user-1001.slice Memory Fix (Bead bf-41xv)

Summary

Replaced MemoryMax hard cap with MemoryHigh soft cap and enabled cgroup swap for user-1001.slice on NixOS.

Problem

The user-1001.slice had:

MemoryMax=32G (hard cap - OOM kill at limit)
MemorySwapMax=0 (swap disabled)

This caused hard process kills when memory hit 32G instead of allowing the kernel to reclaim memory gracefully or use swap.

Solution

Changed to:

MemoryHigh=32G (soft cap - kernel applies pressure and reclaims)
MemoryMax=infinity (no hard kill)
MemorySwapMax=infinity (can use system swap)

Changes Made

NixOS Configuration (`/etc/nixos/configuration.nix`)

Fixed syntax error with writeTextDir (needed parentheses around the call) and configured:

systemd.tmpfiles.rules = [
  "L+ /etc/systemd/system/user-1001.slice.d/50-memory.conf - - - - /etc/systemd/system-user-slices/user-1001.slice.d/50-memory.conf"
];

systemd.packages = [
  (pkgs.writeTextDir "etc/systemd/system-user-slices/user-1001.slice.d/50-memory.conf" ''
    [Slice]
    # MemoryHigh applies soft memory pressure (reclaim) instead of hard OOM kill
    MemoryHigh=32G
    # Do not set MemoryMax - allow emergency usage if needed
    # Do not set MemorySwapMax - allow cgroup to use system swap
  '')
];

Verification

After nixos-rebuild switch:

cat /sys/fs/cgroup/user.slice/user-1001.slice/memory.max      # max (unlimited)
cat /sys/fs/cgroup/user.slice/user-1001.slice/memory.high     # 34359738368 (32G)
cat /sys/fs/cgroup/user.slice/user-1001.slice/memory.swap.max # max (unlimited)

The NixOS-generated config in /nix/store/.../user-1001.slice.d/50-memory.conf contains the correct settings.

Why This Matters

Before: MemoryMax=32G meant hard OOM kill at the limit
After: MemoryHigh=32G allows the kernel to apply gradual memory pressure and reclaim pages, giving processes time to respond before OOM
Swap overflow provides an additional 24GB buffer for memory spikes

Date

2026-05-30

2 KiB Raw Permalink Blame History