We've got some old out of date docs in some places. This isn't even a full reworking, but at least tries to remove some of the more egregiously wrong things. Change-Id: I9033acb9572e1ce1b3e4426564b92706a4385dcb
4.8 KiB
- title
-
Bridge
Bridge
Bridge is a bastion host that is the starting point for ops operations in OpenDev. It is the server from which Ansible is run, and contains a centralized database that contains secure information such as passwords. The bridge server contains all of the ansible playbooks as well as the scripts to create new servers.
Many of the systems in OpenDev are still configured using puppet, although the trend is away from Puppet to Ansible. For the hosts still using puppet, the process is still driven by Ansible.
At a Glance
- Projects
- Bugs
- Resources
Ansible Hosts
In OpenDev, all host configuration is done via ansible playbooks.
Puppet Hosts
For hosts still using puppet, ansible drives the running of
puppet apply
on hosts in the inventory in the
puppet
group. That process first copies appropriate
hiera
data files to each host.
Hiera uses a systemwide configuration file in
/etc/puppet/hiera.yaml
.
The hiera configuration is placed by ansible into common.yaml in
/etc/puppet/hieradata/production
on each puppet host. The
values are simple key-value pairs in yaml format. The keys needed are
the keys referenced in site.pp
, their values are typically
obvious (strings, lists of strings). /etc/puppet/hieradata/
and below should be owned by puppet:puppet
and have mode
0711
.
Below the hieradata
directory, there should be a
common.yaml
file where settings that should be available to
all servers in the infrastructure go, and then two directories full of
files. The first is fqdn
which should contain a yaml file
for every server in the infrastructure named
${fqdn_of_server}.yaml
. That file has secrets that are only
for that server. Additionally, some servers can have a
$group
defined in manifests/site.pp
. There can
be a correspondingly named yaml file in the group
directory
that contains secrets to be made available to each server in the
group.
All of the actual yaml files should have mode 0600 and be owned by root.
Adding a node
Adding a new node should be done using the
/opt/system-config/launch/launch-node.py
script (see
:git_file:`launch/README.rst`
for full details). If the host is put into the puppet group in the
Ansible inventory, puppet will be run on on the host.
Running Ansible on Nodes
Each service that has been migrated fully to Ansible has its own playbook in :git_file:`playbooks named service_{ service_name }.yaml`.
Because the playbooks are normally run by zuul, to run them manually,
first touch the file /home/zuul/DISABLE-ANSIBLE
. Then make
sure no jobs are currently executing ansible. Ensure that
/home/zuul/src/opendev.org/opendev/system-config
and
/home/zuul/src/opendev.org/openstack/project-config
are in
the appropriate states, then run:
cd /home/zuul/src/opendev.org/opendev/system-config
ansible-playbook --limit="$HOST:localhost" playbooks/service-$SERVICE.yaml
as root, where $HOST is the host you want to run puppet on. The :localhost is important as some of the plays depend on performing a task on the localhost before continuing to the host in question, and without it in the limit section, the tasks for the host will have undefined values.
When done, don't forget to remove
/home/zuul/DISABLE-ANSBILE
Running Puppet on Nodes
In OpenDev, puppet is run by ansible from a Zuul job running on bridge which in turn runs a single run of puppet apply on each host using puppet.
The entry point for this process is :git_file:`playbooks/remote_puppet_else.yaml`
If an admin needs to run puppet by hand, it's a simple matter of
following the instructions in running-ansible-on-nodes
but using
playbooks/remote_puppet_else.yaml
as the playbook.
Alternately, if local iteration is desired, it's possible to log in to the server in question and running puppet apply /opt/system-config/manifests/site.pp.
There is also a script, tools/kick.sh that takes the host as an argument and runs the above command.
Testing new puppet code can be done via puppet apply --noop or by constructing a VM with a puppet install in it and just running puppet apply on the code in question. This should actually make it fairly easy to test how production works in a more self-contained manner.
Disabling Ansible on Nodes
In the case of needing to disable the running of ansible on a node,
it's a simple matter of adding an entry to the ansible inventory
"disabled" group. See the disable-enable-ansible
section for more details.