The kernel can be compiled without support for any memory.swap.* files, or
it can be disabled at boot time with the 'swapaccount=0' boot parameter,
so if the file doesn't exist log warning indicating the kernel doesn't
support the file and the user may need to try using the 'swapaccount=1'
boot param.
Note that the actual error from the call to fopen() is ENOENT, but
that is translated into ENODATA in cg_get_attribute_as_uint64()
In some instances, particularly with swap on zram, swap used will be high
while there is still a lot of memory available. FB OOMD handles this by
thresholding kills to X% of total swap usage. Let's do the same thing here.
Anecdotally with these thresholds and my laptop which is exclusively swap
on zram I can sit at 0K / 4G swap free with most of memory free and
systemd-oomd doesn't kill anything.
Partially addresses aggressive kill behavior from
https://bugzilla.redhat.com/show_bug.cgi?id=1941170
When this test is run in mkosi, the previously tested cgroup that we write
xattrs into and the root cgroup are the same.
Since the root cgroup is a live cgroup anyways (vs. the test cgroups which are
remade each time) let's generate the expected preference values from reading
the xattrs instead of assuming it will be NONE.
Since this is only changed the first time the limit is hit (and remains
set as long as the pressure remains over), I changed the name to better
reflect that.
Keeps consistent with "last_had_mem_reclaim" which is actually updated
every time there is reclaim activity.
systemd-oomd only monitors and kills within a selected cgroup subtree
For memory pressure kills, this means it's unnecessary to get the
pgscan rate across all the monitored memory pressure cgroups.
The increase will show up whether we do a total sum or not, but since
we only care about the increase in the subtree we're about to target
for a kill, we can simplify the code a bit by not doing this total sum.
One thing that came out of the test week is that systoomd needs to poll
more frequently so as not to race with the kernel oom killer in
situations where memory is eaten quickly. Memory pressure counters are
lagging so it isn't worthwhile to change the current read rate; however swap
is not lagging and can be checked more frequently.
So let's split these into 2 different timer events. As a result, swap
now also doesn't have to be subject to the post-action (post-kill) delay
that we need for memory pressure events.
Addresses some of slowness to kill discussed in
https://bugzilla.redhat.com/show_bug.cgi?id=1941340
Improve the logging to only print if systemd-oomd killed something. And
also print which cgroup was targeted.
Demote general swap above/pressure above messages to debug.
[zjs: fix some issuelets found in review]
This adds the same line to most of our .conf files.
Not for systemd/user.conf though, since we can't correctly display it right
now:
$ systemd-analyze cat-config --user systemd/user.conf
Option --user is not supported for cat-config right now.
For sysusers.d, tmpfiles.d, rules.d, etc, there is no single file. Maybe
we should short READMEs in /usr/lib/sysusers.d, /usr/lib/tmpfiles.d, etc.?
Inspired by #19118.
For pressure based killing we want to target who has the highest
increase in pgscan from the previous interval (vs. the previous logic
which used raw pgscan). This will prevent biasing towards long running
cgroups as mentioned in #19007.
PID1 already logs about the service being started, so this line isn't necessary
in normal use. Also, by the time it is emitted, the service has already
signalled readiness, so let's not say "starting" but "started".
oomd.conf has two parameters with fractionals: SwapUsedLimit= and
DefaultMemoryPressureLimit=, but one accepts permyriads, the other only
percentages, for no apparent reason. One carries the "Percent" in the
name, the other doesn't.
Let's clean this up: always accept permyriads, and drop the suffix,
given that it is misleading.
I figure we should internally try to focus on scaling everything
relative to UINT32_MAX, and if that isn't in the cards at least 10000,
but never permille nor percent unless there's a really really good
reason for it (e.g. interface defined by someone else).
So far OOMD limits used permyriads, as an upgrade from the original
percent.
The rest of our codebase typically scales stuff relative to UINT32_MAX.
Let's clean this up, an make sure this happens here too. This is
particularly relevant, as this is exposed in unit files and API, and
before we mark this stable we should get the APIs right.
I started working on this because I wanted to change how
DEFINE_TRIVIAL_CLEANUP_FUNC is defined. Even independently of that change, it's
nice to make make things more consistent and predictable.
There may be situations where a cgroup should be protected from killing
or deprioritized as a candidate. In FB oomd xattrs are used to bias oomd
away from supervisor cgroups and towards worker cgroups in container
tasks. On desktops this can be used to protect important units with
unpredictable resource consumption.
The patch allows systemd-oomd to understand 2 xattrs:
"user.oomd_avoid" and "user.oomd_omit". If systemd-oomd sees these
xattrs set to 1 on a candidate cgroup (i.e. while attempting to kill something)
AND the cgroup is owned by root, it will either deprioritize the cgroup as
a candidate (avoid) or remove it completely as a candidate (omit).
Usage is restricted to root owned cgroups to prevent situations where an
unprivileged user can set their own cgroups lower in the kill priority than
another user's (and prevent them from omitting their units from
systemd-oomd killing).
I think this formatting was originally used because it simplified
adding new options to the help messages. However, these days, most
tools their help message end with "\nSee the %s for details.\n" so
the final line almost never has to be edited which eliminates the
benefit of the custom formatting used for printf() help messages.
Let's make things more consistent and use the same formatting for
printf() help messages that we use everywhere else.
Prompted by https://github.com/systemd/systemd/pull/18355#discussion_r567241580
This should have been checking for any reclaim activity within a larger interval
of time rather than within the past second. On systems with swap this
doesn't seem to have mattered too much as reclaim would always increase when
memory pressure was elevated. But testing in the no swap case having
this larger interval made a difference between oomd killing or not.
By default, systemd installs various sample configuration files
containing commented-out defaults. Systems seeking to minimize the
number of files in /etc may wish to install directories and
configuration files that have semantic effects, but not install not
commented-out sample configuration files.
Turn install-sysconfdir into a multi-valued option, with a "no-samples"
value to skip installing sample-only configuration files.
Dan Streetman
Jóhann B. Guðmundsson
Zbigniew Jędrzejewski-Szmek
Anita Zhang
Luca Boccassi
Yu Watanabe
Lennart Poettering
Daan De Meyer
Josh Triplett