Dynamic inventory to run ansible plays in jenkins slaves
We can't always work with those fancy cloud environments in our CI/CD with jenkins and be able to create and destroy full environments at will. Sometimes we have to deal with baremetal machines or just persistent nodes that we need to configure and reconfigure every now and then. Hence, sometimes we need to run maintenance scripts on those nodes, and that's the reason why I developed this plugin. Everybody has thought "we can just create a job and run it in every node", I wanted something different.
This inventory plugin needs ansible >= 2.5
At least, the user needs:
- Global Read: in order to read 'computer/api/json'
- Slave Configure: in order to read 'computer/{SLAVENAME}/config.xml'
Depending on the authentication method used, this plugin will have a different performance. You can:
-
Authenticate using user/password and not use the jsessionid:
This is, BY FAR, the slowest method. Unless you have no other option because of a bug or whatever, do always use the jsessionid (
jenkins_jsessionid=True) if you use user/password to authenticate. -
Authenticate using user/password and use the jsessionid:
This is the way I get the best performance. You will authenticate once, and use the jsessionid from this moment on and never authenticate again.
-
Authenticate using user/apitoken:
If you cannot use the user/password for any reason, you cannot use the jsessionid (since we get the jsessionid in the login screen, we cannot use the apitoken for that). Anyway, this method will have a performance between the other two.
Using the jsessionid you should have no speed problems even with HUGE jenkins servers. Anyway, if you find it too slow, remember that you can always use the cache.
Copy the jenkins.py file in your ansible's inventory plugins directory.
Enable the jenkins plugin in your ansible's configuration file:
[inventory]
# enable inventory plugins, default: 'host_list', 'script', 'yaml', 'ini'
#enable_plugins = host_list, virtualbox, yaml, constructed
enable_plugins = jenkins
Instead of changing your ansible configuration and copying the plugin in your ansible's inventory plugin directory, you can run your command like that:
ANSIBLE_INVENTORY_PLUGINS=. ANSIBLE_INVENTORY_ENABLED=jenkins ansible -i example_inventory.jenkins.yml -m debug -a "var=hostvars[inventory_hostname]" all
This is the file that we will use when invoking the ansible command with the -i option.
ansible-playbook -i my_jenkins_inventory_file_described_here.jenkins.yml my_fancy_playbook.yml
It's name has to end with .jenkins.yml or .jenkins.yaml.
Parameters regarding our jenkins instance.
The name of this plugin: "jenkins".
Mandatory.
URL to the jenkis you want to control with this inventory
Example: http://127.0.0.1:8080
Mandatory.
The username that will be used to query the nodes connected to our jenkins.
You can omit this user if your computers can be seen by anonymous.
The password for jenkins_user.
You can use the apitoken too instead of the user password. The performance should be better with the apitoken.
You can omit this value if you want the password to be requested via prompt when executing the command.
Remember that you can use vault here. Do not use plain text.
If "True", it will login and save the jsessionid cookie.
This option only can be used with the password, do NOT use it with the apitoken.
This option should dramatically improve the performance when using the password.
Maximum time in seconds that it will wait for a single request. Because of ansible's code, the timeout cannot
be set to infinite. It must be greater than 0. None value will result in a default 10 seconds timeout.
Parameters regarding our cache configuration. We can use any of the cache plugins to store our inventory, so that we don't have to communicate with the jenkins instance every time we want to execute a playbook.
Enables (True) / Disables (False) the use of cache when executing this plugin.
Default: False
Once you have enabled the use of the cache, you can use whatever cache plugin you want to use. Each plugin will use their own parameters, you can find detailed documentation about cache plugins and their parameters here: cache doc
Create vars from jinja2 expressions
add hosts to a group based on Jinja2 conditionals
add hosts to a group based on the values of a variable
If true makes invalid entries a fatal error, otherwise skip and continue
Since it is possible to use facts in the expressions they might not always be available
and we ignore those errors by default.
You can find detailed documentation about Constructed params here: constructed doc
This is how a jenkins inventory file using jsoncache plugin would look like:
plugin: jenkins
# Jenkins plugin configuration
# Jenkins server url configuration
jenkins_host: http://127.0.0.1:8080/
# Jenkins user that must have permissions to see the nodes connected to jenkins
# Omit user and password if computers can be seen by anonymous in your jenkins server.
jenkins_user: user
# The password/apitoken for this user. Remember that ansible-vault is supported here.
# It's not a good idea to put plain text here.
# If you omit this field, the password will be gotten via prompt
jenkins_pass: secretpassword
# Do NOT use this option with an apitoken, or the authentication will fail.
# Use it if you are using the user's password, since it will improve the performance.
jenkins_jsessionid: True
# Cache configuration.
# Sample for json file chache plugin configuration, just configure the chache plugin
# that suites better for you
cache_connection: /path/to/the/cache/directory
cache_plugin: jsonfile
cache: True
# Constructed configuration
# Create new hostvars using existing hostvars
compose:
ansible_connection: ('indows' in launcher_plugin)|ternary('winrm', 'ssh')
# You can define your port here for windows too, if you want to.
# Create groups depending on jinja2 filters
groups:
temporary_offline: (temporary_offline)
# Create groups depending on variable values
keyed_groups:
- prefix: oss
key: launcher_plugin.split('@')[0]
# Requests timeout. Maximum time in seconds that we will wait for a request
# None = 10 seconds by default because of ansible code.
# Cannot configure infinte timeout. Must be > 0
timeout: 60
You can find a sample here
The plugin will include several hostvars.
The plugin used by jenkins to launch scripts in this slave. I.e: "windows-slaves@1.3.1",
"command-launcher@1.2" or "ssh-slaves@1.26", for instance.
When we can get this information from the launcher plugin, we will set this hostvar.
It will not be set if we cant find its value inspecting the launcher
(in command-launcher for instance we can't get its value)
If we can get the port from the launcher, we will set it. If not, we won't.
(Basically, we will only set it when using ssh-slaves launcher).
If you need to set this variable in a windows slave, you will have to do it via 'compose'.
It will store "True" or "False" depending on if the node is set temporary offline in jenkins.
It will store "True" or "False" depending on if the node is offline in jenkins. A node can be offline
and not temporary offline, it usually means that there is a problem with the connection to jenkins.
If a node is set temporary offline, this value will be "True" too.
It will store "True" if all node executors are idle, or "False" if any of them is working.
The number of executors in this node. It doesn't matter if the node is online or offline,
this value does not change because of that.
The plugin will read the node properties defined in jenkins and will set them as hostvars in this node,
so that you can access them from your ansible whenever you need them.
You need to be aware that there are variables that will be ignored. Those variables are:
- 'launcher_plugin'
- 'inventory_hostname'
- 'ansible_host'
- 'ansible_port'
- 'temporary_offline'
If you define any node property with those names, it will be ignored, and the actual
value will NOT be overwritten.
By default this plugin will create as many groups as tags are defined in jenkins.
There are 2 mandatory groups in ansible, "all" and "ungrouped". Hence, please, we should NOT use tags with those names in jenkins if we want to use this plugin. It is a limitation that we just can't avoid, we should NEVER use those names as tags in our jenkins (if we are using this plugin), I repeat.
Apart from the tags, we can create any extra groups using the 'constructed' interface. For instance, as we can see in our jenkins-inventory-file example, we can create those groups:
# Group by launcher (windows/linux/command....)
keyed_groups:
- prefix: oss
key: launcher_plugin.split('@')[0]
# Group those nodes that are temporary offline
groups:
temporary_offline: (temporary_offline)
Some basic examples about how to use the plugin:
Shows all groups in the inventory:
ansible -i example_inventory.jenkins.yml -m debug -a "var=hostvars[inventory_hostname]['groups']" localhost
Show all hostvars in every node:
ansible -i example_inventory.jenkins.yml -m debug -a "var=hostvars[inventory_hostname]" all
Ping all nodes:
ansible -i example_inventory.jenkins.yml -m ping all
If we want to ping only those nodes labeled with "mylabel"
ansible -i example_inventory.jenkins.yml -m ping mylabel
Add the -k parameter to the examples if you didn't share the keys.
I repeat what I have said about groups names before:
There are 2 mandatory groups in ansible, "all" and "ungrouped". Hence, please, we should NOT use tags with those names in jenkins if we want to use this plugin. It is a limitation that we just can't avoid, we should NEVER use those names as tags in our jenkins (if we are using this plugin), I repeat.
There is another limitation:
When constructing variables or groups in jenkins inventory file, we only can use the variables defined by this plugin. We cannot use hostvars like 'ansible_version', for instance, to construct new ones defined in the jenkins inventory file. Hence, we should only use the variables defined in the "hostvars" section of this document (including those ones defined via node property in jenkins) to construct new ones (or new groups) via 'constructed'.
Unittests only work with python 2.
It doesn't seem to be a very interesting plugin, since people usually don't need to run ansible in their jenkins slaves because they use jenkins to do so.
Although it is an original way of collecting information or setting up the slaves, I will not invest more time in this plugin. Nevertheless I will try to fix the bugs if I find any of them. Please, if you find any bug, don't hesitate to report it.