The oom-killer generally has a bad reputation among Linux users. This may be part of the reason Linux invokes it only when it has absolutely no other choice. It will swap out the desktop environment, drop the whole page cache and empty every buffer before it will ultimately kill a process. At least that's what I think what it will do. I have yet to be patient enough to wait for it.
Instead of sitting in front of an unresponsive system, listening to the grinding disk for minutes, I usually press the reset button and get back to what I was doing quickly.
If you want to see what I mean, open something like the Epic Citatel HTML5 Demo in a few Firefox windows (the demo is now offline, it looked like this). Save your work to disk beforehand, though.
The downside of the reset button is that it kills all processes, whereas it would probably have been enough to kill a single one. This made people wonder if the oom-killer could be configured to step in earlier: superuser.com , unix.stackexchange.com.
As it turns out, no, it can't. At least using the in-kernel oom killer.
In the user space however, we can do whatever we want.
earlyoom checks the amount of available memory and (since version 0.5) free swap 10 times a second. If both are below 10%, it will kill the largest process. The percentage value is configurable via command line arguments.
In the free -m output below, the available memory is 2170 MiB and
the free swap is 231 MiB.
total used free shared buff/cache available
Mem: 7842 4523 137 841 3182 2170
Swap: 1023 792 231
Why is "available" memory checked as opposed to "free" memory? On a healthy Linux system, "free" memory is supposed to be close to zero, because Linux uses all available physical memory to cache disk access. These caches can be dropped any time the memory is needed for something else.
The "available" memory accounts for that. It sums up all memory that is unused or can be freed immediately.
Note that you need a recent version of
free and Linux kernel 3.14+ to see the "available" column. If you have
a recent kernel, but an old version of free, you can get the value
from cat /proc/meminfo | grep MemAvailable.
When both your available memory and free swap drop below 10% of the total,
it will kill -9 the process that uses the most memory in the opinion of
the kernel (/proc/*/oom_score). It can optionally (-i option) ignore
any positive adjustments set in /proc/*/oom_score_adj to protect innocent
victims (see below).
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. Here you can see how this looks like on my machine (Intel integrated graphics).
About 0.6MB RSS. All memory is locked using mlockall() to make sure
earlyoom does not slow down in low memory situations.
Easy:
git clone https://github.com/rfjakob/earlyoom.git
cd earlyoom
make
sudo make install # Optional, if you want earlyoom to start
# automatically as a service (works on Fedora)For Arch Linux, there's an AUR package:
yaourt -S earlyoom
sudo systemctl enable earlyoom
sudo systemctl start earlyoomJust start the executable you have just compiled:
./earlyoomIt will inform you how much memory and swap you have, what the minimum is, how much memory is available and how much swap is free.
earlyoom v0.10
mem total: 7842 MiB, min: 784 MiB (10 %)
swap total: 1023 MiB, min: 102 MiB (10 %)
mem avail: 5115 MiB (65 %), swap free: 1023 MiB (100 %)
mem avail: 5115 MiB (65 %), swap free: 1023 MiB (100 %)
mem avail: 5115 MiB (65 %), swap free: 1023 MiB (100 %)
[...]
If the values drop below the minimum, processes are killed until it is above the minimum again. Every action is logged to stderr. If you are on running earlyoom as a systemd service, you can view the last 10 lines using
systemctl status earlyoom./earlyoom -h
earlyoom v0.12
Usage: earlyoom [OPTION]...
-m PERCENT set available memory minimum to PERCENT of total (default 10 %)
-s PERCENT set free swap minimum to PERCENT of total (default 10 %)
-M SIZE set available memory minimum to SIZE KiB
-S SIZE set free swap minimum to SIZE KiB
-k use kernel oom killer instead of own user-space implementation
-i user-space oom killer should ignore positive oom_score_adj values
-d enable debugging messages
-v print version information and exit
-r INTERVAL memory report interval in seconds (default 1), set to 0 to
disable completely
-p set niceness of earlyoom to -20 and oom_score_adj to -1000
-h this help text
Bug reports and pull requests are welcome via github. In particular, I am glad to accept
- Use case reports and feedback
- v0.12: Add
-Mand-Soptions (@nailgun); add man page, parameterize Makefile (@yangfl) - v0.11: Fix undefined behavoir in get_entry_fatal (missing return, commit)
- v0.10: Allow to override Makefile's VERSION variable to make packaging easier,
add
-vcommand-line option - v0.9: If oom_score of all processes is 0, use VmRss to find a victim
- v0.8: Use a guesstimate if the kernel does not provide MemAvailable
- v0.7: Select victim by oom_score instead of VmRSS, add options
-iand-d - v0.6: Add command-line options
-m,-s,-k - v0.5: Add swap support
- v0.4: Add SysV init script (thanks @joeytwiddle), use the new
MemAvailablefrom/proc/meminfo(needs Linux 3.14+, commit) - v0.2: Add systemd unit file
- v0.1: Initial release