DevStack configuration is modified via the file
local.conf. It is
a modified INI format file that introduces a meta-section header to
carry additional information regarding the configuration files to be
A sample is provided in
The new header is similar to a normal INI section header but with double
[[ ... ]]) and two internal fields separated by a pipe
|). Note that there are no spaces between the double brackets and the
internal fields. Likewise, there are no spaces between the pipe and the
'[[' <phase> '|' <config-file-name> ']]'
<phase> is one of a set of phase names defined by
<config-file-name> is the configuration filename. The filename
is eval’ed in the
stack.sh context so all environment variables are
available and may be used. Using the project config file variables in
the header is strongly suggested (see the
NOVA_CONF example below).
If the path of the config file does not exist it is skipped.
The defined phases are:
local - extracts
post-config - runs after the layer 2 services are configured and before they are started
extra - runs after services are started and before any files in
post-extra - runs after files in
test-config - runs after tempest (and plugins) are configured
The file is processed strictly in sequence; meta-sections may be specified more than once but if any settings are duplicated the last to appear in the file will be used.
[[post-config|$NOVA_CONF]] [DEFAULT] use_syslog = True [osapi_v3] enabled = False
A specific meta-section
local|localrc is used to provide a default
localrc file (actually
.localrc.auto). This allows all custom
settings for DevStack to be contained in a single file. If
exists it will be used instead to preserve backward-compatibility.
[[local|localrc]] IPV4_ADDRS_SAFE_TO_USE=10.254.1.0/24 ADMIN_PASSWORD=speciale LOGFILE=$DEST/logs/stack.sh.log
Q_PLUGIN_CONF_FILE is unique in that it is assumed to
NOT start with a
/ (slash) character. A slash will need to be
Also note that the
localrc section is sourced as a shell script
fragment and MUST conform to the shell requirements, specifically no
openrc configures login credentials suitable for use with the
OpenStack command-line tools.
stackrc at the
beginning (which in turn sources the
localrc section of
local.conf) in order to pick up
to use in the endpoints. The values shown below are the default values.
- OS_PROJECT_NAME (OS_TENANT_NAME)
Keystone has standardized the term project as the entity that owns resources. In some places references still exist to the previous term tenant for this use. Also, project_name is preferred to project_id. OS_TENANT_NAME remains supported for compatibility with older tools.
In addition to the owning entity (project), OpenStack calls the entity performing the action user.
Keystone’s default authentication requires a password be provided. The usual cautions about putting passwords in environment variables apply, for most DevStack uses this may be an acceptable tradeoff.
- HOST_IP, SERVICE_HOST
Set API endpoint host using
SERVICE_HOSTmay also be used to specify the endpoint, which is convenient for some
HOST_IPis set in the
Authenticating against an OpenStack cloud using Keystone returns a Token and Service Catalog. The catalog contains the endpoints for all services the user/tenant has access to - including Nova, Glance, Keystone and Swift.
- KEYSTONECLIENT_DEBUG, NOVACLIENT_DEBUG
Set command-line client log level to
DEBUG. These are commented out by default.
# export KEYSTONECLIENT_DEBUG=1 # export NOVACLIENT_DEBUG=1
stack.sh is happy to run without a
localrc section in
local.conf, devlife is better when there are a few minimal variables
set. This is an example of a minimal configuration that touches the
values that most often need to be set.
pre-set the passwords to prevent interactive prompts
move network ranges away from the local network (
FLOATING_RANGE, commented out below)
set the host IP if detection is unreliable (
HOST_IP, commented out below)
[[local|localrc]] ADMIN_PASSWORD=secret DATABASE_PASSWORD=$ADMIN_PASSWORD RABBIT_PASSWORD=$ADMIN_PASSWORD SERVICE_PASSWORD=$ADMIN_PASSWORD #IPV4_ADDRS_SAFE_TO_USE=172.31.1.0/24 #FLOATING_RANGE=192.168.20.0/25 #HOST_IP=10.3.4.5
*_PASSWORD variables are not set here you will be prompted to
enter values for them by
Only use alphanumeric characters in your passwords, as some services fail to work when using special characters.
The network ranges must not overlap with any networks in use on the host. Overlap is not uncommon as RFC-1918 ‘private’ ranges are commonly used for both the local networking and Nova’s fixed and floating ranges.
HOST_IP is normally detected on the first run of
often is indeterminate on later runs due to the IP being moved from an
Ethernet interface to a bridge on the host. Setting it here also makes it
openrc to set
HOST_IP is not set
HOST_IPV6 is normally detected on the first run of
will not be set if there is no IPv6 address on the default Ethernet interface.
Setting it here also makes it available for
openrc to set
HOST_IPV6 is not set by default.
For architecture specific configurations which differ from the x86 default here, see arch-configuration.
Historically DevStack obtained all local configuration and
customizations from a
localrc file. In Oct 2013 the
local.conf configuration method was introduced (in review 46768) to simplify this
The Git repositories used to check out the source for each service are
controlled by a pair of variables set for each service.
points to the repository and
*_BRANCH selects which branch to
check out. These may be overridden in
local.conf to pull source
from a different repo for testing, such as a Gerrit branch
GIT_BASE points to the primary repository server.
To pull a branch directly from Gerrit, get the repo and branch from the Gerrit review page:
git fetch https://review.opendev.org/openstack/nova \ refs/changes/50/5050/1 && git checkout FETCH_HEAD
The repo is the stanza following
fetch and the branch is the
stanza following that:
The DevStack install directory is set by the
DEST variable. By
default it is
By setting it early in the
localrc section you can reference it in
later variables. It can be useful to set it even though it is not
changed from the default value.
stack.sh output is only written to the console where it
runs. It can be sent to a file in addition to the console by setting
LOGFILE to the fully-qualified name of the destination log file. A
timestamp will be appended to the given filename for each run of
Old log files are cleaned automatically if
LOGDAYS is set to the
number of days of old log files to keep.
Some coloring is used during the DevStack runs to make it easier to see what is going on. This can be disabled with:
When using the logfile, by default logs are sent to the console and
the file. You can set
false if you only wish the
logs to be sent to the file (this may avoid having double-logging in
some cases where you are capturing the script output and the log
true you can additionally set
VERBOSE_NO_TIMESTAMP to avoid timestamps being added to each
output line sent to the console. This can be useful in some
situations where the console output is being captured by a runner or
framework (e.g. Ansible) that adds its own timestamps. Note that the
log lines sent to the
LOGFILE will still be prefixed with a
Logging the Service Output¶
By default, services run under
systemd and are natively logging to
the systemd journal.
To query the logs use the
journalctl command, such as:
sudo journalctl --unit devstack@*
More examples can be found in Querying Logs.
Example Logging Configuration¶
For example, non-interactive installs probably wish to save output to a file, keep service logs and disable color in the stored files.
[[local|localrc]] DEST=/opt/stack/ LOGFILE=$LOGDIR/stack.sh.log LOG_COLOR=False
Multiple database backends are available. The available databases are defined
in the lib/databases directory.
mysql is the default database, choose a different one by putting the
following in the
disable_service mysql enable_service postgresql
mysql is the default database.
Support for a RabbitMQ RPC backend is included. Additional RPC
backends may be available via external plugins. Enabling or disabling
RabbitMQ is handled via the usual service functions and
Example disabling RabbitMQ in
The Apache web server can be enabled for wsgi services that support
being deployed under HTTPD + mod_wsgi. By default, services that
recommend running under HTTPD + mod_wsgi are deployed under Apache. To
use an alternative deployment strategy (e.g. eventlet) for services
that support an alternative to HTTPD + mod_wsgi set
False in your
Each service that can be run under HTTPD + mod_wsgi also has an
override toggle available that can be set in your
Keystone is run under Apache with
mod_wsgi by default.
Libraries from Git¶
By default devstack installs OpenStack server components from git,
however it installs client libraries from released versions on pypi.
This is appropriate if you are working on server development, but if
you want to see how an unreleased version of the client affects the
system you can have devstack install it from upstream, or from local
git trees by specifying it in
LIBS_FROM_GIT. Multiple libraries
can be specified as a comma separated list.
Setting the variable to
ALL will activate the download for all
Enable the use of Python virtual environments by setting
True. This will enable the creation of venvs for each project
that is defined in the
Each entry in the
PROJECT_VENV array contains the directory name
of a venv to be used for the project. The array index is the project
name. Multiple projects can use the same venv if desired.
ADDITIONAL_VENV_PACKAGES is a comma-separated list of additional
packages to be installed into each venv. Often projects will not have
certain packages listed in its
requirements.txt file because they
are ‘optional’ requirements, i.e. only needed for certain
configurations. By default, the enabled databases will have their
Python bindings added when they are enabled.
A clean install every time¶
stack.sh only clones the project repos if they do not
stack.sh will freshen each repo on each run if
RECLONE is set to
yes. This avoids having to manually remove
repos in order to get the current branch from
Upgrade packages installed by pip¶
stack.sh only installs Python packages if no version is
currently installed or the current version does not match a specified
PIP_UPGRADE is set to
True then existing
required Python packages will be upgraded to the most recent version
that matches requirements.
Images provided in URLS via the comma-separated
variable will be downloaded and uploaded to glance by DevStack.
Default guest-images are predefined for each type of hypervisor and
their testing-requirements in
DOWNLOAD_DEFAULT_IMAGES=False will prevent DevStack downloading
these default images; in that case, you will want to populate
IMAGE_URLS with sufficient images to satisfy testing-requirements.
DOWNLOAD_DEFAULT_IMAGES=False IMAGE_URLS="http://foo.bar.com/image.qcow," IMAGE_URLS+="http://foo.bar.com/image2.qcow"
DEFAULT_INSTANCE_TYPE can be used to configure the default instance
type. When this parameter is not specified, Devstack creates additional
micro & nano flavors for really small instances to run Tempest tests.
For guests with larger memory requirements,
should be specified in the configuration file so Tempest selects the
default flavors instead.
KVM on Power with QEMU 2.4 requires 512 MB to load the firmware - QEMU 2.4 - PowerPC so users running instances on ppc64/ppc64le can choose one of the default created flavors as follows:
IP_VERSION can be used to configure Neutron to create either an
IPv4, IPv6, or dual-stack self-service project data-network by with
The following optional variables can be used to alter the default IPv6 behavior:
IPV6_RA_MODE=slaac IPV6_ADDRESS_MODE=slaac IPV6_ADDRS_SAFE_TO_USE=fd$IPV6_GLOBAL_ID::/56 IPV6_PRIVATE_NETWORK_GATEWAY=fd$IPV6_GLOBAL_ID::1
can be configured with any valid IPv6 prefix. The default values make
use of an auto-generated
IPV6_GLOBAL_ID to comply with RFC4193.
Service IP Version¶
DevStack can enable service operation over either IPv4 or IPv6 by
SERVICE_IP_VERSION to either
When set to
4 devstack services will open listen sockets on
0.0.0.0 and service endpoints will be registered using
as the address.
When set to
6 devstack services will open listen sockets on
and service endpoints will be registered using
HOST_IPV6 as the
The default value for this setting is
4. Dual-mode support, for
4+6 is not currently supported.
optionally be used to alter the default IPv6 address:
Tunnel IP Version¶
DevStack can enable tunnel operation over either IPv4 or IPv6 by
TUNNEL_IP_VERSION to either
When set to
4 Neutron will use an IPv4 address for tunnel endpoints,
When set to
6 Neutron will use an IPv6 address for tunnel endpoints,
The default value for this setting is
4. Dual-mode support, for
4+6 is not supported, as this value must match the address
family of the local tunnel endpoint IP(v6) address.
The value of
TUNNEL_IP_VERSION has a direct relationship to the
TUNNEL_ENDPOINT_IP, which will default to
when set to
HOST_IPV6 when set to
See the multi-node lab guide
See the neutron configuration guide for details on configuration of Neutron
Swift is disabled by default. When enabled, it is configured with only one replica to avoid being IO/memory intensive on a small VM.
If you would like to enable Swift you can add this to your
enable_service s-proxy s-object s-container s-account
If you want a minimal Swift install with only Swift and Keystone you
can have this instead in your
disable_all_services enable_service key mysql s-proxy s-object s-container s-account
If you only want to do some testing of a real normal swift cluster
with multiple replicas you can do so by customizing the variable
SWIFT_REPLICAS in your
localrc section (usually to 3).
You can manually override the ring building to use specific storage
nodes, for example when you want to test a multinode environment. In
this case you have to set a space-separated list of IPs in
SWIFT_STORAGE_IPS in your
localrc section that should be used
as Swift storage nodes.
Please note that this does not create a multinode setup, it is only
used when adding nodes to the Swift rings.
SWIFT_STORAGE_IPS="192.168.1.10 192.168.1.11 192.168.1.12"
If you are enabling
ENABLED_SERVICES DevStack will
install the s3api middleware emulation. Swift will be configured to
act as a S3 endpoint for Keystone so effectively replacing the
Only Swift proxy server is launched in the systemd system all other
services are started in background and managed by
If tempest has been successfully configured, a basic set of smoke tests can be run as follows:
$ cd /opt/stack/tempest $ tox -e smoke
By default tempest is downloaded and the config file is generated, but the
tempest package is not installed in the system’s global site-packages (the
package install includes installing dependences). So tempest won’t run
outside of tox. If you would like to install it add the following to your
The logical volume group used to hold the Cinder-managed volumes is
VOLUME_GROUP_NAME, the logical volume name prefix is set with
VOLUME_NAME_PREFIX and the size of the volume backing file is set
VOLUME_GROUP_NAME="stack-volumes" VOLUME_NAME_PREFIX="volume-" VOLUME_BACKING_FILE_SIZE=24G
When running highly concurrent tests, the default per-project quotas
for volumes, backups, or snapshots may be too small. These can be
adjusted by setting
CINDER_QUOTA_SNAPSHOTS to the desired value. (The default for
each is 10.)
DevStack’s Cinder LVM configuration module currently supports both iSCSI and
NVMe connections, and we can choose which one to use with options
Defaults use iSCSI with the LIO target manager:
CINDER_TARGET_HELPER="lioadm" CINDER_TARGET_PROTOCOL="iscsi" CINDER_TARGET_PREFIX="iqn.2010-10.org.openstack:" CINDER_TARGET_PORT=3260
Additionally there are 3 supported transport protocols for NVMe,
nvmet_fc, and when the
is selected the protocol, prefix, and port defaults will change to more
sensible defaults for NVMe:
CINDER_TARGET_HELPER="nvmet" CINDER_TARGET_PROTOCOL="nvmet_rdma" CINDER_TARGET_PREFIX="nvme-subsystem-1" CINDER_TARGET_PORT=4420
When selecting the RDMA transport protocol DevStack will create on Cinder nodes
a Software RoCE device on top of the
HOST_IP_IFACE and if it is not defined
then on top of the interface with IP address
This Soft-RoCE device will always be created on the Nova compute side since we cannot tell beforehand whether there will be an RDMA connection or not.
We want to setup two devstack (RegionOne and RegionTwo) with shared keystone (same users and services) and horizon. Keystone and Horizon will be located in RegionOne. Full spec is available at: https://wiki.openstack.org/wiki/Heat/Blueprints/Multi_Region_Support_for_Heat.
disable_service horizon KEYSTONE_SERVICE_HOST=<KEYSTONE_IP_ADDRESS_FROM_REGION_ONE> REGION_NAME=RegionTwo KEYSTONE_REGION_NAME=RegionOne
In the devstack for RegionOne, we set REGION_NAME as RegionOne, so region of the services started in this devstack are registered as RegionOne. In devstack for RegionTwo, similarly, we set REGION_NAME as RegionTwo since we want services started in this devstack to be registered in RegionTwo. But Keystone service is started and registered in RegionOne, not RegionTwo, so we use KEYSTONE_REGION_NAME to specify the region of Keystone service. KEYSTONE_REGION_NAME has a default value the same as REGION_NAME thus we omit it in the configuration of RegionOne.
The default image size quota of 1GiB may be too small if larger images are to be used. Change the default at setup time with:
or at runtime via:
openstack --os-cloud devstack-system-admin registered limit set \ --service glance --default-limit 5000 --region RegionOne image_size_total
The upstream CI runs exclusively on nodes with x86 architectures, but OpenStack supports even more architectures. Some of them need to configure Devstack in a certain way.
KVM on s390x (IBM z Systems)¶
KVM on s390x (IBM z Systems) is supported since the Kilo release. For
an all-in-one setup, these minimal settings in the
[[local|localrc]] ADMIN_PASSWORD=secret DATABASE_PASSWORD=$ADMIN_PASSWORD RABBIT_PASSWORD=$ADMIN_PASSWORD SERVICE_PASSWORD=$ADMIN_PASSWORD DOWNLOAD_DEFAULT_IMAGES=False IMAGE_URLS="https://cloud-images.ubuntu.com/xenial/current/xenial-server-cloudimg-s390x-disk1.img" # Provide a custom etcd3 binary download URL and ints sha256. # The binary must be located under '/<etcd version>/etcd-<etcd-version>-linux-s390x.tar.gz' # on this URL. # Build instructions for etcd3: https://github.com/linux-on-ibm-z/docs/wiki/Building-etcd ETCD_DOWNLOAD_URL=<your-etcd-download-url> ETCD_SHA256=<your-etcd3-sha256> enable_service n-sproxy disable_service n-novnc [[post-config|$NOVA_CONF]] [serial_console] base_url=ws://$HOST_IP:6083/ # optional
The default image of Devstack is x86 only, so we deactivate the download with
DOWNLOAD_DEFAULT_IMAGES. The referenced guest image in the code above (
IMAGE_URLS) serves as an example. The list of possible s390x guest images is not limited to that.
This platform doesn’t support a graphical console like VNC or SPICE. The technical reason is the missing framebuffer on the platform. This means we rely on the substitute feature serial console which needs the proxy service
n-sproxy. We also disable VNC’s proxy
n-novncfor that reason . The configuration in the
post-configsection is only needed if you want to use the serial console outside of the all-in-one setup.
A link to an etcd3 binary and its sha256 needs to be provided as the binary for s390x is not hosted on github like it is for other architectures. For more details see https://bugs.launchpad.net/devstack/+bug/1693192. Etcd3 can easily be built along https://github.com/linux-on-ibm-z/docs/wiki/Building-etcd.
To run Tempest against this Devstack all-in-one, you’ll need to use a guest image which is smaller than 1GB when uncompressed. The example image from above is bigger than that!