test/units/testsuite-55.sh

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