Taming Zombie Cgroups - Should Reparenting Be Simple or Smart?

Explore a critical Linux kernel issue in this 16-minute conference talk that addresses the persistent "zombie memory cgroup" problem affecting system memory management. Learn about the technical challenge where destroyed memory cgroups cannot be properly freed due to lingering LRU pages, particularly shared file pages, that remain charged to defunct cgroups and accumulate over time in environments with frequent cgroup creation and destruction. Discover the community-driven solution currently being upstreamed that transfers ownership of LRU pages from memory cgroups to object cgroups, enabling proper reparenting to parent cgroups when memory cgroups go offline. Examine the core mechanism's correctness guarantees, including binding stability between folios and memory cgroups while critical locks are held, ensuring safe and race-free reparenting operations. Analyze the central design trade-off between implementing simple versus sophisticated reparenting processes, considering how merging child cgroup LRU lists into parent cgroups can disrupt generational ordering essential for effective page reclaim. Understand the implications for Multi-Gen LRU (MGLRU) support where generations must be adjusted during reparenting, and engage with the ongoing community discussion about optimizing the final reparenting step for real-world performance while maintaining correctness and maintainability.