Christian Berendt c3d8262ec5 Add doc8 test and improve rst syntax
This will test all rst files inside the doc directory for style issues with
doc8 (an opinionated style checker for rst styles of documentation).

This will fix all syntax issues identified by doc8 and will improve
the syntax.

Change-Id: Id1b9563e07e77e306aef5a0767c98c27f87c5c0e
2016-08-04 15:09:10 +02:00

20 KiB

Quick Start

This guide provides a step by step of how to deploy Kolla on bare metal or a virtual machine.

Host machine requirements

The recommended deployment target requirements:

  • 2 (or more) network interfaces.
  • At least 8gb main memory
  • At least 40gb disk space.


Some commands below may require root permissions (e.g. pip, apt-get).

If developing or evaluating Kolla, the community strongly recommends using bare metal or a virtual machine. Follow the instructions in this document to get started with deploying OpenStack on bare metal or a virtual machine with Kolla. There are other deployment environments referenced below in Additional Environments.

Install Dependencies

Kolla is tested on CentOS, Oracle Linux, RHEL and Ubuntu as both container OS platforms and bare metal deployment targets.

Fedora: Kolla will not run on Fedora 22 and later as a bare metal deployment target. These distributions compress kernel modules with the .xz compressed format. The guestfs system in the CentOS family of containers cannot read these images because a dependent package supermin in CentOS needs to be updated to add .xz compressed format support.

Ubuntu: For Ubuntu based systems where Docker is used it is recommended to use the latest available LTS kernel. The latest LTS kernel available is the wily kernel (version 4.2). While all kernels should work for Docker, some older kernels may have issues with some of the different Docker backends such as AUFS and OverlayFS. In order to update kernel in Ubuntu 14.04 LTS to 4.2, run:

apt-get -y install linux-image-generic-lts-wily


Install is very sensitive about version of components. Please review carefully because default Operating System repos are likely out of date.

Component Min Version Max Version Comment
Ansible 2.0.0 none On deployment host
Docker 1.10.0 none On target nodes
Docker Python 1.6.0 none On target nodes
Python Jinja2 2.8.0 none On deployment host

Make sure the pip package manager is installed and upgraded to latest before proceeding:

# CentOS 7
yum -y install epel-release
yum -y install python-pip

# Ubuntu 14.04 LTS
apt-get -y install python-pip

# Upgrade pip and check version
pip install -U pip
pip -V

Since Docker is required to build images as well as be present on all deployed targets, the Kolla community recommends installing the official Docker, Inc. packaged version of Docker for maximum stability and compatibility with the following command:

curl -sSL https://get.docker.io | bash

This command will install the most recent stable version of Docker, but please note that Kolla releases are not in sync with docker in any way, so some things could stop working with new version. The latest release of Kolla is tested to work with docker-engine >= 1.10.0. To check your docker version run this command:

docker --version

When running with systemd, setup docker-engine with the appropriate information in the Docker daemon to launch with. This means setting up the following information in the docker.service file. If you do not set the MountFlags option correctly then kolla-ansible will fail to deploy the neutron-dhcp-agent container and throws APIError/HTTPError. After adding the drop-in unit file as follows, reload and restart the docker service:

# Create the drop-in unit directory for docker.service
mkdir -p /etc/systemd/system/docker.service.d

# Create the drop-in unit file
tee /etc/systemd/system/docker.service.d/kolla.conf <<-'EOF'

# Run these commands to reload the daemon
systemctl daemon-reload
systemctl restart docker

For Ubuntu 14.04 which uses upstart instead of systemd, run the following:

mount --make-shared /run


If centos/fedora/oraclelinux container images are built on an Ubuntu host, the backend storage driver must not be AUFS (see the known issues in image-building).


On ubuntu 16.04, please uninstall lxd and lxc packages. (issue with cgroup mounts, mounts exponentially increasing when restarting container).

On the target hosts you also need an updated version of the Docker python libraries:


The old docker-python is obsoleted by python-docker-py.

yum install -y python-docker-py

Or using pip to install a latest version:

pip install -U docker-py

OpenStack, RabbitMQ, and Ceph require all hosts to have matching times to ensure proper message delivery. In the case of Ceph, it will complain if the hosts differ by more than 0.05 seconds. Some OpenStack services have timers as low as 2 seconds by default. For these reasons it is highly recommended to setup an NTP service of some kind. While ntpd will achieve more accurate time for the deployment if the NTP servers are running in the local deployment environment, chrony is more accurate when syncing the time across a WAN connection. When running Ceph it is recommended to setup ntpd to sync time locally due to the tight time constraints.

To install, start, and enable ntp on CentOS execute the following:

# CentOS 7
yum -y install ntp
systemctl enable ntpd.service
systemctl start ntpd.service

To install and start on Debian based systems execute the following:

apt-get install ntp

Libvirt is started by default on many operating systems. Please disable libvirt on any machines that will be deployment targets. Only one copy of libvirt may be running at a time.

# CentOS 7
systemctl stop libvirtd.service
systemctl disable libvirtd.service

# Ubuntu
service libvirt-bin stop
update-rc.d libvirt-bin disable

On Ubuntu, apparmor will sometimes prevent libvirt from working.

/usr/sbin/libvirtd: error while loading shared libraries: libvirt-admin.so.0: cannot open shared object file: Permission denied

If you are seeing the libvirt container fail with the error above, disable the libvirt profile.

sudo apparmor_parser -R /etc/apparmor.d/usr.sbin.libvirtd

Kolla deploys OpenStack using Ansible. Install Ansible from distribution packaging if the distro packaging has recommended version available.

Some implemented distro versions of Ansible are too old to use distro packaging. Currently, CentOS and RHEL package Ansible >2.0 which is suitable for use with Kolla. Note that you will need to enable access to the EPEL repository to install via yum -- to do so, take a look at Fedora's EPEL docs and FAQ.

On CentOS or RHEL systems, this can be done using:

yum -y install ansible

Many DEB based systems do not meet Kolla's Ansible version requirements. It is recommended to use pip to install Ansible >2.0. Finally Ansible >2.0 may be installed using:

pip install -U ansible

If DEB based systems include a version of Ansible that meets Kolla's version requirements it can be installed by:

apt-get install ansible

Install Kolla

To clone the Kolla repo:

git clone https://git.openstack.org/openstack/kolla

To install Kolla tools and Python dependencies use:

pip install kolla/

Kolla holds configurations files in etc/kolla. Copy the configuration files to /etc:

cd kolla
cp -r etc/kolla /etc/

Install Python Clients

On the system where the OpenStack CLI/Python code is run, the Kolla community recommends installing the OpenStack python clients if they are not installed. This could be a completely different machine then the deployment host or deployment targets. The following requirements are needed to build the client code:

# Ubuntu
apt-get install -y python-dev libffi-dev libssl-dev gcc

# CentOS 7
yum -y install python-devel libffi-devel openssl-devel gcc

To install the clients use:

yum install -y python-openstackclient python-neutronclient

Or using pip to install:

pip install -U python-openstackclient python-neutronclient

Local Registry

A local registry is not required for an all-in-one installation. Check out the multinode for more information on using a local registry. Otherwise, the Docker Hub Image Registry contains all images from each of Kolla's major releases. The latest release tag is 2.0.0 for Mitaka.

Additional Environments

Two virtualized development environment options are available for Kolla. These options permit the development of Kolla without disrupting the host operating system.

If developing Kolla on an OpenStack cloud environment that supports Heat, follow the heat-dev-env.

If developing Kolla on a system that provides VirtualBox or Libvirt in addition to Vagrant, use the Vagrant virtual environment documented in vagrant-dev-env.

Currently the Heat development environment is entirely non-functional. The Kolla core reviewers have debated removing it from the repository but have resisted to provide an opportunity for contributors to make Heat usable for Kolla development. The Kolla core reviewers believe Heat would offer a great way to develop Kolla in addition to Vagrant, bare metal, or a manually setup virtual machine.

For more information refer to _bug 1562334.

Building Container Images

The Kolla community builds and pushes tested images for each tagged release of Kolla, but if running from master, it is recommended to build images locally.

Checkout the image-building for more advanced build configuration.

Before running the below instructions, ensure the docker daemon is running or the build process will fail. To build images using default parameters run:


By default kolla-build will build all containers using CentOS as the base image and binary installation as base installation method. To change this behavior, please use the following parameters with kolla-build:

--base [ubuntu|centos|fedora|oraclelinux]
--type [binary|source]

If pushing to a local registry (recommended) use the flags:

kolla-build --registry registry_ip_address:registry_ip_port --push

Note --base and --type can be added to the above kolla-build command if different distributions or types are desired.

It is also possible to build individual containers. As an example, if the glance containers failed to build, all glance related containers can be rebuilt as follows:

kolla-build glance

In order to see all available parameters, run:

kolla-build -h

For more information about building Kolla container images, check the detailed instruction in image-building.

Deploying Kolla

The Kolla community provides two example methods of Kolla deploy: all-in-one and multinode. The all-in-one deploy is similar to devstack deploy which installs all OpenStack services on a single host. In the multinode deploy, OpenStack services can be run on specific hosts. This documentation only describes deploying all-in-one method as most simple one. To setup multinode see the multinode.

Each method is represented as an Ansible inventory file. More information on the Ansible inventory file can be found in the Ansible inventory introduction.

All variables for the environment can be specified in the files: /etc/kolla/globals.yml and /etc/kolla/passwords.yml.

Generate passwords for /etc/kolla/passwords.yml using the provided kolla-genpwd tool. The tool will populate all empty fields in the /etc/kolla/passwords.yml file using randomly generated values to secure the deployment. Optionally, the passwords may be populate in the file by hand.


Start by editing /etc/kolla/globals.yml. Check and edit, if needed, these parameters: kolla_base_distro, kolla_install_type. These parameters should match what you used in the kolla-build command line. The default for kolla_base_distro is centos and for kolla_install_type is binary. If you want to use ubuntu with source type, then you should make sure globals.yml has the following entries:

kolla_base_distro: "ubuntu"
kolla_install_type: "source"

Please specify an unused IP address in the network to act as a VIP for kolla_internal_vip_address. The VIP will be used with keepalived and added to the api_interface as specified in the globals.yml :

kolla_internal_vip_address: ""

The network_interface variable is the interface to which Kolla binds API services. For example, when starting up Mariadb it will bind to the IP on the interface list in the network_interface variable. :

network_interface: "eth0"

The neutron_external_interface variable is the interface that will be used for the external bridge in Neutron. Without this bridge the deployment instance traffic will be unable to access the rest of the Internet. In the case of a single interface on a machine, a veth pair may be used where one end of the veth pair is listed here and the other end is in a bridge on the system. :

neutron_external_interface: "eth1"

If using a local docker registry, set the docker_registry information where the local registry is operating on IP address and the port 4000.

docker_registry: ""

For all-in-one deploys, the following commands can be run. These will setup all of the containers on the localhost. These commands will be wrapped in the kolla-script in the future.


even for all-in-one installs it is possible to use the docker registry for deployment, although not strictly required.

First, check that the deployment targets are in a state where Kolla may deploy to them:

kolla-ansible prechecks

Verify that all required images with appropriate tags are available:

kolla-ansible pull

Run the deployment:

kolla-ansible deploy

If APIError/HTTPError is received from the neutron-dhcp-agent container, remove the container and recreate it:

docker rm -v -f neutron_dhcp_agent
kolla-ansible deploy

In order to see all available parameters, run:

kolla-ansible -h


In case of deploying using the _nested environment (eg. Using Virtualbox VM's, KVM VM's), if your compute node supports hardware acceleration for virtual machines.

For this, run the follow command in compute node:

$ egrep -c '(vmx|svm)' /proc/cpuinfo

If this command returns a value of zero, your compute node does not support hardware acceleration and you must configure libvirt to use QEMU instead of KVM.

For this, change the virt_type option in the [libvirt] section of nova-compute.conf file inside the /etc/kolla/config/ directory.


A bare metal system with Ceph takes 18 minutes to deploy. A virtual machine deployment takes 25 minutes. These are estimates; different hardware may be faster or slower but should be near these results.

After successful deployment of OpenStack, the Horizon dashboard will be available by entering IP address or hostname from kolla_external_fqdn, or kolla_internal_fqdn. If these variables were not set during deploy they default to kolla_internal_vip_address.

Useful tools

After successful deployment of OpenStack, run the following command can create an openrc file /etc/kolla/admin-openrc.sh on the deploy node. Or view tools/openrc-example for an example of an openrc that may be used with the environment.

kolla-ansible post-deploy

After the openrc file is created, use the following command to initialize an environment with a glance image and neutron networks:

source /etc/kolla/admin-openrc.sh


Nearly always when Kolla fails, it is caused by a CTRL-C during the deployment process or a problem in the globals.yml configuration.

To correct the problem where Operators have a misconfigured environment, the Kolla developers have added a precheck feature which ensures the deployment targets are in a state where Kolla may deploy to them. To run the prechecks, execute:

kolla-ansible prechecks

If a failure during deployment occurs it nearly always occurs during evaluation of the software. Once the Operator learns the few configuration options required, it is highly unlikely they will experience a failure in deployment.

Deployment may be run as many times as desired, but if a failure in a bootstrap task occurs, a further deploy action will not correct the problem. In this scenario, Kolla's behavior is undefined.

The fastest way during evaluation to recover from a deployment failure is to remove the failed deployment:

On each node where OpenStack is deployed run:


The Operator will have to copy via scp or some other means the cleanup scripts to the various nodes where the failed containers are located.

Any time the tags of a release change, it is possible that the container implementation from older versions won't match the Ansible playbooks in a new version. If running multinode from a registry, each node's Docker image cache must be refreshed with the latest images before a new deployment can occur. To refresh the docker cache from the local Docker registry:

kolla-ansible pull

Debugging Kolla

The container's status can be determined on the deployment targets by executing:

docker ps -a

If any of the containers exited, this indicates a bug in the container. Please seek help by filing a launchpad bug or contacting the developers via IRC.

The logs can be examined by executing:

docker exec -it heka bash

The logs from all services in all containers may be read from /var/log/kolla/SERVICE_NAME

If the stdout logs are needed, please run:

docker logs <container-name>

Note that most of the containers don't log to stdout so the above command will provide no information.

To learn more about Docker command line operation please refer to Docker documentation.

When enable_central_logging is enabled, to view the logs in a web browser using Kibana, go to:

or http://<kolla_external_vip_address>:<kibana_server_port>

and authenticate using <kibana_user> and <kibana_password>.

The values <kolla_internal_vip_address>, <kolla_external_vip_address> <kibana_server_port> and <kibana_user> can be found in <kolla_install_path>/kolla/ansible/group_vars/all.yml or if the default values are overridden, in /etc/kolla/globals.yml. The value of <kibana_password> can be found in /etc/kolla/passwords.yml.


When you log in to Kibana web interface for the first time, you are prompted to create an index. Please create an index using the name log-*. This step is necessary until the default Kibana dashboard is implemented in Kolla.