Contents
- 1
cc-oci-runtime
- 1.1 Overview
- 1.2 Quick Start
- 1.3 Requirements
- 1.4 Platform Support
- 1.5 Supported Application Versions
- 1.6 Design
- 1.7 Limitations
- 1.8 Running under
docker
- 1.9 Running under
containerd
(without Docker) - 1.10 Running stand-alone
- 1.11 Community
- 1.12 Building
- 1.13 Tests
- 1.14 Configuration
- 1.15 Logging
- 1.16 Command-line Interface
- 1.17 Development
- 1.18 Debugging
- 1.19 Links
cc-oci-runtime
is an Open Containers Initiative (OCI) "runtime"
that launches an Intel VT-x secured Clear Containers 2.0 hypervisor,
rather than a standard Linux container. It leverages the highly
optimised Clear Linux technology to achieve this goal.
The tool is licensed under the terms of the GNU GPLv2 and aims to be compatible with the OCI runtime specification [1], allowing Clear Containers to be launched transparently by Docker (using containerd) and other OCI-conforming container managers.
Henceforth, the tool will simply be referred to as "the runtime".
See the canonical cc-oci-runtime home page for the latest information.
If you are eager to install and start using Clear Containers, please refer to the installation instructions on the wiki:
- A Qemu hypervisor that supports the
pc-lite
machine type (see qemu-lite). - CONFIG_VHOST_NET enabled in the host kernel
the runtime supports running Clear Containers on Intel 64-bit (x86-64) Linux systems.
The runtime has been tested with the following application versions:
- Docker version 1.12.1.
- Containerd version 0.2.2.
See the design document.
See:
Assuming a Docker 1.12 environment, start the Docker daemon specifying
the "--add-runtime $alias=$path
" option. For example:
$ sudo dockerd --add-runtime cor=/usr/bin/cc-oci-runtime
Then, to run a Clear Container using the runtime, specify "--runtime cor
".
For example:
$ sudo docker run --runtime cor -ti busybox
Note that if you wish to pass options to the runtime such as
--global-log
(see Logging) and --debug
(see Debugging), you
should instead configure Docker to invoke the helper script like this:
$ sudo dockerd --add-runtime cor=/usr/bin/cc-oci-runtime.sh
The helper script will call the real runtime binary with the options you
specify. Runtime options can either be passed to the runtime binary by
creating a cc-oci-runtime.sh.cfg
configuration file, or by modifying
/usr/bin/cc-oci-runtime.sh
directly. See the contents of this script
for instructions.
If you are running Containerd directly, without Docker:
Start the server daemon:
$ sudo /usr/local/bin/containerd --debug --runtime $PWD/cc-oci-runtime
Launch a hypervisor:
$ name=foo # XXX: path to directory containing atleast the following: # # config.json # rootfs/ # $ bundle_dir=... $ sudo /usr/local/bin/ctr --debug containers start --attach "$name" "$bundle_dir"
Forcibly stop the hypervisor:
$ name=foo $ sudo ./cc-oci-runtime stop "$name"
The runtime can be run directly, without the need for either docker
or containerd
:
$ name=foo $ pidfile=/tmp/cor.pid $ logfile=/tmp/cor.log $ sudo ./cc-oci-runtime --debug create --console $(tty) --bundle "$bundle_dir" "$name" $ sudo ./cc-oci-runtime --debug start "$name"
Or, to simulate how containerd
calls the runtime:
$ sudo ./cc-oci-runtime --log "$logfile" --log-format json create --bundle "$bundle_dir" --console $(tty) --pid-file "$pidfile" "$name" $ sudo ./cc-oci-runtime --log "$logfile" --log-format json start "$name"
Assuming the following provisos, the runtime can be run as a non-root
user:
- User has read+write permissions for the Clear Containers root
filesystem image specified in the
vm
JSON object (see Configuration). - User has read+execute permissions for the Clear Containers kernel
image specified in the
vm
JSON object (see Configuration). - The bundle configuration file ("
config.json
") does not specify any mounts that the runtime must honour. - The runtime is invoked with the "
--root=$dir
" option where "$dir
" is a pre-existing directory that the user has write permission to.
To run non-privileged:
$ name=foo $ dir=/tmp/cor $ mkdir -p "$dir" $ ./cc-oci-runtime --root "$dir" create --console $(tty) --bundle "$oci_bundle_directory" "$name" $ ./cc-oci-runtime --root "$dir" start "$name"
Ensure you have the development versions of the following packages installed on your system:
- check
- glib
- json-glib
- uuid
- libmnl
Quick start, just run:
$ ./autogen.sh && make
If you have specific requirements, run:
$ ./configure --help
... then add the extra "configure
" flags you want to use:
$ ./autogen.sh --enable-foo --disable-bar && make
To run the basic unit tests, run:
$ make check
To configure the command above to also run the functional tests (recommended), see the functional tests README.
At the time of writing, the OCI had not agreed on how best to handle VM-based runtimes such as this (see [3]).
Until the OCI specification clarifies how VM runtimes will be defined, the runtime will search a number of different data sources for its VM configuration information.
Unless otherwise specified, each configuration file in this section will be looked for in the following directories (in order):
The bundle directory, specified by "
--bundle $bundle_dir
".The system configuration directory ("
./configure --sysconfdir=...
").With no
--prefix
or with--prefix=/
, the file will be looked for in/etc/cc-oci-runtime/
".The defaults directory.
This is a directory called "
defaults/cc-oci-runtime/
" below the configured data directory ("./configure --datadir=...
").With no
--prefix
or with--prefix=/
, the file will be looked for in/usr/share/defaults/cc-oci-runtime/
".
The first file found will be used and the runtime will log the full path to each configuration file used (see Logging).
Example files will be available in the "data/
" directory after the
build has completed. To influence the way these files are generated,
consider using the following "configure
" options:
--with-qemu-path=
--with-cc-kernel=
--with-cc-image=
Note
You may still need to make adjustments to this file to work for your environment.
The runtime will consult the OCI configuration file config.json
for a "vm
" object, according to the proposed OCI specification
[3]
If no "vm
" object is found in config.json
, the file
"vm.json
" will be looked for which should contain a stand-alone
JSON "vm
" object specifying the virtual machine configuration.
This file specifies both the full path to the hypervisor binary to use and all the arguments to be passed to it. The runtime supports specifying specific options using variables (see Variable Expansion).
Currently, the runtime will expand the following special tags found in
hypervisor.args
appropriately:
@COMMS_SOCKET@
- path to the hypervisor control socket (QMP socket for qemu).@CONSOLE_DEVICE@
- hypervisor arguments used to control where console I/O is sent to.@IMAGE@
- Clear Containers rootfs image path (read fromconfig.json
).@KERNEL_PARAMS@
- kernel parameters (fromconfig.json
).@KERNEL@
- path to kernel (fromconfig.json
).@NAME@
- VM name.@PROCESS_SOCKET@
- required to detect when hypervisor has started running, and when it has shut down.@SIZE@
- size of @IMAGE@ which is auto-calculated.@UUID@
- VM uuid.@WORKLOAD_DIR@
- path to workload chroot directory that will be mounted (via 9p) inside the VM.@AGENT_CTL_SOCKET@
- path to the guest agent control socket ( control serial port for hyperstart)@AGENT_TTY_SOCKET@
- path to the guest agent multiplex tty I/O socket ( tty serial port for hyperstart)
The runtime logs to the file specified by the global --log
option.
However, it can also write to a global log file if the
--global-log
option is specified. Note that if both log options are
specified, both log files will be appended to.
The global log potentially provides more detail than the standard log since it is always written to in ASCII format and includes Process ID details. Also note that all instances of the runtime will append to the global log.
Additionally exist the possibility to log hypervisor's stderr and stdout into
$hypervisorLogDir/$containerId-hypervisor.stderr
and
$hypervisorLogDir/$containerId-hypervisor.stdout
respectively if the
--hypervisor-log-dir
option is specified. Note that $hypervisorLogDir
and $containerId
are variables provided by user through
--hypervisor-log-dir
option and create
command respectively.
At the time of writing, the OCI has provided recommendations for the runtime command line interface (CLI) (see [2]).
However, the OCI runtime reference implementation, runc, has a CLI which deviates from the recommendations.
This issue has been raised with OCI (see [4]), but until the situation is clarified, the runtime strives to support both the OCI CLI and the runc CLI interfaces.
Details of the runc command line options can be found in the runc manpage.
Note: Next arguments are unique to the runtime at present:
- --global-log
- --hypervisor-log-dir
- --shim-path
- --proxy-socket-path
The list
command supports a "--all
" option that provides
additional information including details of the resources used by the
virtual machine.
Follow the instructions in Building, but you will also want to install:
- doxygen
- lcov
- valgrind
To build the API documentation:
$ doxygen Doxyfile
Then, point your browser at /tmp/doxygen-cc-oci-runtime
. If you
don't like that location, change the value of OUTPUT_DIRECTORY
in
the file Doxyfile
.
Specify the
--enable-debug
configure option to theautogen.sh
script which enable debug output, but also disable all compiler and linker optimisations.If you want to see the hypervisor boot messages, remove "quiet" from the hypervisor command-line in "
hypervisor.args
".Run with the "
--debug
" global option.If you want to debug as a non-
root
user, specify the "--root
" global option. For example:$ gdb --args ./cc-oci-runtime \ --debug \ --root /tmp/cor/ \ --global-log /tmp/global.log \ --hypervisor-log-dir /tmp/ \ start --console $(tty) $container $bundle_path
Consult the global Log (see Logging).
[1] | https://github.com/opencontainers/runtime-spec |
[2] | https://github.com/opencontainers/runtime-spec/blob/master/runtime.md |
[3] | (1, 2) opencontainers/runtime-spec#405 |
[4] | opencontainers/runtime-spec#434 |