Puppet users - Apr 2012 - Includes and parametrized class redefinition

Hello

I have some code that works like the simplified clase shown below. The idea 
is to have a define "foo" that includes a class "foo::pre"
which contains
resources that need to be executed before the define is called. The define 
can be called multiple times but the initialization has to be done only 
once, which is why it''s implemented as a class:

class foo::pre {
  notice("foo::pre")
}
define foo() {
  include ''foo::pre''
  notice("foo")
}
class x {
  notice("x")
  foo { ''x foo'':
  }
}
class y {
  notice("y")
  foo { ''y foo'':
  }
}
include x
include y

The issue is that now I need to parametrize foo::pre so that its behavior 
depends on a variable that exists in foo:

class foo::pre($blah) {
  notice("foo::pre")
}
define foo() {
  class { ''foo::pre'':
    blah => 1,
  }
  notice("foo")
}
class x {
  notice("x")
  foo { ''x foo'':
  }
}
class y {
  notice("y")
  foo { ''y foo'':
  }
}
include x
include y

With this code I get "Duplicate definition: Class[Foo::Pre] is already 
defined". This seems weird to me because I thought the "class {
''myclass'':
}" syntax was semantically equivalent to "include myclass".
Puppet,
however, complains that it''s being defined twice, even though
there''s no
definition happening there, just inclusion.

So, is there a way to redesign this to match the original behavior? I know 
the current trend is to keep this kind of thing in hiera but this is 
already a fairly large code base that can''t be changed quickly...

Thanks in advance,
Andre

-- 
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/-/dll5ilVD_60J.
To post to this group, send email to puppet-users@googlegroups.com.
To unsubscribe from this group, send email to
puppet-users+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.

On Apr 27, 6:40 am, Andre Nathan <andre...@gmail.com>
wrote:
> Hello
>
> I have some code that works like the simplified clase shown below. The idea
> is to have a define "foo" that includes a class
"foo::pre" which contains
> resources that need to be executed before the define is called. The define
> can be called multiple times but the initialization has to be done only
> once, which is why it''s implemented as a class:
>
> class foo::pre {
>   notice("foo::pre")}
>
> define foo() {
>   include ''foo::pre''
>   notice("foo")}
>
> class x {
>   notice("x")
>   foo { ''x foo'':
>   }}
>
> class y {
>   notice("y")
>   foo { ''y foo'':
>   }}
>
> include x
> include y

Great, no problem.

> The issue is that now I need to parametrize foo::pre so that its behavior
> depends on a variable that exists in foo:
>
> class foo::pre($blah) {
>   notice("foo::pre")}
>
> define foo() {
>   class { ''foo::pre'':
>     blah => 1,
>   }
>   notice("foo")}
>
> class x {
>   notice("x")
>   foo { ''x foo'':
>   }}
>
> class y {
>   notice("y")
>   foo { ''y foo'':
>   }}
>
> include x
> include y
>
> With this code I get "Duplicate definition: Class[Foo::Pre] is already
> defined". This seems weird to me because I thought the "class {
''myclass'':}" syntax was semantically equivalent to
"include myclass". Puppet,
>
> however, complains that it''s being defined twice, even though
there''s no
> definition happening there, just inclusion.

This is one of the several drawbacks of Puppet''s implementation of
parameterized classes. It may be that "class {
''myclass'':}" is
semantically equivalent to "include ''myclass''", but
that''s beside the
point.  Unlike ordinary classes, parameterized classes can only be
included / declared once.  It''s not necessarily the declaration syntax
that makes the difference, but rather the nature of the class being
declared.

> So, is there a way to redesign this to match the original behavior? I know
> the current trend is to keep this kind of thing in hiera but this is
> already a fairly large code base that can''t be changed quickly...

I know it''s not what you want to hear, but Hiera is your best bet.  I
don''t think the code base size is relevant, because I don''t
think the
time and effort to implement an Hiera-based class data source is
likely to differ much from what parameterizing all the same classes
would require.  Or to put it a different way, solving your problem via
class parameterization would require *at least* as much shakeup of
your code base as would implementing an hiera-based solution.

If you insist on using parameterized classes, then you have to come up
with a way to ensure that they are declared exactly once if they are
needed.  If it is harmless to declare them when they are unneeded,
then you could declare them unconditionally for every node; otherwise
you have a mess to sort out.  It is my impression that people with
that sort of mess usually end up relying on a complex ENC to deal with
it.


John

-- 
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+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.

Hello John

On Friday, April 27, 2012 9:58:09 AM UTC-3, jcbollinger
wrote:
>
> I know it''s not what you want to hear, but Hiera is your best bet.
I
> don''t think the code base size is relevant, because I
don''t think the
> time and effort to implement an Hiera-based class data source is 
> likely to differ much from what parameterizing all the same classes 
> would require.  Or to put it a different way, solving your problem via 
> class parameterization would require *at least* as much shakeup of 
> your code base as would implementing an hiera-based solution. 
>
Well, the thing is that we''re in the middle of a transition process
that is
moving from "everything is a global variable in the node" to
parametrized
classes, and while they are not perfect, I found that our new code is much 
saner and easier to debug when using them.

So the question is, are parametrized classes now considered
"deprecated"? I
remember reading somewhere that improvements were being made for Puppet 
2.8... While I can see the advantages of Hiera, it seems to me that
it''s
another instance of the global variable problem if it''s used to load
values
inside some class, and I''d rather not lose the benefit of being able to
check a class "signature" to see immediately what variables it needs,
and
having the code fail if any is not provided.

Thanks,
Andre

-- 
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/-/eA93Ge_3z2kJ.
To post to this group, send email to puppet-users@googlegroups.com.
To unsubscribe from this group, send email to
puppet-users+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.

> to me that it''s another instance of the global variable problem if
it''s
> used to load values inside some class, and I''d rather not lose the
> benefit of being able to check a class "signature" to see
immediately
> what variables it needs, and having the code fail if any is not provided.
If I''m understanding you right, then I get around that problem using
the
%{calling_module} variable passed from hiera-puppet.

For example

class foo::data ( $somevar = hiera("bar") ) {

When used with a hierarchy similar to

%{environment}/%{calling_module} in my

Means I can store all my "foo" variables up into, for example,
dev/foo.yaml

It''s a nice way to be able to see at-a-glance how "foo" is
configured in
my environment, and also helps with ambiguous variable names (eg: $port) 
as grouping them this way offers a kind of scope.

Craig


-- 
Craig Dunn | http://www.craigdunn.org
Yahoo/Skype: craigrdunn | Twitter: @crayfishX

-- 
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+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.

On Apr 27, 8:12 am, Andre Nathan <andre...@gmail.com>
wrote:
> Hello John
>
> On Friday, April 27, 2012 9:58:09 AM UTC-3, jcbollinger wrote:
>
> > I know it''s not what you want to hear, but Hiera is your best
bet.  I
> > don''t think the code base size is relevant, because I
don''t think the
> > time and effort to implement an Hiera-based class data source is
> > likely to differ much from what parameterizing all the same classes
> > would require.  Or to put it a different way, solving your problem via
> > class parameterization would require *at least* as much shakeup of
> > your code base as would implementing an hiera-based solution.
>
> Well, the thing is that we''re in the middle of a transition
process that is
> moving from "everything is a global variable in the node" to
parametrized
> classes, and while they are not perfect, I found that our new code is much
> saner and easier to debug when using them.
>
> So the question is, are parametrized classes now considered
"deprecated"?

No, they are not deprecated.  "Raw" would be a better description: the
implementation of parameterized classes in 2.6 (where they were
introduced) and 2.7 has always had problems, especially including the
one you asked about.

> I
> remember reading somewhere that improvements were being made for Puppet
> 2.8...

I have it on good authority that parameterized classes will be greatly
improved for Puppet''s next major release, "Telly".  I have
been told
that among other things, they will be integrated with Hiera in a way
that sounds very useful.  I am uncertain, however, whether the use
case you seem to be looking for will be supported.  That is, in Telly
you should be able to include a parameterized class as many times as
you want if you rely only on parameter defaults (which themselves will
be improved for Telly), but I don''t know whether you will be able to
do the same if you specify custom parameter values.

> While I can see the advantages of Hiera, it seems to me that it''s
> another instance of the global variable problem if it''s used to
load values
> inside some class, and I''d rather not lose the benefit of being
able to
> check a class "signature" to see immediately what variables it
needs, and
> having the code fail if any is not provided.

I think you will like Telly''s improvements in this area, at least if
they work as was explained to me.  Until then, however, you simply
cannot use parameterized classes in the way you are trying to do.  It
does not work.

Use external data from Hiera instead of class parameters, with Hiera
keys corresponding to the class and parameter names that you otherwise
would implement.  That''s a sane and manageable design in its own
right, and it will also position you well to transition to Telly when
it come out.  Document your classes with plain comments (which you
should do anyway), and you have most of what you were looking for,
except Puppet automatically failing if a class parameter is omitted.


John

-- 
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+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/puppet-users?hl=en.