toralf
Developer
Joined: 01 Feb 2004
Posts: 3934
Location: Hamburg
|
 Posted: Sun Jan 24, 2021 11:40 am Post subject: switch from cgroups v1 to v2 |
 |
|
I do wonder if it has an advantage and if - how can I convert my current logic: | Code: | #!/bin/sh
# set -x
# set cgroup v1 limits
set -euf
export PATH="/usr/sbin:/usr/bin:/sbin:/bin"
export LANG=C.utf8
# use cgroup v1 if available
if ! hash -r cgcreate || ! hash -r cgset || ! -d /sys/fs/cgroup; then
exit 0
fi
# make this script available for non-tinderbox consumers too
cp /opt/tb/bin/cgroup-release-agent.sh /usr/local/bin/
chmod 755 /usr/local/bin/cgroup-release-agent.sh
echo 1 > /sys/fs/cgroup/memory/memory.use_hierarchy
for i in cpu memory
do
echo "/usr/local/bin/cgroup-release-agent.sh" > /sys/fs/cgroup/$i/release_agent
done
# prefer a generic name b/c "tinderbox" is just one consumer of CGroups
name=/local
cgcreate -g cpu,memory:$name
# reserve 8 vCPU for members of $name, period=0.1 sec
cgset -r cpu.cfs_quota_us=800000 $name
# reserve 120 GB RAM (8GB left) and 140 GB RAM+swap
cgset -r memory.limit_in_bytes=120G $name
cgset -r memory.memsw.limit_in_bytes=140G $name
|
to have asubgroup called local which is used later as | Code: | function CgroupCreate() {
local name=$1
local pid=$2
# use cgroup v1 if available
if ! hash -r cgcreate || ! hash -r cgset || ! -d /sys/fs/cgroup; then
return 0
fi
cgcreate -g cpu,memory:$name
cgset -r cpu.cfs_quota_us=150000 $name
cgset -r memory.limit_in_bytes=20G $name
cgset -r memory.memsw.limit_in_bytes=30G $name
for i in cpu memory
do
echo 1 > /sys/fs/cgroup/$i/$name/notify_on_release
echo "$pid" > /sys/fs/cgroup/$i/$name/tasks
done
} |
|
|
Kernel & Hardware
