If you’re learning to use ansible, as I am, you might find this helpful.
I just discovered that you can do stuff like this:
pretend my_rpms.txt is a text file full of the names of rpm packages that I want to install, one name per line
- name: install some rpms from yum
yum: name=$item
with_file: my_rpms.txt
There’s also a with_pipe and with_fileglob and all kinds of stuff in lib/ansible/runner/lookup_plugins/ in the ansible source tree.
I didn’t find this documented anywhere, so I thought I’d mention it in the list to help anyone else who is in the same situation as me.
Thanks!
-Dylan
Ouch, they don't seem to be documented on
http://ansible.cc/docs/playbooks2.html !
They should be.
I'll file a bug.
--Michael
In this case does it run the rpm command once for each line in the file?
I am running into this situation with apt-get, running apt-get against a large list of packages is much faster than running it once against each package, also, you can solve dependency issues this way (if there is a choice of dependency for a given package).
Is there a way to make this option aware of comments, like only using lines that do not start with a hash sign?
The yum and apt modules have extra brains that lets them use $item
efficiently. So, it's not running yum a separate time for each
module.
My example was wrong: with_lines does what I described.
If you wanted to skip lines with a hash, you could probably use with_pipe like
with_pipe: grep -v '^#" my_rpms.txt
but I'm not sure if you'd need to json-ifty the output.
It's not presently possible, though as I can't think of any reason why
this would ever be a bad thing the way it was originally intended (to
keep a package list as a separate file, for instance), I'll gladly
take patches to enable this.
FWIW, it seems it should also skip blank lines.
filter= option that defaults to blank lines and those that start with # or ;?
I'd make it default to ignoring blanks and "#".
Only worry about a filter arg if we need it later.
Michael DeHaan wrote:
I'd make it default to ignoring blanks and "#".
Only worry about a filter arg if we need it later.
with_file only gets complete files. For processing line-based data, there
is with_lines which takes a command to execute, where you can use grep as
you see fit.
Daniel
What I'd really like is some way to pipe with_X plugins through each
other. with_file = reads the contents of a file but doesn't output
json. with_lines = reads contents of a file and outputs json, one
item per line. with_pipe = puts output into one var. Should we write
with_pipe_lines to make with_pipe output json? Maybe something like
with:
- file: $filename
- grep: -v "^#"
- grep: -v "^[[:space:]]\*$"
- lines_to_json:
I've used imaginary with_grep and with_lines_to_json
We know this.
However, executing grep is gross here, adds additional syntax, and we
will be smarter than this.
If you want this, I'd suggest you write a plugin to do it, but it's
going to have to look a little different.
I am not interested in seeing this complication in core, and want to
see the number of concepts not continue to grow unbounded 
Roger. Sorry. Got carried away.
I keep wanting to do everything inside the playbook.