zuul/doc/source/admin/installation.rst
James E. Blair a1d8fc7a51 Reorganize admin manual
This makes a number of changes to the installation/configuration
documentation in the admin manual.

Remove quick-start guide.  The process of quick-starting is
covered by the installation and setup tutorial, which is now the
first of the installation sections.  The reference material from
quick-start is now in the tutorial.

Rename the tutorial quick-start.  It's nice to have something
named quick-start, and the tutorial fits the bill.

Rename the installation section "Installation Reference".  This
now has more detailed information about installation and
deployment choices, but has very little procedural documentation.

Make zuul-from-scratch more internally consistent in style (use
code-block:: shell and heredocs wherever possible).

Change-Id: I7e4714ce5e775dc9ac0988c3470eef1f74fb36d6
2018-10-10 09:54:03 -07:00

335 lines
11 KiB
ReStructuredText

Installation Reference
======================
Install Zuul
------------
To install a Zuul release from PyPI, run::
pip install zuul
Or from a git checkout, run::
pip install .
That will also install Zuul's python dependencies. To minimize
interaction with other python packages installed on a system, you may
wish to install Zuul within a Python virtualenv.
Zuul has several system-level dependencies as well. You can find a
list of operating system packages in ``bindep.txt`` in Zuul's source
directory.
Zuul Components
---------------
Zuul provides the following components:
- **zuul-scheduler**: The main Zuul process. Handles receiving
events, executing jobs, collecting results and posting reports.
Coordinates the work of the other components. It also provides
a gearman daemon which the other components use for
coordination.
- **zuul-merger**: Scale-out component that performs git merge
operations. Zuul performs a large number of git operations in
the course of its work. Adding merger processes can help speed
Zuul's processing. This component is optional (zero or more of
these can be run).
- **zuul-executor**: Scale-out component for executing jobs. At
least one of these is required. Depending on system
configuration, you can expect a single executor to handle up to
about 100 simultaneous jobs. Can handle the functions of a
merger if dedicated mergers are not provided. One or more of
these must be run.
- **zuul-web**: A web server that receives "webhook" events from
external providers, supplies a web dashboard, and provides
websocket access to live streaming of logs.
- **zuul-fingergw**: A gateway which provides finger protocol
access to live streaming of logs.
For more detailed information about these, see :ref:`components`.
External Dependencies
---------------------
Zuul interacts with several other systems described below.
Gearman
~~~~~~~
Gearman is a job distribution system that Zuul uses to communicate
with its distributed components. The Zuul scheduler distributes work
to Zuul mergers and executors using Gearman. You may supply your own
gearman server, but the Zuul scheduler includes a built-in server
which is recommended. Ensure that all Zuul hosts can communicate with
the gearman server.
Zuul distributes secrets to executors via gearman, so be sure to
secure it with TLS and certificate authentication. Obtain (or
generate) a certificate for both the server and the clients (they may
use the same certificate or have individual certificates). They must
be signed by a CA, but it can be your own CA.
Nodepool
~~~~~~~~
In order to run all but the simplest jobs, Zuul uses a companion
program, Nodepool, to supply the nodes (whether dynamic cloud
instances or static hardware) used by jobs. Before starting Zuul,
ensure you have Nodepool installed and any images you require built.
Zuul only makes one requirement of these nodes: that it be able to log
in given a username and ssh private key.
.. TODO: SpamapS any zookeeper config recommendations?
Nodepool uses Zookeeper to communicate internally among its
components, and also to communicate with Zuul. You can run a simple
single-node Zookeeper instance, or a multi-node cluster. Ensure that
the host running the Zuul scheduler has access to the cluster.
Ansible
~~~~~~~
Zuul uses Ansible to run jobs. Each version of Zuul is designed to
work with a specific, contemporary version of Ansible. Zuul specifies
that version of Ansible in its python package metadata, and normally
the correct version will be installed automatically with Zuul.
Because of the close integration of Zuul and Ansible, attempting to
use other versions of Ansible with Zuul is not recommended.
Zuul Setup
----------
At minimum you need to provide ``zuul.conf`` and ``main.yaml`` placed
in ``/etc/zuul/``. The following example uses the builtin gearman
service in Zuul, and a connection to Gerrit.
**zuul.conf**::
[scheduler]
tenant_config=/etc/zuul/main.yaml
[gearman_server]
start=true
[gearman]
server=127.0.0.1
[connection my_gerrit]
driver=gerrit
server=git.example.com
port=29418
baseurl=https://git.example.com/gerrit/
user=zuul
sshkey=/home/zuul/.ssh/id_rsa
See :ref:`components` and :ref:`connections` for more details.
The following tells Zuul to read its configuration from and operate on
the *example-project* project:
**main.yaml**::
- tenant:
name: example-tenant
source:
my_gerrit:
untrusted-projects:
- example-project
Starting Zuul
-------------
You can run any zuul process with the **-d** option to make it not
daemonize. It's a good idea at first to confirm there's no issues with
your configuration.
To start, simply run::
zuul-scheduler
Once run you should have two zuul-scheduler processes (if using the
built-in gearman server, or one process otherwise).
Before Zuul can run any jobs, it needs to load its configuration, most
of which is in the git repositories that Zuul operates on. Start an
executor to allow zuul to do that::
zuul-executor
Zuul should now be able to read its configuration from the configured
repo and process any jobs defined therein.
.. _web-deployment-options:
Web Deployment Options
----------------------
The ``zuul-web`` service provides an web dashboard, a REST API and a websocket
log streaming service as a single holistic web application. For production use
it is recommended to run it behind a reverse proxy, such as Apache or Nginx.
More advanced users may desire to do one or more exciting things such as:
White Label
Serve the dashboard of an individual tenant at the root of its own domain.
https://zuul.openstack.org is an example of a Zuul dashboard that has been
white labeled for the ``openstack`` tenant of its Zuul.
Static Offload
Shift the duties of serving static files, such as HTML, Javascript, CSS or
images to the Reverse Proxy server.
Static External
Serve the static files from a completely separate location that does not
support programmatic rewrite rules such as a Swift Object Store.
Sub-URL
Serve a Zuul dashboard from a location below the root URL as part of
presenting integration with other application.
https://softwarefactory-project.io/zuul/ is an example of a Zuul dashboard
that is being served from a Sub-URL.
None of those make any sense for simple non-production oriented deployments, so
all discussion will assume that the ``zuul-web`` service is exposed via a
Reverse Proxy. Where rewrite rule examples are given, they will be given
with Apache syntax, but any other Reverse Proxy should work just fine.
Basic Reverse Proxy
~~~~~~~~~~~~~~~~~~~
Using Apache as the Reverse Proxy requires the ``mod_proxy``,
``mod_proxy_http`` and ``mod_proxy_wstunnel`` modules to be installed and
enabled. Static Offload and White Label additionally require ``mod_rewrite``.
All of the cases require a rewrite rule for the websocket streaming, so the
simplest reverse-proxy case is::
RewriteEngine on
RewriteRule ^/api/tenant/(.*)/console-stream ws://localhost:9000/api/tenant/$1/console-stream [P]
RewriteRule ^/(.*)$ http://localhost:9000/$1 [P]
Static Offload
~~~~~~~~~~~~~~
To have the Reverse Proxy serve the static html/javascript assets instead of
proxying them to the REST layer, register the location where you unpacked
the web application as the document root and add rewrite rules::
<Directory /usr/share/zuul>
Require all granted
</Directory>
Alias / /usr/share/zuul/
<Location />
RewriteEngine on
RewriteBase /
# Rewrite api to the zuul-web endpoint
RewriteRule api/tenant/(.*)/console-stream ws://localhost:9000/api/tenant/$1/console-stream [P,L]
RewriteRule api/(.*)$ http://localhost:9000/api/$1 [P,L]
# Backward compatible rewrite
RewriteRule t/(.*)/(.*).html(.*) /t/$1/$2$3 [R=301,L,NE]
# Don't rewrite files or directories
RewriteCond %{REQUEST_FILENAME} !-f
RewriteCond %{REQUEST_FILENAME} !-d
RewriteRule . /index.html [L]
</Location>
Sub directory serving
~~~~~~~~~~~~~~~~~~~~~
The web application needs to be rebuild to update the internal location of
the static files. Set the homepage setting in the package.json to an
absolute path or url. For example, to deploy the web interface through a
'/zuul/' sub directory:
.. note::
The web dashboard source code and package.json are located in the ``web``
directory. All the yarn commands need to be executed from the ``web``
directory.
.. code-block:: bash
sed -e 's#"homepage": "/"#"homepage": "/zuul/"#' -i package.json
yarn build
Then assuming the web application is unpacked in /usr/share/zuul,
add the following rewrite rules::
<Directory /usr/share/zuul>
Require all granted
</Directory>
Alias /zuul /usr/share/zuul/
<Location /zuul>
RewriteEngine on
RewriteBase /zuul
# Rewrite api to the zuul-web endpoint
RewriteRule api/tenant/(.*)/console-stream ws://localhost:9000/api/tenant/$1/console-stream [P,L]
RewriteRule api/(.*)$ http://localhost:9000/api/$1 [P,L]
# Backward compatible rewrite
RewriteRule t/(.*)/(.*).html(.*) /t/$1/$2$3 [R=301,L,NE]
# Don't rewrite files or directories
RewriteCond %{REQUEST_FILENAME} !-f
RewriteCond %{REQUEST_FILENAME} !-d
RewriteRule . /zuul/index.html [L]
</Location>
White Labeled Tenant
~~~~~~~~~~~~~~~~~~~~
Running a white-labeled tenant is similar to the offload case, but adds a
rule to ensure connection webhooks don't try to get put into the tenant scope.
.. note::
It's possible to do white-labelling without static offload, but it is more
complex with no benefit.
Assuming the zuul tenant name is "example", the rewrite rules are::
<Directory /usr/share/zuul>
Require all granted
</Directory>
Alias / /usr/share/zuul/
<Location />
RewriteEngine on
RewriteBase /
# Rewrite api to the zuul-web endpoint
RewriteRule api/connection/(.*)$ http://localhost:9000/api/connection/$1 [P,L]
RewriteRule api/console-stream ws://localhost:9000/api/tenant/example/console-stream [P,L]
RewriteRule api/(.*)$ http://localhost:9000/api/tenant/example/$1 [P,L]
# Backward compatible rewrite
RewriteRule t/(.*)/(.*).html(.*) /t/$1/$2$3 [R=301,L,NE]
# Don't rewrite files or directories
RewriteCond %{REQUEST_FILENAME} !-f
RewriteCond %{REQUEST_FILENAME} !-d
RewriteRule . /index.html [L]
</Location>
Static External
~~~~~~~~~~~~~~~
.. note::
Hosting zuul dashboard on an external static location that does not support
dynamic url rewrite rules only works for white-labeled deployments.
In order to serve the zuul dashboard code from an external static location,
``REACT_APP_ZUUl_API`` must be set at javascript build time:
.. code-block:: bash
REACT_APP_ZUUL_API='http://zuul-web.example.com' yarn build