in-module data with hiera

[R.I.Pienaar]

hello,

Till now hiera-puppet was the only way I know that allowed hiera data to
be loaded from inside a module.

The problem with this was that it was still subject to the site specific
hierarchy which means a module author had a pretty hard time to store
his data in a proper way in his module thus perpetuating the use of the
params classes pattern.

Now that Puppet 3 is out and it's gem extensible I can finally try some
ideas I had but put off implementing because it was too hard to install
and distribute these extensions.

I propose extending the module layout with a data/ directory that can go
into each module and in this data directory would live a hiera config
file (optionally) and module specific data:

your_module
├── data
│ ├── hiera.json
│ └── osfamily
│ ├── Debian.json
│ └── RedHat.json
└── manifests
└── init.pp

Here the data/hiera.json is optional and specifies a hierarchy that the
module author chooses and is unique to the specific backend.

The default contents would be this is the file is absent:

{"hierarchy": ["osfamily/%{osfamily}", "common"]}

But a module author can pick anything there, should even be able to rely
on facts that is shipped with the module in its lib dir since that'll
get pluginsynced out before compile time:

Now given the data files for Redhat:

{"apache::package" : "httpd"}

...and Debian:

{"apache::package" : "apache2"}

And your main hiera site config being something like:

:backends: - json
- module_json

You should be able to just write module code like this:

class apache($package="apache") {
package{$package: ensure => present}
}

If no data is specified in the site hiera backends then this will use
the in-module hierarchy and data and just do the right thing on RedHat
and Debian systems but as always the site can still override the data
using hard coding, site specific data, ENCs etc.

So the important thing here is the module author has control over the
hierarchy that gets used when the data in his module gets loaded. The
site can has its own hierarchy policy but this backend will only use
the hierarchy that is recorded in the module by its author.

If you want to play with this idea on your Puppet 3 install just do 'gem
install hiera-module-json'

So I am looking for feedback from the community on this pattern, will it
solve the problem of author-supplied module data better than we have
today? I've heard this problem brought up quite a lot so keen to hear
feedback.

I'd imagine eventually a backend like this might be a hard-coded backend
shipped with puppet and always there as the lowest priority backend
below any that the site might specify in their site wide hiera config so
everyone can rely on this being there and with the new lookup helpers
this should also be backward compatible - old Puppets or ones who
specifically disable the hiera indirector will just not have data and
will need to supply it some other way.

---
R.I.Pienaar

[Ashley Penney]

This is -exactly- what I've wanted in my modules since I heard of
Hiera and I am strongly supporting this proposal. I'll be
experimenting with your gem. I can't give this enough support as this
is what I wanted since Hiera began. It'll make supporting multiple
distros/operating systems much easier for modules on the forge.

> --
> You received this message because you are subscribed to the Google Groups "Puppet Users" group.
> To post to this group, send email to puppet...@googlegroups.com.
> To unsubscribe from this group, send email to puppet-users...@googlegroups.com.
> For more options, visit this group at http://groups.google.com/group/puppet-users?hl=en.
>

[jcbollinger]

On Sunday, September 30, 2012 4:37:29 AM UTC-5, R.I. Pienaar wrote:

I propose extending the module layout with a data/ directory that can go
into each module and in this data directory would live a hiera config
file (optionally) and module specific data:

That sounds really attractive, but I'm not in a position to test it right now.  Also (separately), I hope you have more distribution plans than just gem, because gem is a complete non-starter for me.


John

[R.I.Pienaar]

personally the end goal would be to just merge it with hiera or puppet.

[Peter Meier]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 09/30/2012 05:01 PM, Ashley Penney wrote:
> This is -exactly- what I've wanted in my modules since I heard of
> Hiera and I am strongly supporting this proposal. I'll be
> experimenting with your gem. I can't give this enough support as
> this is what I wanted since Hiera began. It'll make supporting
> multiple distros/operating systems much easier for modules on the
> forge.

+1. Nice idea!

~pete

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://www.enigmail.net/

iEYEARECAAYFAlBqkw0ACgkQbwltcAfKi398AACbBbGxDdWcpHBySXCSUY++yydb
D4cAn318SEhFYPqjyKymaH45vKYUzxvi
=3lWJ
-----END PGP SIGNATURE-----

[Eric Sorenson]

Hi, I'm catching up on a backlog of tickets and wondered if any of the folks who'd expressed interest in this change have had the opportunity to try it out from RI's branch:

https://github.com/ripienaar/puppet/tree/feature/master/16856

If you have, can you please put your feedback in the associated redmine ticket:

http://projects.puppetlabs.com/issues/16856

Seems like a pretty excellent feature but I don't have a production use-case I can try it against so there might be unexpected consequences. It'd be great to hear from you.

-=Eric

[ZipKid]

Hello,

I have tested the branch https://github.com/ripienaar/puppet/tree/feature/master/16856.

The basic functionality of the data in the module seems to work as desired.

Just a few things

- If only using the param auto-lookup not having a hiera.yaml file is ok

- if there is NO hiera.yaml config file, calling hiera() causes an error "Error 400 on SERVER: undefined method `include?' for nil:NilClass"

- Having an empty hiera.yaml is ok when using a hiera(call)

- The lookup in the moduledata does not happen if only using the auto-lookup of the class params, unless you manually specify "module_data" as hiera backend.

- When using the hiera() call, specifying "module_data" as hiera backend is not needed.

I hope this helps.

Regards,

Stefan Goethals.

[R.I.Pienaar]

----- Original Message -----
> From: "ZipKid" <zipki...@gmail.com>
> To: puppet...@googlegroups.com
> Sent: Tuesday, December 4, 2012 11:10:13 PM
> Subject: Re: [Puppet Users] in-module data with hiera
>
> Hello,
>
> I have tested the branch
> https://github.com/ripienaar/puppet/tree/feature/master/16856.
> The basic functionality of the data in the module seems to work as
> desired.
> Just a few things
> - If only using the param auto-lookup not having a hiera.yaml file is
> ok
> - if there is NO hiera.yaml config file, calling hiera() causes an
> error "Error 400 on SERVER: undefined method `include?' for
> nil:NilClass"
> - Having an empty hiera.yaml is ok when using a hiera(call)
> - The lookup in the moduledata does not happen if only using the
> auto-lookup of the class params, unless you manually specify
> "module_data" as hiera backend.
> - When using the hiera() call, specifying "module_data" as hiera
> backend is not needed.
>
> I hope this helps.

awesome, thanks a lot for taking the time to test it, I'll incorporate
this feedback in a few weeks.

do you think the basic approach will improve your workflow and lead
to easier to share modules?

[Garrett Honeycutt]

Absolutely leads to modules that are easier to share. This functionality
is critical for releasing modules that work on different system types
without writing a bunch of conditional logic that replicates hiera.
Outstanding work.

-g

--
Garrett Honeycutt

206.414.8658
http://puppetlabs.com

[Stefan Goethals]

This functionality is a fantastic improvement over the, already useful, data.pp functionality of the hiera_puppet backend.

For me this means i can create modules with defaults without limiting the module to the specific defaults i can think of and 

have them easily over-ridable without modifying the module itself.

Great work!

Thanks,

Stefan Goethals.

ps. Just let me know if you need more testing to be done.

[ZipKid]

Hello, 

Maybe i should be clearer and specify that the hiera.yaml i mention in my comment is  /etc/puppet/hiera.yaml and not the <module>/data/hiera.yaml version.

That last one indeed only requires a hierarchy setting for things to work.

Stefan.

[jcbollinger]

On Sunday, September 30, 2012 4:37:29 AM UTC-5, R.I. Pienaar wrote:

I propose extending the module layout with a data/ directory that can go
into each module and in this data directory would live a hiera config
file (optionally) and module specific data:

[...]

Here the data/hiera.json is optional and specifies a hierarchy that the
module author chooses and is unique to the specific backend.

The default contents would be this is the file is absent:

   {"hierarchy": ["osfamily/%{osfamily}", "common"]}

Having thought about this some more, I still think it's a great idea, but giving special support to 'osfamily' in the default in-module hierarchy seems a bit smelly to me.  I don't doubt that categorizing data by osfamily would be a frequent practice, but making a special case for it straight off the bat doesn't feel right to me.

Certainly the default osfamily support wouldn't be directly harmful to users, inasmuch as they can easily ignore or override it, but I nevertheless urge you to keep it simple.  The backlog of open tickets attests to the fact that PL doesn't really have enough engineering capacity as it is.


John

[R.I.Pienaar]

On 5 Dec 2012, at 16:43, jcbollinger <John.Bo...@stJude.org> wrote:

On Sunday, September 30, 2012 4:37:29 AM UTC-5, R.I. Pienaar wrote:

I propose extending the module layout with a data/ directory that can go
into each module and in this data directory would live a hiera config
file (optionally) and module specific data:

[...]

Here the data/hiera.json is optional and specifies a hierarchy that the
module author chooses and is unique to the specific backend.

The default contents would be this is the file is absent:

   {"hierarchy": ["osfamily/%{osfamily}", "common"]}

Having thought about this some more, I still think it's a great idea, but giving special support to 'osfamily' in the default in-module hierarchy seems a bit smelly to me.  I don't doubt that categorizing data by osfamily would be a frequent practice, but making a special case for it straight off the bat doesn't feel right to me.

Certainly the default osfamily support wouldn't be directly harmful to users, inasmuch as they can easily ignore or override it, but I nevertheless urge you to keep it simple. 

Do you have an alternative default hierarchy you'd prefer?

The backlog of open tickets attests to the fact that PL doesn't really have enough engineering capacity as it is.

Not sure what the relevance is here..

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/-/Lhdp3eelUZEJ.

[Stefan Goethals]

Not having any problem with osfamily i agree with John. 

A default to 'common' would suffice i believe.

Stefan.

[jcbollinger]

On Wednesday, December 5, 2012 8:47:32 AM UTC-6, R.I. Pienaar wrote:

On 5 Dec 2012, at 16:43, jcbollinger <John.Bo...@stJude.org> wrote:

Certainly the default osfamily support wouldn't be directly harmful to users, inasmuch as they can easily ignore or override it, but I nevertheless urge you to keep it simple. 

Do you have an alternative default hierarchy you'd prefer?

As Stefan said, just "common" would be fine, or some similar single, fixed-name file.

The backlog of open tickets attests to the fact that PL doesn't really have enough engineering capacity as it is.

Not sure what the relevance is here..

I probably shouldn't even have brought it up.  My apologies.  My point, however, was that new features are not free, even after they are written -- every line of code in Puppet has an ongoing maintenance cost.


John

[Jakov Sosic]

On 12/05/2012 09:45 PM, Stefan Goethals wrote:
> Not having any problem with osfamily i agree with John.
> A default to 'common' would suffice i believe.

Agree, common is more than enough as default.

[R.I.Pienaar]

I've updated my pull request[1] with this feedback and the bugs
Zipkid reported, any testers and more feedback welcome

[1] https://github.com/puppetlabs/puppet/pull/1217

[Stefan Goethals]

The good news:

Without /etc/puppet/hiera.yaml & Without <module>/data/hiera.yaml = OK!
Without /etc/puppet/hiera.yaml but with empty <module>/data/hiera.yaml = OK!
Without /etc/puppet/hiera.yaml but with <module>/data/hiera.yaml with a hierarchy = OK!

With empty /etc/puppet/hiera.yaml & Without <module>/data/hiera.yaml = OK!
With /etc/puppet/hiera.yaml with a hierarch & Without <module>/data/hiera.yaml = OK!
With /etc/puppet/hiera.yaml with a hierarch & with empty <module>/data/hiera.yaml = OK!
With /etc/puppet/hiera.yaml with a hierarch & with <module>/data/hiera.yaml with a hierarchy = OK!

Lookups via hiera() in a define also seem to work well!


Comment:

The default file looked up in the in-module data backend is 'default.yaml', not 'common.yaml' and
that is different from the yaml backend so it is somewhat confusing.
It is also not what was 'agreed' in this thread.

All this is based on a simple test with a module with one class and one define so it is not very in-depth
but the results are clear and as expected....

Thank you Volcane ! We're getting there. :-)

Regards,

Stefan - Zipkid - Goethals.

> --
> You received this message because you are subscribed to the Google Groups "Puppet Users" group.

[R.I.Pienaar]

yes good, actually common is a default set by hiera, this should be consistant
here too, I agree will fix.

https://github.com/puppetlabs/hiera/blob/master/lib/hiera/config.rb#L13-14

> All this is based on a simple test with a module with one class and
> one define so it is not very in-depth
> but the results are clear and as expected....
>
> Thank you Volcane ! We're getting there. :-)

no thank you, community members testing this stuff is super awesome and
help us a lot.

[R.I.Pienaar]

yay, confused myself - this is of course in the backend too, doh.

I updated the pull with this fix, thanks again.

[Stefan Goethals]

If anyone is interested, here is the puppet env i used to test this.

https://github.com/zipkid/puppet3-hiera_data_in_module

Regards,

Stefan - Zipkid - Goethals.

[ZipKid]

Hello,

What is still needed to get this issue 'moving'?

I have 2 customers where i have to start a new Puppet environment starting in January 2013 and i would really like to be able to use this functionality for those.