earlyoom - Man Page

Early OOM Daemon

Synopsis

earlyoom [OPTION]...

Description

The oom-killer generally has a bad reputation among Linux users. One may have to sit in front of an unresponsive system, listening to the grinding disk for minutes, and press the reset button to quickly get back to what one was doing after running out of patience.

earlyoom checks the amount of available memory and free swap up to 10 times a second (less often if there is a lot of free memory). If both memory and swap (if any) are below 10%, it will kill the largest process (highest oom_score).

The percentage values are configurable via command line arguments.

If there is a failure when trying to kill a process, earlyoom sleeps for 1 second to limit log spam due to recurring errors.

Options

-m PERCENT[,KILL_PERCENT]

set available memory minimum to PERCENT of user mem total (default 10 %).

user mem total, introduced in earlyoom v1.8, is the memory accessible by userspace (MemAvailable+AnonPages as reported in /proc/meminfo). When a tmpfs ramdisk fills up, user mem total shrinks accordingly.

By using a percentage of user mem total as opposed to total memory, the set memory minimum can always be achieved by killing processes, even when tmpfs fills a large portion of memory.

earlyoom sends SIGTERM once both available memory and free swap are below their respective PERCENT settings.

It sends SIGKILL once both are below their respective KILL_PERCENT setting (default PERCENT/2).

Use the same value for PERCENT and KILL_PERCENT if you always want to use SIGKILL.

Examples:

earlyoom              # sets PERCENT=10, KILL_PERCENT=5
earlyoom -m 30        # sets PERCENT=30, KILL_PERCENT=15
earlyoom -m 20,18     # sets PERCENT=20, KILL_PERCENT=18

-s PERCENT[,KILL_PERCENT]

set free swap minimum to PERCENT of total (default 10 %). Send SIGKILL if at or below KILL_PERCENT (default PERCENT/2), otherwise SIGTERM.

You can use -s 100 to have earlyoom effectively ignore swap usage: Processes are killed once available memory drops below the configured minimum, no matter how much swap is free.

Use the same value for PERCENT and KILL_PERCENT if you always want to use SIGKILL.

-m Size[,Kill_size]

As an alternative to specifying a percentage of total memory, -M sets the available memory minimum to SIZE KiB. The value is internally converted to the percentage of mem total as reported on startup. user mem total is NOT used for the startup calculation because that would make the outcome dependent on how filled tmpfs is at that moment.

If you pass both -M and -m, the lower value is used. Example: Reserve 10% of RAM but at most 1 GiB:

earlyoom -m 10 -M 1048576

earlyoom sends SIGKILL if at or below KILL_SIZE (default SIZE/2), otherwise SIGTERM.

-s Size[,Kill_size]

As an alternative to specifying a percentage of total swap, -S sets the free swap minimum to SIZE KiB. The value is internally converted to a percentage. If you pass both -S and -s, the lower value is used.

Send SIGKILL if at or below KILL_SIZE (default SIZE/2), otherwise SIGTERM.

-k

removed in earlyoom v1.2, ignored for compatibility

-i

removed in earlyoom v1.7, ignored for compatibility

-d, –debug

enable debugging messages

-v

print version information and exit

-r INTERVAL

Time between printing periodic memory reports, in seconds (default 1.0). A memory report looks like this:

mem avail: 21790 of 23909 MiB (91.14%), swap free:    0 of    0 MiB ( 0.00%)

Set to 3600 to print a report every hour, to 86400 to print once a day etc. Set to 0 to disable printing periodic memory reports. Free memory monitoring and low-memory killing runs independently of this option at an adaptive poll rate that only depends on free memory. Due to the adaptive poll rate, when there is a lot of free memory, the actual interval may be up to 1 second longer than the setting.

-p

Increase earlyoom’s priority: set niceness of earlyoom to -20 and oom_score_adj to -100.

When earlyoom is run through its default systemd service, the -p switch doesn’t work. To achieve the same effect, enter the following three lines into sudo systemctl edit earlyoom:

[Service]
OOMScoreAdjust=-100
Nice=-20

-n

Enable notifications via d-bus.

To actually see the notifications in your GUI session, you need to have systembus-notify (https://github.com/rfjakob/systembus-notify) running as your user.

-n /Path/to/Script

Run the given script for each process killed. Must be an absolute path.

Within the script, information about the killed process can be obtained via the following environment variables:

EARLYOOM_PID     Process PID
EARLYOOM_NAME    Process name truncated to 16 bytes (as reported in /proc/PID/comm)
EARLYOOM_CMDLINE Process cmdline truncated to 256 bytes (as reported in /proc/PID/cmdline)
EARLYOOM_UID     UID of the user running the process

WARNING: EARLYOOM_NAME can contain spaces, newlines, special characters and is controlled by the user, or it can be empty! Make sure that your notification script can handle that!

-g

Kill all processes that have same process group id (PGID) as the process with excessive memory usage.

For example, with this flag turned on, the whole application will be killed when one of its subprocess consumes too much memory (as long as they all have the same PGID, of course).

Enable this flag when completely cleaning up the “entire application” is more desirable, and you are sure that the application puts all its processes in the same PGID.

Note that some desktop environments (GNOME, for example) put all desktop application in the same process group as gnome-shell. earlyoom might kill all such processes including gnome-shell when this flag is turned on.

Be sure to check how your environment behaves beforehand. Use

pstree -gT

to show all processes with the PGID in brackets.

--prefer REGEX

Prefer killing processes whose comm name matches REGEX (adds 300 to oom_score).

The comm name is the string in /proc/pid/comm. It is the first 15 bytes of the process name. Longer names are truncated to 15 bytes.

The comm name is also what top, pstree, ps -e show. Use any of these tools to find the proper comm name.

Example: You want to match gnome-control-center, which is longer than 15 bytes:

earlyoom --prefer '^gnome-control-c$'

--avoid REGEX

avoid killing processes whose comm name matches REGEX (subtracts 300 from oom_score).

--ignore REGEX

ignore processes whose comm name matches REGEX.

Unlike the --avoid option, this option disables any potential killing of the matched processes that might have occurred due to the processes attaining a high oom_score.

Use this option with caution as other processes might be sacrificed in place of the ignored processes when earlyoom determines to kill processes.

--sort-by-rss

find process with the largest rss (default oom_score)

--dryrun

dry run (do not kill any processes)

--syslog

use syslog instead of std streams.

Usually this is not needed as systemd handles logging of all output.

The --syslog option may be useful for minimal embedded systems that don’t run systemd. See https://github.com/rfjakob/earlyoom/pull/292 for some background info.

-h, --help

this help text

Exit Status

0: Successful program execution.

1: Other error - inspect message for details

2: Switch conflict.

4: Could not cd to /proc

5: Could not open proc

7: Could not open /proc/sysrq-trigger

13: Unknown options.

14: Wrong parameters for other options.

15: Wrong parameters for memory threshold.

16: Wrong parameters for swap threshold.

102: Could not open /proc/meminfo

103: Could not read /proc/meminfo

104: Could not find a specific entry in /proc/meminfo

105: Could not convert number when parse the contents of /proc/meminfo

Why not trigger the kernel oom killer?

Earlyoom does not use echo f > /proc/sysrq-trigger because the Chrome people made their browser always be the first (innocent!) victim by setting oom_score_adj very high. Instead, earlyoom finds out itself by reading through /proc/*/status (actually /proc/*/statm, which contains the same information but is easier to parse programmatically).

Additionally, in recent kernels (tested on 4.0.5), triggering the kernel oom killer manually may not work at all. That is, it may only free some graphics memory (that will be allocated immediately again) and not actually kill any process.

Memory Usage

About 2 MiB VmRSS. All memory is locked using mlockall() to make sure earlyoom does not slow down in low memory situations.

Bugs

If there is zero total swap on earlyoom startup, any -S (uppercase “S”) values are ignored, a warning is printed, and default swap percentages are used.

For processes matched by --prefer, negative oom_score_adj values are not taken into account, and the process gets an effective oom_score of at least 300. See https://github.com/rfjakob/earlyoom/issues/159 for details.

Author

The author of earlyoom is Jakob Unterwurzacher ⟨jakobunt@gmail.com⟩.

This manual page was written by Yangfl ⟨mmyangfl@gmail.com⟩, for the Debian project (and may be used by others).

See Also

nohang(8)

Info

General Commands Manual