Skip to content

MRuby on Container / A Linux container runtime using mruby DSL for configuration, control and hooks

License

GPL-3.0 and 2 other licenses found

Licenses found

GPL-3.0
LICENSE
LGPL-2.1
LICENSE_libcgroup
MIT
LICENSE_mruby
Notifications You must be signed in to change notification settings

haconiwa/haconiwa

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Haconiwa

The logo

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

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:

VERSION=<specify version>
wget https://github.com/haconiwa/haconiwa/releases/download/v${VERSION}/haconiwa-v${VERSION}.x86_64-pc-linux-gnu.tgz
tar xzf haconiwa-v${VERSION}.x86_64-pc-linux-gnu.tgz
sudo install hacorb hacoirb haconiwa /usr/local/bin
haconiwa
# 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.

Example

Bootstraping container filesystem

Create mount points:

$ mkdir -p /var/another/root/etc
$ mkdir -p /var/another/root/home

Create the file example.haco:

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

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

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

  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"
end

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.

Running

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

Bootstrap

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"

Provision

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.network - Specify networking config. Network#namespace and #container_ip must be set. Other attributes are #bridge_name, #bridge_ip, #veth_host, #veth_guest
  • 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
  • config.support_reload - Specify reloadable parameters when invoked haconiwa reload command. Only :cgroup and :resource are available for now and it is active only when they are defined following configuration blocks. See test cases and examples
  • config.wait_interval - Specify the sleep interval in wait and watchdog loops by milli seconds
  • config.metadata - Add container's metadata(tagging) by ruby Hash
  • config.lxcfs_root - Set your host's lxcfs mount point to cooperate with containers

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.

Hooks

  • config.add_general_hook(hookpoint, &block) - Define hook codes that are invoked through the Haconiwa's spawning process. Hook points are below:
    • :before_fork - Hooked just before the container process is forked
    • :after_fork - Hooked just after the container process is forked, in forked process
    • :before_chroot - Hooked just after container settins are applied (e.g. namespace, cgroup, caps, fs mounting) and just before do chroot in forked process
    • :after_chroot - Hooked just after the chroot is successful, in forked process. This is the last timing before doing exec() and becoming a new program
    • :before_start_wait - Hooked before starting to wait() the container process. Hook itself is invoked in the parent process
    • :teardown_container - Hooked after the container process has quitted, in the parent process. base.exit_status is set
    • :after_reload - Hooked just after haconiwa reload is invoked and successful
    • :after_failure - Hooked just after tha container process is exited with failure. base.exit_status is set
    • Every hook can accept one argument base, which is Haconiwa::Base object.
  • hooks below are system hooks, which are invoked in across-supervisor layer.
    • :setup - Hooked just before the supervisor processes are going to be forked
    • :teardown - Hooked after the supervisor processes have quitted all
    • :system_failure - Hooked just after tha container process is exited with failure
      • Either barn.exit_status or barn.system_exception will be available
    • Every hook can accept one argument barn, which is Haconiwa::Barn object.
  • config.add_async_hook(option, &block) - Define timer handler. Supported options:
    • :msec/:sec/:min/:hour - First timeout to invoke hook.
    • :interval_msec - Define the interval timeout hooks, if this paramete exists.
    • NOTE: Async hook uses POSIX timer and real-time signals internally. Please do not queue real-time signals directly.
  • 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.
  • config.validate_real_id(&block{|ruid, rgid| ... }) - Validates the Real UID/GID who invoked haconiwa command. return true if OK, false/nil to mark invalid. This is useful when haconiwa command is set-user-ID root

Please check out sample directory.

Programming the container world by mruby

e.g.:

Namespace.unshare(Namespace::CLONE_NEWNS)
Namespace.unshare(Namespace::CLONE_NEWPID)

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

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

c = Process.fork {
  Mount.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.

Development

  • 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

TODOs

  • Haconiwa DSL compiler
  • Networking helpers
  • P2P containers

License

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

Haconinwa's logo is originally created by @takeshige, and is under Creative Commons License BY-NC-ND 4.0. CC BY-NC-ND

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 github.com repository.