The No Hang Daemon

Nohang is a highly flexible full-featured daemon for Linux which is able to correctly prevent out of memory conditions.

What is the problem?

OOM killer doesn't prevent OOM conditions.

Solution

Use of earlyoom or nohang, but nohang is more featured.

Some features

• convenient configuration with a well commented config file (there are 38 parameters in config)
• SIGKILL and SIGTERM as signals that can be sent to the victim
• zram support (mem_used_total as a trigger)
• customizable intensity of monitoring
• desktop notifications: results of preventings OOM and low memory warnings
• black, white, prefer, avoid lists via regex
• possibility of restarting processes via command like systemctl restart something if the process is selected as a victim
• look at the config to find more features

Demo

Video: nohang prevents OOM after command while true; do tail /dev/zero; done has been executed.

An exaple of output

MemAvail: 2976 M, 50.6 % | SwapFree: 10758 M, 100.0 %
MemAvail:    0 M,  0.0 % | SwapFree: 10281 M,  95.6 %
MemAvail:    0 M,  0.0 % | SwapFree:  9918 M,  92.2 %
MemAvail:    0 M,  0.0 % | SwapFree:  8659 M,  80.5 %
MemAvail:    0 M,  0.0 % | SwapFree:  7235 M,  67.3 %
MemAvail:   19 M,  0.3 % | SwapFree:  6851 M,  63.7 %
MemAvail:    0 M,  0.0 % | SwapFree:  5780 M,  53.7 %
MemAvail:    0 M,  0.0 % | SwapFree:  5008 M,  46.6 %
MemAvail:    0 M,  0.0 % | SwapFree:  4199 M,  39.0 %
MemAvail:    0 M,  0.0 % | SwapFree:  3502 M,  32.6 %
MemAvail:    0 M,  0.0 % | SwapFree:  2929 M,  27.2 %
MemAvail:    0 M,  0.0 % | SwapFree:  2446 M,  22.7 %
MemAvail:    0 M,  0.0 % | SwapFree:  2093 M,  19.5 %
MemAvail:    0 M,  0.0 % | SwapFree:  1573 M,  14.6 %
MemAvail:    0 M,  0.0 % | SwapFree:  1320 M,  12.3 %
MemAvail:    0 M,  0.0 % | SwapFree:  1117 M,  10.4 %
MemAvail:    0 M,  0.0 % | SwapFree:   943 M,   8.8 %

2018-07-06 Fri 03:04:37
  MemAvailable (0 MiB, 0.0 %) < mem_min_sigterm (588 MiB, 10.0 %)
  SwapFree (943 MiB, 8.8 %) < swap_min_sigterm (1076 MiB, 10.0 %)
  Preventing OOM: trying to send the SIGTERM signal to tail,
  Pid: 14636, Badness: 777, VmRSS: 4446 MiB, VmSwap: 8510 MiB
  Success
MemAvail:  173 M,  2.9 % | SwapFree:  3363 M,  31.3 %
MemAvail: 4700 M, 80.0 % | SwapFree:  8986 M,  83.5 %
MemAvail: 4668 M, 79.4 % | SwapFree:  8997 M,  83.6 %
MemAvail: 4610 M, 78.5 % | SwapFree:  9024 M,  83.9 %

Requirements

• Linux 3.14+
• Python 3.4+

Memory and CPU usage

• VmRSS is about 12 MiB
• CPU usage depends on the level of available memory (the frequency of memory status checks increases as the amount of available memory decreases) and monitorong intensity (can be changed by user via config)

Status

Now is unstable, but works great on my PC and tested on Fedora 28 x86_64, Debian 9 x86_64, Debian 8 i386. The first stable version will be released soon.

Download

git clone https://github.com/hakavlad/nohang.git
cd nohang

Installation and start for systemd users

sudo ./install.sh

Purge

sudo ./purge.sh

Command line options

./nohang --help
usage: nohang [-h] [-c CONFIG]

optional arguments:
  -h, --help            show this help message and exit
  -c CONFIG, --config CONFIG
                        path to the config file, default values:
                        ./nohang.conf, /etc/nohang/nohang.conf

How to configure nohang

Read config and edit values. Execute sudo systemctl restart nohang for apply changes.

Feedback

Please, create issues.