devstack/doc/source/faq.rst

178 lines
8.3 KiB
ReStructuredText

===
FAQ
===
- `General Questions <#general>`__
- `Operation and Configuration <#ops_conf>`__
- `Miscellaneous <#misc>`__
General Questions
=================
Q: Can I use DevStack for production?
A: No. We mean it. Really. DevStack makes some implementation
choices that are not appropriate for production deployments. We
warned you!
Q: Then why selinux in enforcing mode?
A: That is the default on current Fedora and RHEL releases. DevStack
has (rightly so) a bad reputation for its security practices; it has
always been meant as a development tool first and system integration
later. This is changing as the security issues around OpenStack's
use of root (for example) have been tightened and developers need to
be better equipped to work in these environments. ``stack.sh``'s use
of root is primarily to support the activities that would be handled
by packaging in "real" deployments. To remove additional protections
that will be desired/required in production would be a step
backward.
Q: But selinux is disabled in RHEL!
A: Today it is, yes. That is a specific exception that certain
DevStack contributors fought strongly against. The primary reason it
was allowed was to support using RHEL6 as the Python 2.6 test
platform and that took priority time-wise. This will not be the case
with RHEL 7.
Q: Why a shell script, why not chef/puppet/...
A: The script is meant to be read by humans (as well as ran by
computers); it is the primary documentation after all. Using a
recipe system requires everyone to agree and understand chef or
puppet.
Q: Why not use Crowbar?
A: DevStack is optimized for documentation & developers. As some of
us use `Crowbar <https://github.com/dellcloudedge/crowbar>`__ for
production deployments, we hope developers documenting how they
setup systems for new features supports projects like Crowbar.
Q: I'd like to help!
A: That isn't a question, but please do! The source for DevStack is
at
`git.openstack.org <https://git.openstack.org/cgit/openstack-dev/devstack>`__
and bug reports go to
`LaunchPad <http://bugs.launchpad.net/devstack/>`__. Contributions
follow the usual process as described in the `developer
guide <http://docs.openstack.org/infra/manual/developers.html>`__. This Sphinx
documentation is housed in the doc directory.
Q: Why not use packages?
A: Unlike packages, DevStack leaves your cloud ready to develop -
checkouts of the code and services running in screen. However, many
people are doing the hard work of packaging and recipes for
production deployments. We hope this script serves as a way to
communicate configuration changes between developers and packagers.
Q: Why isn't $MY\_FAVORITE\_DISTRO supported?
A: DevStack is meant for developers and those who want to see how
OpenStack really works. DevStack is known to run on the
distro/release combinations listed in ``README.md``. DevStack is
only supported on releases other than those documented in
``README.md`` on a best-effort basis.
Q: What about Fedora/RHEL/CentOS?
A: Fedora and CentOS/RHEL are supported via rpm dependency files and
specific checks in ``stack.sh``. Support will follow the pattern set
with the Ubuntu testing, i.e. only a single release of the distro
will receive regular testing, others will be handled on a
best-effort basis.
Q: Are there any differences between Ubuntu and Fedora support?
A: Neutron is not fully supported prior to Fedora 18 due lack of
OpenVSwitch packages.
Q: Why can't I use another shell?
A: DevStack now uses some specific bash-ism that require Bash 4, such
as associative arrays. Simple compatibility patches have been accepted
in the past when they are not complex, at this point no additional
compatibility patches will be considered except for shells matching
the array functionality as it is very ingrained in the repo and project
management.
Q: But, but, can't I test on OS/X?
A: Yes, even you, core developer who complained about this, needs to
install bash 4 via homebrew to keep running tests on OS/X. Get a Real
Operating System. (For most of you who don't know, I am referring to
myself.)
Operation and Configuration
===========================
Q: Can DevStack handle a multi-node installation?
A: Indirectly, yes. You run DevStack on each node with the
appropriate configuration in ``local.conf``. The primary
considerations are turning off the services not required on the
secondary nodes, making sure the passwords match and setting the
various API URLs to the right place.
Q: How can I document the environment that DevStack is using?
A: DevStack includes a script (``tools/info.sh``) that gathers the
versions of the relevant installed apt packages, pip packages and
git repos. This is a good way to verify what Python modules are
installed.
Q: How do I turn off a service that is enabled by default?
A: Services can be turned off by adding ``disable_service xxx`` to
``local.conf`` (using ``n-vol`` in this example):
::
disable_service n-vol
Q: Is enabling a service that defaults to off done with the reverse of the above?
A: Of course!
::
enable_service qpid
Q: How do I run a specific OpenStack milestone?
A: OpenStack milestones have tags set in the git repo. Set the appropriate tag in the ``*_BRANCH`` variables in ``local.conf``. Swift is on its own release schedule so pick a tag in the Swift repo that is just before the milestone release. For example:
::
[[local|localrc]]
GLANCE_BRANCH=stable/juno
HORIZON_BRANCH=stable/juno
KEYSTONE_BRANCH=stable/juno
NOVA_BRANCH=stable/juno
GLANCE_BRANCH=stable/juno
NEUTRON_BRANCH=stable/juno
SWIFT_BRANCH=2.2.1
Q: Why not use [STRIKEOUT:``tools/pip-requires``]\ ``requirements.txt`` to grab project dependencies?
[STRIKEOUT:The majority of deployments will use packages to install
OpenStack that will have distro-based packages as dependencies.
DevStack installs as many of these Python packages as possible to
mimic the expected production environment.] Certain Linux
distributions have a 'lack of workaround' in their Python
configurations that installs vendor packaged Python modules and
pip-installed modules to the SAME DIRECTORY TREE. This is causing
heartache and moving us in the direction of installing more modules
from PyPI than vendor packages. However, that is only being done as
necessary as the packaging needs to catch up to the development
cycle anyway so this is kept to a minimum.
Q: What can I do about RabbitMQ not wanting to start on my fresh new VM?
A: This is often caused by ``erlang`` not being happy with the
hostname resolving to a reachable IP address. Make sure your
hostname resolves to a working IP address; setting it to 127.0.0.1
in ``/etc/hosts`` is often good enough for a single-node
installation. And in an extreme case, use ``clean.sh`` to eradicate
it and try again.
Q: How can I set up Heat in stand-alone configuration?
A: Configure ``local.conf`` thusly:
::
[[local|localrc]]
HEAT_STANDALONE=True
ENABLED_SERVICES=rabbit,mysql,heat,h-api,h-api-cfn,h-api-cw,h-eng
KEYSTONE_SERVICE_HOST=<keystone-host>
KEYSTONE_AUTH_HOST=<keystone-host>
Q: Why are my configuration changes ignored?
A: You may have run into the package prerequisite installation
timeout. ``tools/install_prereqs.sh`` has a timer that skips the
package installation checks if it was run within the last
``PREREQ_RERUN_HOURS`` hours (default is 2). To override this, set
``FORCE_PREREQ=1`` and the package checks will never be skipped.
Miscellaneous
=============
Q: ``tools/fixup_stuff.sh`` is broken and shouldn't 'fix' just one version of packages.
A: [Another not-a-question] No it isn't. Stuff in there is to
correct problems in an environment that need to be fixed elsewhere
or may/will be fixed in a future release. In the case of
``httplib2`` and ``prettytable`` specific problems with specific
versions are being worked around. If later releases have those
problems than we'll add them to the script. Knowing about the broken
future releases is valuable rather than polling to see if it has
been fixed.