Guillem, Sorry for the delayed response. Anyway, John is correct about what you're trying to do with Hiera. I can say as far as the ENC Gary has written follows all the rules of what an ENC 'should do' as per the documentation.
http://docs.puppetlabs.com/guides/external_nodes.html Keep in mind also, when the ENC is run, the only information it has access to is the Facts from the node. The manifests are not compiled until after the ENC returns the classes it has determined should be declared. This ENC does it in one call to hiera for classes. When this Hiera ENC processes, it never adds the discovered parameters and other variables to the current running context. While that could achieve what you want, it could also complicate things in unexpected ways. So as I said above, when the ENC runs, you only have just the facts from the host as your current context. John's suggestion of a new usage model that aligns better with Hiera's design is probably the best answer. That being a case, I would suggest custom facts to help you navigate your hiera tree in a more controlled/granular manner. -- Tom Linkin Professional Services Engineer http://puppetlabs.com/ twitter: @trlinkin On Monday, October 1, 2012 at 10:47 AM, jcbollinger wrote: > > > On Wednesday, September 26, 2012 10:52:19 AM UTC-5, Guillem Liarte wrote: > > This is the situation I have: > > > > All my hosts are the same OS. > > All my host are in the same puppet environment, so I cannot use > > %{environment} > > > > I have a module that sets all the basic functionality for the OS, > > resolution, authentication, security, packages, etc > > I have a module for each application hosted. > > > > At the moment all the 'data' is in Puppet, mostly in parametrised classes > > in site.pp. > > > > What I want to get is a hiera set-up that allows me to use this structure: > > > I suspect that one of the reasons you are having trouble is that you are > trying to employ a usage paradigm that is inconsistent with hiera's design > (more below). > > > > > :hierarchy: > > - global # source of application names (classes? modules?) and > > environments list > > - basic # data for the basic class > > There's nothing wrong with those levels. > > > - prod/%{application}/%{hostname} # hostname files for specific > > data > > - prod/%{application}/%{env} # environmental data for > > each application (module) > > - prod/%{application}/default # default data for an > > application > > > But there is a problem with those. It may be possible to make it work, but > it's shaky to use variable hierarchy levels for data selection. That's what > keys are for. With that said, recent Puppet releases provide automatic > $calling_module and $calling_class variables, one of which you could probably > use in place of $application. As I understand it, that's intended to provide > (better) support for module-specific data, which might be a good way to cast > that part of your problem. > > > > - nonprod/%{sysclass}/%{hostname} > > - nonprod/%{sysclass}/%{env} > > - nonprod/%{sysclass}/default > > > You additionally have a fundamental problem with %{env}. Hiera will attempt > to resolve that as a Puppet variable, to which the presence of a matching key > somewhere in the Hiera hierarchy is irrelevant. Hiera needs to know the > value to resolve the hierarchy (as you have defined it), and it would need, > in principle, to resolve the hierarchy before it could look up the value in > your data store. > > What actually happens, I'm sure, is that hiera uses the value of $::env that > it looks up in Puppet at function entry. You might be able to work around > that by setting that variable in Puppet before looking up other data, such as > by putting > > $env = hiera('env') > > at top scope near the beginning of your site.pp. > > > > > > So in short, I would like hiera to be a source of facts, where I can get > > information that feeds Puppet in order to classify the nodes and to feed > > the parametrised classes. > > > As an aside, throwing parametrized classes into this mix has only downside as > far as I am concerned, except inasmuch as you may want to use parametrized > classes that are (unwisely) provided by modules written by others. Since you > want to rely on hiera (which is good), it is superior to write your classes > like this wherever you are in control of module interfaces: > > class mymodule::class1 { > $param1 = hiera('mymodule::class1::param1') > $param2 = hiera('mymodule::class1::param2') > # or with simpler keys enabled by use of > # %{calling_module} and/or %{calling_class} > } > > There are several advantages, among them that you can encode interclass > parse-order dependencies via the built-in 'include' function, and that you > can use hiera's 'hiera_include()' function to assign such classes to nodes. > > > > > I recently found this blog entry: > > > > http://garyhetzel.com/2012/04/12/hiera_as_a_puppet_enc > > > > > Gary appears to have done some cool work there, but as you have discovered, > it's not going to overcome the inherent problem with self-referrential data. > It might be possible to work around this by augmenting Gary's hiera > additions/modifications with a separate pre-lookup of needed extra variables, > but you're then talking about a distinctly non-trivial effort and a > substantial branch away from stock hiera. > > > > I would like to make emphasis in this: Gary's hiera as an ENC works, but > > for a more simple scenario than the one I am proposing, if I only wanted to > > classify Classes and Hosts, it does work fine. Where I have not been able > > to succeed is in adding an 'env' layer after the application (classes, > > organised in modules). > > > You are classifying based only on hostname, because that's the only data you > want to consider that actually originates from the node being classified. > Everything else is logic and structure of the ENC you are trying to build. > > The problem is that you are trying to implement a data structure that Hiera > does not natively support. You can make it work, but you will need either > significant changes in hiera, or a different usage mode. I would suggest the > latter. > > Were I you, I would consider writing a separate, hiera-based ENC instead of > trying to build all the ENC features you want directly into hiera itself. > Among other things, you would not then need to worry about maintaining a > private hiera fork. It looks like the basic idea would need to involve the > ENC making multiple (at least two) lookups in hiera, using possibly different > subsets of the overall data hierarchy for your site, to retrieve the > variables to and classes (optionally with parameters) that should be > assigned. Anywhere the assigned classes perform their own hiera lookups, > they would use the full hierarchy. > > Although Gary's work doesn't do everything you want, it would probably serve > as a good guide to much of what is needed. > > > John > > -- > You received this message because you are subscribed to the Google Groups > "Puppet Users" group. > To view this discussion on the web visit > https://groups.google.com/d/msg/puppet-users/-/IT2oIrddP-QJ. > To post to this group, send email to puppet-users@googlegroups.com > (mailto:puppet-users@googlegroups.com). > To unsubscribe from this group, send email to > puppet-users+unsubscr...@googlegroups.com > (mailto:puppet-users+unsubscr...@googlegroups.com). > For more options, visit this group at > http://groups.google.com/group/puppet-users?hl=en. -- You received this message because you are subscribed to the Google Groups "Puppet Users" group. To post to this group, send email to puppet-users@googlegroups.com. To unsubscribe from this group, send email to puppet-users+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.
