheduling of Concurrent Elixir Programs
45mSchrödinger's Pokémon: Observability for Chaotic Load Testing
34mChaos Engineering For Hybrid Targets With LitmusChaos
28mExtreme OS Kernel Testing
19mLearnings from Chaos experiments