How can I use the results from jinja2-templated file into a variable without some hack to write a file out and read it in again?
I’m using the ec2 module to create new EC2 instances. For this module, you can specify a multi-line string for the user_data field. I’d like to use a template to create the string to use for this field. How can I do this?
The template module doesn’t seem to fit because it wants to write to an absolute destination at the target host, but I don’t want to write a file, I just need the templated output to use in a variable to pass along to the ec2 module.
I can write the data with {{ jinja2 variables }} directly in a playbook, but I want to re-use the same template for multiple playbooks that set up different instances with different values.
Tried the lookup(‘file’, ‘my-user-data.yml’) but it does not appear to be parsed by jinja2.
include_vars helps by making a chunk of work re-usable; thanks! However in my use case, I would like to include text in one file into the variable file to finally be included in a task. I’ve updated the git repository to reflect this setup with use of include_vars. The problem is that the output from dumping one multiline variable into another file in YAML format doesn’t indent each line to comply with the YAML multiline format.
The actual use case here is the use of the ec2 module to create new instances, and I am setting the user_data field to some cloud-init YAML for use in the initial provisioning of CoreOS (I would prefer to use Ansible, but some of it can only be done as part of this cloudinit process). I want to write files during this process that are included directly in the cloud-config YAML, and so I would prefer to have those files as separate files in my Ansible setup, such as “bashrc.yml” for example, and include these inside a “cloudinit.yml” file that constructs the final text to use for the user_data variable.
I found an awkward way to deal with this, though it does at least work. One can use the Jinja2 indent filter as I have in the code:
multilinevar_from_included_varfile: |
var_tests:
name: include some script in yaml format
script: |
{{ script_content**|indent(4)** }}"
If anyone knows of a cleaner way to do this I’d love to hear it.
Output from updated test without using the Jinja2 ‘indent’ filter:
name: include some script in yaml format
script: #!/bin/bash
echo this is my script!
echo notice that the lines in the final output are not indented currently to be parsed correctly in the task file
I’m sorry, I found your use case too confusing for me to understand - no offense implied - with what you were doing with lookup plugins.
I can’t help you explore it further.
This may just be the case of a mailing list making it hard to communicate, but things like "The problem is that the output from dumping one multiline variable into another file in YAML format doesn’t indent each line to comply with the YAML multiline format. " don’t make sense to me.
Embedding YAML in YAML isn’t really a thing, normally you would just keep a datastructure inside the YAML.
In ansible, you can also use the template module to dump YAML on demand like so:
“{{ foo | to_yaml }}”
Which is really great if you have a datastructure.
Michael, no need to apologize, you’ve already helped a lot, thank you!
One thing that would really help here and is general enough I’m sure it would find use elsewhere as well is to have the template module capable of outputting to a variable rather than a file. That certainly would help me in my case, and I’ve seen cases elsewhere while trying to improve my code for this where it would also have been helpful.
I don't know if you know this, but variables can themselves be
templated. You can do this in a set_fact or even in a variable file
(group_vars, host_vars, var_files, etc).
For example, say I have file, .bashrc that I want to write to my target host during the cloud-init process. In some cases, like with CoreOS which sets files like this one read-only once cloud-init is complete, there are things you can do during cloud-init that you cannot do afterwards (otherwise I’d just use template module as usual). To make this happen with Ansible, I need to include the content of the file within the cloud-init YAML string which is passed in to the user_data field of the ec2 ansible module. I would like to store a template like bashrc.j2 so I could modify it before it gets used within the user_data string as part of that cloud_init text, rather than awkwardly as part of a variable file in YAML format with a bunch of other files.
"One thing that would really help here and is general enough I’m sure it would find use elsewhere as well is to have the template module capable of outputting to a variable rather than a file. That certainly would help me in my case, and I’ve seen cases elsewhere while trying to improve my code for this where it would also have been helpful. "
Yep, as indicated above, you can put templates directly in variable files, so this might not be needed.
Excerpts from AJ Daws's message of 2014-08-08 17:30:33 -0400:
Michael, no need to apologize, you've already helped a lot, thank you!
One thing that would really help here and is general enough I'm sure it
would find use elsewhere as well is to have the template module capable of
outputting to a variable rather than a file. That certainly would help me
in my case, and I've seen cases elsewhere while trying to improve my code
for this where it would also have been helpful.
There is also a template lookup plugin, no? This should evaluate the
contents of a template:
This actually sounds like precisely what I was hoping for. Michael, can you let me know why you don’t recommend it? I assume by “evaluate” the template, you mean it will parse it with Jinja2, which seems to make sense…? Thanks for the help!