containers-arent-magic.sh

# you have to run this as root
# It only runs on Linux (namespaces and cgroups only exist on Linux)
# if you don't have it, cgcreate is in the libcgroup package

set -eux # let's be safe

# Download the container (it's in a github gist published by my github account)
# This is just the frapsoft/fish Fish Docker container flattened into a single tarball
# You can also easily make your own tarball to run instead of this one  with `docker export`
wget bit.ly/fish-container -O fish.tar

# extract fish.tar into a directory
mkdir container-root
cd container-root
tar -xf ../fish.tar
# generate a random cgroup id
uuid="cgroup_$(shuf -i 1000-2000 -n 1)"
# create the cgroup
cgcreate -g "cpu,cpuacct,memory:$uuid"
# assign CPU/memory limits to the cgroup
cgset -r cpu.shares=512 "$uuid"
cgset -r memory.limit_in_bytes=1000000000 "$uuid"
# The following line does a lot of work:
# 1. cgexec: use our new cgroup
# 2. unshare: make and use a new PID, network, hostname, and mount namespace
# 3. chroot: change root directory to current directory
# 4. mount: use the right /proc in our new mount namespace
# 5. hostname: change the hostname in the new hostname namespace to something fun
cgexec -g "cpu,cpuacct,memory:$uuid" \
    unshare -fmuipn --mount-proc \
    chroot "$PWD" \
    /bin/sh -c "/bin/mount -t proc proc /proc && hostname container-fun-times && /usr/bin/fish"

# Here are ome fun things to try once you're running your container!
# Run them both in the container and in a normal shell and see the difference
# - ps aux
# - ifconfig
# - hostname

Hi Julia! First of all, this is awesome!

I just wanted to add that cgcreate comes in the libcgroup-tools package(rhel based distros), to save some time to the ppl that are trying to run the script and already have the libcgroup package but don't have the cgroup utilities.

This didn't work on Oracle Linux Server release 7.4 may be some commands are different ?

What commands didn't work exactly? You might need to install some tools, like the above comment mentioned about the libcgroup-tools package

@jvns I've updated it a bit to check for existing files/directories when run multiple times. But also updates the commands for cgroups v2!

There's some new controllers and a single hierarchy design. For example, for memory there are these two pertinent values

• memory.high: the kernel will attempt to keep memory usage below this configuration.
• memory.max: if memory reaches this level the OOM killer (a system used to sacrifice one or more processes to free up memory for the system when all else fails) is invoked on the cgroup.

For cpu's, I haven't quite figured it out yet haha. There is no more cpuacct, the new cpu controller superseded cpu and cpuacct. However, I don't understand exactly how cpu.max works yet. It can take a range from 0 to 100000. 20000 seems to work fine for this demo..

# you have to run this as root
# It only runs on Linux (namespaces and cgroups only exist on Linux)
# if you don't have it, cgcreate is in the libcgroup package

set -eux # let's be safe

# Download the container (it's in a github gist published by my github account)
# This is just the frapsoft/fish Fish Docker container flattened into a single tarball
# You can also easily make your own tarball to run instead of this one  with `docker export`
if [[ ! -f "$(pwd)/fish.tar" ]]; then
  wget bit.ly/fish-container -O fish.tar
fi

# prepare cgroup directory
if [[ -d "$(pwd)/container-root" ]]; then
  rm -r container-root
fi

mkdir container-root
cd container-root

# extract fish.tar into a directory
tar -xf ../fish.tar

# generate a random cgroup id
uuid="cgroup_$(shuf -i 1000-2000 -n 1)"

# create the cgroup
cgcreate -g "cpu,memory:$uuid"

# assign CPU/memory limits to the cgroup
cgset -r cpu.max=20000 "$uuid"
cgset -r memory.max=1000000000 "$uuid"

# The following line does a lot of work:
# 1. cgexec: use our new cgroup
# 2. unshare: make and use a new PID, network, hostname, and mount namespace
# 3. chroot: change root directory to current directory
# 4. mount: use the right /proc in our new mount namespace
# 5. hostname: change the hostname in the new hostname namespace to something fun
cgexec -g "cpu,memory:$uuid" \
  unshare -fmuipn --mount-proc \
  chroot "$PWD" \
  /bin/sh -c "/bin/mount -t proc proc /proc && hostname container-fun-times && /usr/bin/fish"

# Here are ome fun things to try once you're running your container!
# Run them both in the container and in a normal shell and see the difference
# - ps aux
# - ifconfig
# - hostname

Runs out of the box on an up-to-date Arch system (after installing libcgroup). Ubuntu 22.04 should also have cgroups v2 by default now.