Configuration of prometheus::server fails when looking up configname

[madelaney]

Affected Puppet, Ruby, OS and module versions/distributions

• Puppet: 4.10.12
• Ruby: ruby 2.1.9p490 (2016-03-30 revision 54437) [x86_64-linux]
• Distribution: Ubuntu 16.04.5 LTS
• Module version: v6.2.0

How to reproduce (e.g Puppet code you use)

node {
include ::prometheus
include ::prometheus::server
}

What are you seeing

The puppet catalog fails to compile.

What behaviour did you expect instead

I'd expect the catalog to compile and run.

Output log

Error: Could not retrieve catalog from remote server: Error 500 on SERVER: Server Error: Evaluation Error: Error while evaluating a Function Call, Lookup of key 'prometheus::server::configname' failed: Data Provider type mismatch: Got String when a hash-like object was expected to access value using 'distro' from key 'os.distro.codename' at /etc/puppetlabs/code/environments/production/manifests/site.pp:41:1 on node <machinename>

Any additional information you'd like to impart

I also tried to follow the example to see if that would help but it was all for not.

[bastelfreak]

Hi @madelaney, can you share your site.pp? At least the code around line 41.

[madelaney]

@bastelfreak Sure. This will be a two part, for full context:

site.pp:41 is
hiera_include('classes', [])

The classes array is

---
classes:
- prometheus
- prometheus::server

[bastelfreak]

I don't think that the error relates to this module. However I can add an acceptance test for this later.

[madelaney]

Um. I can play around with this, but I'm not sure I have enough configuration for prometheus to cause an issue. I will admit, this is my first foray into prometheus so I do claim ignorance on the subject matter.

[madelaney]

There is a similar issue reported to the puppet-ntp (see MODULES-4103). This seems to point to an issue on my side, and to your point, not this module.

Unlink the aforementioned issue, I'm on a newer version of Puppet/Facter

sudo /opt/puppetlabs/bin/puppet agent --version
4.10.12
sudo /opt/puppetlabs/bin/facter --version
3.6.10 (commit ecf783f2ac05eb026d820524df45ee3dbf01bfe1)

The root cause seems to come from the hiera lookup file.

For what it's worth, I'm also using the NTP module (v7.2.0) but I don't see the error listed in MODULES-4103.

[madelaney]

After some digging, this is a facter bug where the variable $os cannot be overloaded. Would you be open to a P.R. that replaces the variable $os with $os_lc? If not, do you have another term in mind?

[bastelfreak]

@madAndroid which version of facter do you run? The bundled version in Puppet 4.10 already supports hashes. Can you do this on the cmdline please:

which facter
facter --version
facter -p os

[madelaney]

@bastelfreak , I think you're mention had the wrong user (maybe?).

If so, here's the data you requested:

which facter
facter: Command not found.
Exit 1

/opt/puppetlabs/bin/facter --version
3.6.10 (commit ecf783f2ac05eb026d820524df45ee3dbf01bfe1)

/opt/puppetlabs/bin/facter -p os
2018-08-31 09:00:58.005141 WARN  puppetlabs.facter - skipping external facts for "/home/mdelaney/.puppetlabs/opt/puppet/cache/facts.d": No such file or directory
{
  architecture => "amd64",
  distro => {
    codename => "xenial",
    description => "Ubuntu 16.04.5 LTS",
    id => "Ubuntu",
    release => {
      full => "16.04",
      major => "16.04"
    }
  },
  family => "Debian",
  hardware => "x86_64",
  name => "Ubuntu",
  release => {
    full => "16.04",
    major => "16.04"
  },
  selinux => {
    enabled => false
  }
}

[bastelfreak]

Sorry, I wanted to highlight you. Do you somewhere reference the os.distro.codename? this seems to come from your codebase, not from our module. Do you use it in your global hiera.yaml or somewhere in the site.pp or are there more values for the classes hiera key?

[madelaney]

I do not, I do have the following (which is similar) but I do not have os.distro.codename exactly.

        {
          'name'  => 'per-os defaults',
          'paths' => [
            'osfamily/%{facts.os.name}/%{facts.os.distro.codename}/defaults.yaml',
            'osfamily/%{facts.os.name}/%{facts.os.architecture}.yaml',
            'osfamily/%{facts.os.name}/defaults.yaml'
          ]
        },

[bastelfreak]

this code references the s.distro.codename fact so I assume that is the root cause.

[madelaney]

Okay, I'll do some digging and try to get to the bottom of it. Sorry for the spam.

[spali]

I think, I have a similar issue:

Error: Evaluation Error: Error while evaluating a Resource Statement, Lookup of key 'prometheus::node_exporter::download_extension' failed: Data Provider type mismatch: Got String when a hash-like object was expected to access value using 'family' from key 'os.family' at /data/puppetrepo/manifests/init.pp:29:5 on node testnode.domain.net

the manifest at that line contains:

    class { 'prometheus::node_exporter':
      version => '0.17.0',
    }

# which facter
/opt/puppetlabs/bin/facter

# facter -p os
{
  architecture => "x86_64",
  family => "RedHat",
  hardware => "x86_64",
  name => "CentOS",
  release => {
    full => "6.9",
    major => "6",
    minor => "9"
  },
  selinux => {
    enabled => false
  }
}

# puppet agent --version
4.10.9
# facter -v
3.6.8 (commit 25dace2536f6bd339b70b385c7e90e1951870671)
# hiera -v
3.3.2

First, I assumed it has something to do with the lookup in hiera.yaml in the module.

puppet-prometheus/hiera.yaml

Line 12 in 757a5b4

path: "%{facts.os.family}.yaml"

But even after replacing the hiera lookup with hardcoded values, I get the same error... but I can't find where else it it referenced in the module.
As soon I remove the above class reference, it works.

[bastelfreak]

@spali which version of the module do you use? Did you also test the master branch? Can you share the whole /data/puppetrepo/manifests/init.pp file?

[spali]

I used 6.4.0. I can reproduce it with master.
But not if I use a test manifest with only the above call.
I will further dig into it, how it can be reproduced or if I have some weird bug somewhere.

Side note: found another bug on CentOS with sysv init system.
#290

[spali]

Got my error... I had in the main hiera.yaml a reference of "osfamily/%{os.family}.yaml".
As soon as I replaced it with "osfamily/%{facts.os.family}.yaml" it worked...
Don't ask me why it only occurs if I add this module :(

[bastelfreak]

ah, interesting. Thanks for debugging!

[aptituz]

@bastelfreak

I'm a bit confused why this issue has been closed, because the problem still exists. When including prometheus::server from a profile that error pops up with puppet-agent 5.5.18.

The issue seems to stem from the fact that the module defines a parameter $os which apparently conflicts with the facter-provided os-fact. Renaming the parameter to - say - os_ls makes puppet compile a catalog flawlessly.

I really believe that this is related to a bug in facter (as @madelaney suggested in his comment: #254 (comment)) and/or hiera, but given that the module claims compatibility with the puppet-version mentioned above, wouldn't it anyway make sense to use a different name for the parameter?

Regards,
Patrick

[bastelfreak]

I closed this because our tests cannot reproduce this and people reported that they had an issue within their own codebase :)

Renaming the variable might help, but that would be a breaking change and we still cannot reproduce the error.

• Can you please tell us your facter version?
• Which version of the module do you use?
• Do you use the Puppet AIO packages / on which OS are you?
• Please show the Puppet/hiera code you use for this module

[KlavsKlavsen]

Doing this (without any other calls to prometheus) - breaks it for us - on latest puppet 6.

class {'prometheus::node_exporter':
install_method => 'package',
package_name => 'obmondo-node-exporter',
package_ensure => ensure_latest($enable),
bin_dir => '/ops/obmondo/nagios/plugins',
bin_name => 'node_exporter',
collectors_disable => ['loadavg', 'mdadm'],
extra_options => '--collector.ntp.server ntp1.orange.intra',
}

[KlavsKlavsen]

Was actually caused by same bug - as previous poster.. we had one place in a hiera.yaml referring to %{os.family} instead of %{facts.os.family} - so great bug to get flushed out, in our own code :)

[bastelfreak]

always happy to help :D

[lonebrave]

Having same issue as originally reported (Got string when hash-like object was expected...). Determined that it seems to be a conflict with the module's use of $os conflicting with the $os fact. I've resolved in my environment by forking the module and refactoring the $os variable as $os_type. Once the variable name was changed, the errors went away. I'm happy to submit a PR if the team will consider this solution.

My environment is
puppet server:
CentOS 7.9.2009
puppet 7.5.0
facter 4.0.52
puppetserver 7.1.0

prometheus server:
Ubuntu 20.04.2
puppet 7.5.0
facter 4.0.52

[TheMeier]

Puppet 7 EOL is 2 month away. With Puppet 8 this should not be an issue since $os not automatically refer to a fact.

So can we close this? Or are there other reasons to rename the variable everywhere?