How does cgroups v2 memory control differ fundamentally from legacy cgroups v1?

In legacy cgroups v1, memory and swap were managed under separate, uncoordinated controllers (memory.limit_in_bytes vs memory.memsw.limit_in_bytes), with only a single hard limit that abruptly triggered the OOM killer. Cgroups v2 unified the hierarchy and introduced a four-tier protection and throttling model: memory.min (hard protection), memory.low (best-effort overcommit protection), memory.high (proactive reclaim with asynchronous throttling), and memory.max (hard ceiling triggering synchronous reclaim and OOM).

What is the difference between memory.high and memory.max?

When a cgroup exceeds memory.high, the kernel does NOT immediately kill processes. Instead, it subjects allocating threads to progressive proportional sleep penalties (throttling) while the background kernel threads (kswapd) proactively reclaim page cache and swap anonymous memory. In contrast, memory.max is a hard ceiling: exceeding it blocks allocating threads and forces immediate synchronous in-line reclaim; if reclaim fails to bring usage below memory.max, the kernel OOM killer terminates a process inside the cgroup.

What is memory.oom.group and why is it crucial for container runtimes like Kubernetes?

By default, the Linux OOM killer selects and terminates a single process (the one with the highest oom_score) within a cgroup. In multi-process containers (like database replicas with background workers or sidecars), killing a single child process leaves the application in an unstable, corrupted partial state. Enabling memory.oom.group = 1 instructs the kernel to terminate all processes within the cgroup simultaneously, allowing Kubernetes to restart the container cleanly.

How do memory.min and memory.low provide guaranteed memory protection?

memory.min provides hard protection: memory within this boundary will never be reclaimed by the kernel under any system memory pressure, unless the entire system panics. memory.low provides soft protection: memory within this boundary is shielded from reclaim unless unshielded cgroups are completely exhausted. This prevents noisy-neighbor containers from evicting hot page caches of mission-critical databases.

Linux Memory cgroups v2, Swap Accounting & Reclaim Studio

Verified Specification | Standardized Formulas | Instant Precision

Secure & Private (Zero Data Retention) Free Access • No Sign-Up

Linux Memory cgroups v2 & Reclaim Studio

Interactive kernel cgroups v2 simulator modeling memory.min, memory.low, memory.high, memory.max, proactive reclaim throttling, and OOM group semantics.

cgroups v2 / systemd

Simulated Container Workload

memory.high (Proactive Reclaim Limit)

2 GiB 4096 MiB 8 GiB

memory.max (Hard OOM Ceiling)

4 GiB 6144 MiB 12 GiB

cgroup Settings

memory.oom.group = 1 memory.swap.max = 2048 MiB

memory.current (Usage)

4,620 MiB

Anon: 3,200 MiB | File: 1,420 MiB

Proactive Throttling Delay

12.4 ms / alloc

Penalty for exceeding memory.high

Kernel Memory Reclaim

Active (kswapd)

Evicting inactive file cache

OOM Killer Status

IDLE (Safe)

1,524 MiB headroom below max

cgroup v2 Memory Thresholds & Pressure Zones ● Proactive Reclaim Active

systemd / cgroup v2 Unit Configuration

Kernel Reclaim & Throttling Formulas

Linux Memory cgroups v2 & Reclaim Studio

🔍 Companion Workbenches & Verification Tools

Cryptographic Hash Generator (SHA-256, SHA-512, SHA-384, SHA-1, MD5)

Distributed Transactions: CockroachDB Parallel Commits & Hybrid Logical Clocks Studio

Linux KSM (Kernel Samepage Merging) & Memory Deduplication Studio

Apache Iceberg & Parquet Columnar Lakehouse Storage Studio

Why Smart Users Are Switching to This Free Toolkit

The Browser Feature Most People Never Use

Stop Paying for Software You Can Get Free

File Conversion Tricks That Save Hours Every Week

This Free Tool Replaced 5 Paid Apps for Our Team

What Your Browser Can Do Without Any Extensions