MRuby on Container / helper tools with DSL for your handmade linux containers, or container clusters
Ruby HTML Shell Other
Latest commit f64c6d2 Jan 16, 2017 @udzura udzura committed on GitHub Merge pull request #68 from haconiwa/fix-double-dot-cgroup
Fix cannot set cgroup parameters including 2 dots, such as `"memory.memsw.limit_in_bytes"`


Build Status

mRuby on Container / helper tools with DSL for your handmade linux containers.

Haconiwa (箱庭 - a miniature garden) is a container builder DSL, by which you can choose any container-related technologies as you like:

  • Linux namespace
  • Linux control group(cgroup)
  • Linux capabilities
  • Bind mount / chroot
  • Resource limit(rlimit)
  • setuid/setgid
  • ...

Haconiwa is written in mruby, so you can utilize Ruby DSL for creating your own container.

Install binary

haconiwa packages are provided via packagecloud.

Available for: CentOS >= 7 / CentOS ~> 6(Experimental, maybe kernel update required) / Fedora >= 23 / Ubuntu Trusty / Ubuntu Xenial / Debian jessie (which are supported by best effort...)

(PR: We are welcoming package maintainers for other distro!!!)

Other linuxes users can just download binaries from latest:

tar xzf haconiwa-v${VERSION}.x86_64-pc-linux-gnu.tgz
sudo install hacorb hacoirb haconiwa /usr/local/bin
# haconiwa - The MRuby on Container
# commands:
#     run       - run the container
#     attach    - attach to existing container
#     ...

NOTE: If you'd like using cgroup-related features, install cgroup package such as cgroup-lite (Ubuntu) or cgroup-bin (Debian). If you would not, these installation are not required.


Bootstraping container filesystem

Create the file example.haco:

Haconiwa::Base.define do |config| = "new-haconiwa001" # to be hostname

  config.bootstrap do |b|
    b.strategy = "lxc"
    b.os_type = centos

  config.provision do |p|
    p.run_shell "yum -y install git"

  config.cgroup["cpu.shares"] = 2048
  config.cgroup["memory.limit_in_bytes"] = 256 * 1024 * 1024
  config.cgroup["pids.max"] = 1024

  config.add_mount_point "/var/another/root/etc", to: "/var/your_rootfs/etc", readonly: true
  config.add_mount_point "/var/another/root/home", to: "/var/your_rootfs/home"
  config.mount_independent "procfs"
  config.mount_independent "sysfs"
  config.chroot_to "/var/your_rootfs"

  config.namespace.unshare "ipc"
  config.namespace.unshare "uts"
  config.namespace.unshare "mount"
  config.namespace.unshare "pid"

  config.capabilities.allow :all
  config.capabilities.drop "cap_sys_admin"

Then run the haconiwa create command to set up container base root filesystem.

$ haconiwa create example.haco
Start bootstrapping rootfs with lxc-create...

To re-run provisioning, you can use haconiwa provision.


Then use haconiwa run command to make container up.

$ haconiwa run example.haco

When you want to attach existing container:

$ haconiwa attach example.haco

Note: attach subcommand allows to set PID(--target) or container name(--name) for dynamic configuration. And attach is not concerned with capabilities which is granted to container. So you can drop or allow specific caps with --drop/--allow.

DSL spec


config.bootstrap block support 6 strategies now.

  • strategy = "lxc"
    • needs lxc-create command
    • lxc.project_name to set PJ name. default to the dirname
    • lxc.os_type to set OS type installed to
  • strategy = "debootstrap"
    • needs debootstrap command
    • deb.variant to set Debian variant param to pass debootstrap
    • deb.debian_release to set Debian's release name squeeze/jessie/sid and so on...
    • deb.mirror_url to set mirror URL debootstrap uses
    • deb.components to set components installed. eg, 'base'
  • strategy = "git"
    • needs git command :)
    • git.git_url to set the repository URL for clone
    • git.git_options to set extra git options by Array, if necessary
  • strategy = "tarball"
    • needs tar command ;)
    • tb.archive_path to set the source archive path on your host machine
    • tb.tar_options to set extra tar options by Array, if necessary
  • strategy = "shell" / "mruby"
    • shell.code to set a shell or mruby code by string(heredoc is OK). You can pass the mruby code block for "mruby"


config.provision block support some operations(in the future. now run_shell only).

  • run_shell to set plane shell script(automatically set -xe-ed on run)
    • We can declare run_shell multiple times
    • Set name by name: option, then you can specify provision operation by haconiwa provision --run-only=...

Running container environment

  • config.environ - A hash to pass environment variables to a created container. e.g. config.environ = {"FOO_KEY" => "value", ...}
  • config.workdir - The working directory of haconiwa's init command
  • config.command.set_stdout/set_stderr - Emit command's stdout/err to specified files. This is active only on daemon mode
  • config.resource.set_limit - Set the resource limit of container, using setrlimit
  • config.cgroup - Assign cgroup parameters via []=
  • config.namespace.unshare - Unshare the namespaces like "mount", "ipc" or "pid" .... persist_in option make the specified namespace persist in a bind-moounted-file
  • config.capabilities.reset_to_privileged! - Haconiwa has default capability whitelist to use. If you want to use customized black/whitelist, declare this first
  • config.capabilities.allow - Allow capabilities on container root. Setting parameters other than :all should make this acts as whitelist
  • config.capabilities.drop - Drop capabilities of container root. Default to act as blacklist
  • config.add_mount_point - Add the mount point of container. Source directory is resolved from the directory where a user run haconiwa
  • config.mount_independent - Mount the independent filesystems: "procfs", "sysfs", "devtmpfs", "devpts" and "shm" in the newborn container. Useful if "pid" or "net" are unshared
  • config.chroot_to - The new chroot root
  • config.uid=/config.gid= - The new container's running uid/gid. groups= is also respected

You can pick your own parameters for your use case of container. e.g. just using mount namespace unshared, container with common filesystem, limit the cgroups for big resource job and so on.


  • config.after_spawn(option, &block) - Define timer handler. Pass option like msec: 10 * 60 * 1000, then after 10 min passed, the defined hook will be invoked
  • config.add_signal_handler(signame, &block) - Define signal handler at supervisor process(not container itself). Available signals are SIGTTIN/SIGTTOU/SIGUSR1/SIGUSR2. See handler example.

Please look into sample directory.


Container management with etcd

If there is a /etc/haconiwa.conf.rb file like below, as a global config:

Haconiwa.configure do |config|
  config.etcd_url = "http://localhost:2379/v2"

haconiwa cli understands this DSL and uses etcd as a container database backend, with enabled clustering. We recommend to set etcd's server name to host's primary LAN IP(such as: ETCD_NAME="").

You can run haconiwa ps to check container processes (all over the hosts clustered by etcd!).

Programming the container world by mruby



m =

m.make_private "/"
m.bind_mount "/var/lib/myroot", "/var/lib/haconiwa/root"

Dir.chroot "/var/lib/haconiwa"
Dir.chdir "/"

c = Process.fork {
  m.mount "proc", "/proc", :type => "proc"
  Exec.exec "/bin/sh"
pid, ret = Process.waitpid2 c
puts "Container exited with: #{ret.inspect}"

See dependent gem's READMEs.


  • rake compile will create binaries.
  • rake won't be passed unless you are not on Linux.
  • This project is built upon great mruby-cli. Please browse its README.

Release policy

  • Versions whose minor versions are even numbers (0.6, 0.8, 0.10, 1.0...): Stable release
  • Versions whose minor versions are odd numbers (0.7, 0.9, 0.11, 1.1...): Unstable release. Features added at this version should be broken
  • I introduced this policy after version 0.5.x
  • We create branches as 0.6.x-dev for release
  • PRs can be proposed to master branch. Maintainers will pick these to stable/unstable dev branches


Bug reports and pull requests are welcome on GitHub at This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the Contributor Covenant code of conduct.


  • Haconiwa DSL compiler
  • Networking helpers
  • P2P containers


Haconiwa core is under the GPL v3 License: See LICENSE file.

Bundled libraries (libcap, libcgroup, libargtable and mruby) are licensed by each authors. See LICENSE_* file.

For other mgems' licenses, especially ones which are not bundled by mruby-core, please refer their repository.