[chef] Re: Re: attributes constructed from other attributes aren't added to the attribute map

[Stefán Freyr Stefánsson < >]

Well... for future reference I'm replying to myself with a solution that 
seems to work.

We knew that composed attributes were being used in other "official" 
cookbooks (such as the main apache cookbook) so I dug into that to see how 
things were done there. The only thing I noticed was that they used a 
slightly different syntax.

So instead of:

default['chef_attribute_hell']['attr1'] = "attribute1"
default['chef_attribute_hell']['attr2'] = 
"#{node.chef_attribute_hell.attr1}/attr2"
default['chef_attribute_hell']['attr3'] = 
"#{node.chef_attribute_hell.attr2}/attr3"

They were doing it like this:

default['chef_attribute_hell']['attr1'] = "attribute1"
default['chef_attribute_hell']['attr2'] = 
"#{node['chef_attribute_hell']['attr1']}/attr2"
default['chef_attribute_hell']['attr3'] = 
"#{node['chef_attribute_hell']['attr2']}/attr3"

I tried that in my cookbook and voila.... now everything seems to work.

Still not quite sure how the different syntaxes affect the compile time of 
the node attribute map but hey, at least now it works.

Kind regards and merry Christmas,
Stefan Freyr.

________________________________________
From: Stefán Freyr Stefánsson
Sent: Tuesday, December 23, 2014 11:00 AM
To: 

 
Subject: Re: [chef] Re: attributes constructed from other attributes aren't 
added to the attribute map

Hi.

Thanks for your answer. Unfortunately I'm not quite sure if I'm understanding 
it correctly, at least my understanding of the fix does not work.

What I did was I added 'node.' in front of my 'default['foo']['bar'] = ..." 
in the attributes/default.rb file in my cookbook. Here are the full contents 
of the file as they stand after the change:

node.default['chef_attribute_hell']['attr1'] = "attribute1"
node.default['chef_attribute_hell']['attr2'] = 
"#{node.chef_attribute_hell.attr1}/attr2"
node.default['chef_attribute_hell']['attr3'] = 
"#{node.chef_attribute_hell.attr2}/attr3"

This still results in:

       NoMethodError
       -------------
       Undefined method or attribute `attr2' on `node'

       Cookbook Trace:
       ---------------
         
/tmp/kitchen/cookbooks/chef_attribute_hell/attributes/default.rb:3:in 
`from_file'

       Relevant File Content:
       ----------------------
       /tmp/kitchen/cookbooks/chef_attribute_hell/attributes/default.rb:

         1:  node.default['chef_attribute_hell']['attr1'] = "attribute1"
         2:  node.default['chef_attribute_hell']['attr2'] = 
"#{node.chef_attribute_hell.attr1}/attr2"
         3>> node.default['chef_attribute_hell']['attr3'] = 
"#{node.chef_attribute_hell.attr2}/attr3"
         4:

Finally, you say that this is a "very rare edge case". All I did was make a 
cookbook with attributes that are constructed from values of previous 
attributes. Is that really an edge case? Is there some other way to 
accomplish the same thing? Because this seems like a bit of a fundamental 
thing to me.

Kind regards, Stefan Freyr.

________________________________________
From: AJ Christensen 
<
 >
Sent: Monday, December 22, 2014 7:19 PM
To: 

 
Subject: [chef] Re: attributes constructed from other attributes aren't added 
to the attribute map

The syntax is not wrong, this is unfortunately a very rare edge case
with the way that attribute reads are evaluated.

You can move any computed attributes to the Recipe's Compile Phase in
order to Not Be Bitten By Things., e.g.:

``` ruby
node.default['chef_attribute_hell']['attr2'] = "#{...}/..."
node.default['chef_attribute_hell']['attr3'] =
"#{node.chef_attribute_hell.attr2}/attr3"
```

I wish the notes from the "Attributes WAT" community session had been
captured but it was a very high energy discussion -- we went into the
details as to what causes this to occur. Something to do with the
"read time" of the attributes in the attributes DSL being before some
attribute precedence resolution happens, I fear, although I don't
recall. HTH.

cheers,

--aj

On Tue, Dec 23, 2014 at 7:38 AM, Stefán Freyr Stefánsson
<
 >
 wrote:
> Hi.
>
> I just started using test-kitchen again after a bit of a break. I was
> surprised to find out that a very simple internal cookbook that we've been
> using now didn't work.
>
> The error occurred when evaluating our default attribute file. It was
> complaining about an attribute (attr3) that was composed with the value of
> another attribute (attr2), which in turn was composed from the value of
> another attribute (attr1). After adding some "raise" lines to the default.rb
> attribute file in our cookbook and a test recipe we were able to determine
> that attr1 was added to the map, while attr2 was not added and trying to
> construct attr3 then resulted in a failure.
>
> Sorry for the convoluted explanation, I make some sad attempt to describe
> this here: https://gist.github.com/StFS/2a333afe352d0dc8cbb5
>
> As I said I was using test-kitchen for this and in my .kitchen.yml file I
> had:
>   require_chef_omnibus: true
>
> This installed Chef 12.0.3-1 on the vagrant machine (which is a CentOS 7).
> Since this had been working previously I dug around a bit and noticed that
> all of our earlier installations seemed to be using version 11 of Chef. So I
> changed my .kitchen.yml file to specify version 11 of chef and now at least
> I'm getting further and the attributes map seems to be built correctly in my
> recipe.
>
> So my question is, is the following syntax for my attribute file in some way
> wrong? If so, what should I be using? If the syntax is not wrong, is this
> then a bug in Chef 12? Has it been reported (I didn't find anything during
> my googling around for this problem)?
>
> Here is an example of an attribute file that would fail:
>
> default['chef_attribute_hell']['attr1'] = "attribute1"
> default['chef_attribute_hell']['attr2'] =
> "#{node.chef_attribute_hell.attr1}/attr2"
> default['chef_attribute_hell']['attr3'] =
> "#{node.chef_attribute_hell.attr2}/attr3"
>
> Kind regards,
> Stefan Freyr
>
> p.s. I can make a full blown cookbook with a .kitchen.yml file available
> which demonstrates this problem if needed.