Early OOM Cgroups Freezer Daemon

The Linux OOM Killer is a mechanism of last resort. When memory runs out, the kernel sacrifices a process to save the system. It is effective, but it is destructive. It often kills the wrong process, and for background workers processing long-running jobs, "death" means lost state and wasted compute.

earlyfreeze proposes a different contract: Suspension.

By monitoring Linux PSI (Pressure Stall Information), earlyfreeze detects memory contention before the kernel panics. When pressure spikes, it uses the Cgroups V2 freezer controller to atomically pause non-critical workloads. Once the pressure subsides, they resume exactly where they left off.

It is a very simple daemon designed to keep your workers alive, your system responsive, and the OOM Killer asleep.

git clone https://github.com/scale03/earlyfreeze.git
make

./earlyfreeze --target <path> [options]


Options:
  -t, --target <path>     Target cgroup (e.g., /sys/fs/cgroup/user.slice)
  -h, --threshold <val>   Freeze PSI % (default: 20.0)
  -r, --recover <val>     Thaw PSI % (default: 5.0)
  -i, --interval <ms>     Poll interval (default: 100)
  -d, --dry-run           Log only, do not freeze

• Linux Kernel 5.2+
• Cgroup v2 mounted
• PSI enabled

Note for non-systemd users (Alpine, Void, Artix): earlyfreeze is init-agnostic. It works on any cgroup v2 directory. Just point it to your custom group: ./earlyfreeze --target /sys/fs/cgroup/my_custom_group