Taming Zombie Cgroups: Should Reparenting Be Simple or Smart?
summary
Proposes upstreaming a concrete fix that reassigns LRU page charges from memory cgroup to the object cgroup API so zombie cgroups free their metadata, with a correctness guarantee that folio-to-cgroup binding is stable while the lruvec lock is held. Simple reparenting splices child active/inactive lists into parent (moving pages to the list tail), possibly breaking LRU ordering and MGLRU generations; smarter approaches try per-generation merging but add heuristics. Reported impact reaches hundreds of gigabytes.
AI-written summary · about
chapters (7)
- 0:00Introduction to zombie cgroups
- 1:45Simple vs smart reparenting
- 3:33How simple reparenting works
- 6:46Real-world scenarios and concerns
- 10:20Complex reparenting trade-offs
- 13:32Discussion and audience questions
- 15:00Summary and conclusion
related talks
34mCgroups and Enterprise Users
27mWhen Linux Memory Accounting Goes Wrong
35mFuture-Proof Your Applications: Avoid Breakdowns With Cgroup Chan...
42mAllocators for Compressed Pages: In-depth Comparison with z3fold in Focus
38mUnderstanding and Working with the Cgroups Interface
28mMulti-Gen LRU: Current Status & Next Steps
23mImproving data placement for Zoned Linux File systems
20mObject caching on Zoned Storage