[Puppet Users] How to update a managed directory?

You are focusing far too narrowly on a particular implementation. It may
not be possible to achieve specifically what you're asking for, but there
are several ways to reach the configuration objective that is your ultimate
goal.

One thing you might be able to do is subclass the class that manages your
directory, and override the File resource appropriately.

Another thing you might be able to do is split out management of the files
that need to vary (with the help of File's 'ignore' parameter, for
instance) into separate classes, and apply the correct one of those classes
based on whatever logic applies.

Alternatively, perhaps you could template those files such that all nodes
could use the same templates, and you wouldn't need conditional or
special-case behavior in your manifests.

There are other alternatives, too, but those are the most promising I have
come up with.


John

John, thanks for the response. I added my comments inline.
Looking at this now actually. the source and sourceselect attributes of
File are probably what I want. However the source attribute is order
dependent when determining which file gets applied. Would you know if the
+> operator pre or post pends to an existing array? Is there a way to
enforce the array ordering when appending? Haven't found documentation on
this so will be doing some trial and error.

I do not know for sure whether +> prepends or appends, but my guess would
be that it appends. Whichever way it does it, you can rely on Puppet to be
consistent. As far as I know, there is no way to switch the order. Of
course, the subclass can always replace the source array instead of
appending. With judicious use of class variables, you could even do that
without duplicating the source list anywhere.

Alternatively, if your approach is going to be to fiddle with the 'source'
parameter then you might consider how to avoid subclassing at all. One way
would be to load your source(s) via hiera. Another might leverage File's
behavior of skipping sources that don't exist: for example, if one or more
of your sources are declared in terms of the $hostname fact, then you can
create hostname-based overrides entirely within your source data for only
the nodes you want.


John

Agreed. This is what I am focusing on. But have hit a snag I don't know how
to resolve. My end goal is to have a patching mechanism for a JEE
container that does not have an installer. Just a directory of files that
gets copied to the node. I have verified that passing an array with source
directories of the patches before the JEE container files works perfectly.
I can add patches and roll them back easily. This all works great when I
hard code the patch directories in an array before the jee container source
directory. However I don't know when a patch will be released and I need
the flexibility to apply the patch to only certain nodes. This means the
array set for the source has to be created dynamically. I am using a
template to create the array from the patches passed in as parameters.

class jee ($patches = []){
...
file {$install_dir:
ensure => directory,
source => split(template('jee/source.erb'),','),
sourceselect => 'all'
recurse => true,
...
}
}

Everything in mind tells me this should work. However when I try to run
the above code I get the following error.
"Error: /Stage[main]/jee/File[/usr/local/jee]: Failed to generate
additional resources using 'eval_generate: wrong header line format"
I have seen the "wrong header line format" before and it is usually an
incorrect quotation someplace. However in this case I am relying on the
split function to transform the string returned from the template into an
array. I am not sure where to begin debugging this. Any ideas?


I'm not too keen on the $hostname solution. I feel that is what nodes are
for so don't want to obfuscate how files get applied by creating another
means to do the same thing. Plus this would become a management nightmare
given the use case I described earlier.

Apparently a template adds return on the end of the return string.
Guessing this is to support the template concatenation for file content.
Switched to inline_template and all works well. Lesson learned.
Appreciate your comments on this John. It's nice to have second pair of
eyes.