BPFOOM: using eBPF to customize the OOM handling
The OOM killer is supposed to save the system from being deadlocked when there is not enough memory to continue the normal operation, preferably by getting rid of something not particularly useful and without damaging anything really useful.
Previous 25+ years of the OOM killer development showed that there are as many opinions on what the best policy is as there are engineers who are working in this area.
Finally eBPF allows everybody to develop their own policy, so no more arguments, right?
Roman GUSHCHIN
Roman is a kernel engineer with interests in memory management, cgroups, bpf, self-driving cars and random trouble-shooting, with a particular passion to optimize the kernel memory accounting. Co-maintainer for memory cgroups. Currently works at Google.