blob: b04ebca7e21c982631dfeea0e2f34d3f032eb3d9 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
|
#!/usr/bin/env bash
# SPDX-License-Identifier: LGPL-2.1-or-later
set -eux
set -o pipefail
# shellcheck source=test/units/util.sh
. "$(dirname "$0")"/util.sh
. /etc/os-release
# OpenSUSE does not have the stress tool packaged. It does have stress-ng but the stress-ng does not support
# --vm-stride which this test uses.
if [[ "$ID" =~ "opensuse" ]]; then
echo "Skipping due to missing stress package in OpenSUSE" >>/skipped
exit 77
fi
systemd-analyze log-level debug
# Ensure that the init.scope.d drop-in is applied on boot
test "$(cat /sys/fs/cgroup/init.scope/memory.high)" != "max"
# Loose checks to ensure the environment has the necessary features for systemd-oomd
[[ -e /proc/pressure ]] || echo "no PSI" >>/skipped
[[ "$(get_cgroup_hierarchy)" == "unified" ]] || echo "no cgroupsv2" >>/skipped
[[ -x /usr/lib/systemd/systemd-oomd ]] || echo "no oomd" >>/skipped
if [[ -s /skipped ]]; then
exit 77
fi
rm -rf /run/systemd/system/TEST-55-OOMD-testbloat.service.d
# Activate swap file if we are in a VM
if systemd-detect-virt --vm --quiet; then
swapoff --all
if [[ "$(findmnt -n -o FSTYPE /)" == btrfs ]]; then
btrfs filesystem mkswapfile -s 64M /swapfile
else
dd if=/dev/zero of=/swapfile bs=1M count=64
chmod 0600 /swapfile
mkswap /swapfile
fi
swapon /swapfile
swapon --show
fi
# Configure oomd explicitly to avoid conflicts with distro dropins
mkdir -p /run/systemd/oomd.conf.d/
cat >/run/systemd/oomd.conf.d/99-oomd-test.conf <<EOF
[OOM]
DefaultMemoryPressureDurationSec=2s
EOF
mkdir -p /run/systemd/system/-.slice.d/
cat >/run/systemd/system/-.slice.d/99-oomd-test.conf <<EOF
[Slice]
ManagedOOMSwap=auto
EOF
mkdir -p /run/systemd/system/user@.service.d/
cat >/run/systemd/system/user@.service.d/99-oomd-test.conf <<EOF
[Service]
ManagedOOMMemoryPressure=auto
ManagedOOMMemoryPressureLimit=0%
EOF
mkdir -p /run/systemd/system/systemd-oomd.service.d/
cat >/run/systemd/system/systemd-oomd.service.d/debug.conf <<EOF
[Service]
Environment=SYSTEMD_LOG_LEVEL=debug
EOF
systemctl daemon-reload
# enable the service to ensure dbus-org.freedesktop.oom1.service exists
# and D-Bus activation works
systemctl enable systemd-oomd.service
# if oomd is already running for some reasons, then restart it to make sure the above settings to be applied
if systemctl is-active systemd-oomd.service; then
systemctl restart systemd-oomd.service
fi
if [[ -v ASAN_OPTIONS || -v UBSAN_OPTIONS ]]; then
# If we're running with sanitizers, sd-executor might pull in quite a significant chunk of shared
# libraries, which in turn causes a lot of pressure that can put us in the front when sd-oomd decides to
# go on a killing spree. This fact is exacerbated further on Arch Linux which ships unstripped gcc-libs,
# so sd-executor pulls in over 30M of libs on startup. Let's make the MemoryHigh= limit a bit more
# generous when running with sanitizers to make the test happy.
systemctl edit --runtime --stdin --drop-in=99-MemoryHigh.conf TEST-55-OOMD-testchill.service <<EOF
[Service]
MemoryHigh=60M
EOF
# Do the same for the user instance as well
mkdir -p /run/systemd/user/
cp -rfv /run/systemd/system/TEST-55-OOMD-testchill.service.d/ /run/systemd/user/
else
# Ensure that we can start services even with a very low hard memory cap without oom-kills, but skip
# under sanitizers as they balloon memory usage.
systemd-run -t -p MemoryMax=10M -p MemorySwapMax=0 -p MemoryZSwapMax=0 /bin/true
fi
systemctl start TEST-55-OOMD-testchill.service
systemctl start TEST-55-OOMD-testbloat.service
# Verify systemd-oomd is monitoring the expected units
timeout 1m bash -xec 'until oomctl | grep "/TEST-55-OOMD-workload.slice"; do sleep 1; done'
oomctl | grep "/TEST-55-OOMD-workload.slice"
oomctl | grep "20.00%"
oomctl | grep "Default Memory Pressure Duration: 2s"
systemctl status TEST-55-OOMD-testchill.service
# systemd-oomd watches for elevated pressure for 2 seconds before acting.
# It can take time to build up pressure so either wait 2 minutes or for the service to fail.
for _ in {0..59}; do
if ! systemctl status TEST-55-OOMD-testbloat.service; then
break
fi
oomctl
sleep 2
done
# testbloat should be killed and testchill should be fine
if systemctl status TEST-55-OOMD-testbloat.service; then exit 42; fi
if ! systemctl status TEST-55-OOMD-testchill.service; then exit 24; fi
# Make sure we also work correctly on user units.
loginctl enable-linger testuser
systemctl start --machine "testuser@.host" --user TEST-55-OOMD-testchill.service
systemctl start --machine "testuser@.host" --user TEST-55-OOMD-testbloat.service
# Verify systemd-oomd is monitoring the expected units
# Try to avoid racing the oomctl output check by checking in a loop with a timeout
timeout 1m bash -xec 'until oomctl | grep "/TEST-55-OOMD-workload.slice"; do sleep 1; done'
oomctl | grep -E "/user.slice.*/TEST-55-OOMD-workload.slice"
oomctl | grep "20.00%"
oomctl | grep "Default Memory Pressure Duration: 2s"
systemctl --machine "testuser@.host" --user status TEST-55-OOMD-testchill.service
# systemd-oomd watches for elevated pressure for 2 seconds before acting.
# It can take time to build up pressure so either wait 2 minutes or for the service to fail.
for _ in {0..59}; do
if ! systemctl --machine "testuser@.host" --user status TEST-55-OOMD-testbloat.service; then
break
fi
oomctl
sleep 2
done
# testbloat should be killed and testchill should be fine
if systemctl --machine "testuser@.host" --user status TEST-55-OOMD-testbloat.service; then exit 42; fi
if ! systemctl --machine "testuser@.host" --user status TEST-55-OOMD-testchill.service; then exit 24; fi
loginctl disable-linger testuser
# only run this portion of the test if we can set xattrs
if cgroupfs_supports_user_xattrs; then
sleep 120 # wait for systemd-oomd kill cool down and elevated memory pressure to come down
mkdir -p /run/systemd/system/TEST-55-OOMD-testbloat.service.d/
cat >/run/systemd/system/TEST-55-OOMD-testbloat.service.d/override.conf <<EOF
[Service]
ManagedOOMPreference=avoid
EOF
systemctl daemon-reload
systemctl start TEST-55-OOMD-testchill.service
systemctl start TEST-55-OOMD-testmunch.service
systemctl start TEST-55-OOMD-testbloat.service
for _ in {0..59}; do
if ! systemctl status TEST-55-OOMD-testmunch.service; then
break
fi
oomctl
sleep 2
done
# testmunch should be killed since testbloat had the avoid xattr on it
if ! systemctl status TEST-55-OOMD-testbloat.service; then exit 25; fi
if systemctl status TEST-55-OOMD-testmunch.service; then exit 43; fi
if ! systemctl status TEST-55-OOMD-testchill.service; then exit 24; fi
fi
systemd-analyze log-level info
touch /testok
|
