Tiny Puppet Experiments

Tiny puppet is a new idea by Alessandro Franceschi to provide a simple skeleton for basic puppet modules. The intention is to provide a Proper Opinion on core parameters. Alessandro talked to me about how there is still no totally satisfying solution to defining the module data in TP.

Today I’ll take a look at doing so.

Start with activating travis support. And immediately upgrading all gems. Now actually only a single spec fails. Progress! This seems to be exactly the core problem Alessandro’s been talking about.

To get a broader view on the problem space, I read R.I.Pienaar’s module-data module, and try to ingest the potential capabilities of the tp face and everything else in the module.

One problem I see is how tp tries to shoehorn everything through the tp:: namespace, leading to twofold problems:

  1. Complex lookup logic (not yet implemented). This deviates from the lookup structure popularized by hiera. Breaking away from this means that hiera customizations cannot be re-used with this structure and users have to learn new rules.
  2. On the code side, this leads to definitions like

    tp::conf { 'redis::red.conf': ...

    which should better read

    redis::conf { 'red.conf': ...

Keeping with common hiera patterns for #1 would require some kind of translation of actual arguments into the tp world. Since Alessandro’s Handy Grail post from last year, I was toying around with a convention-over-configuration based approach to the issues laid out there. Maybe these ideas could be combined here. Having a one-liner to pass arguments to a core tp define would also make #2 easier.

To reality-check my experiment, I’m looking over puppetlabs-mysql to get a realistic view on the feasibility of converting the module to a tp core.

puppetlabs-mysql notes

  • mysql::params looks like the seventh circle of Dante’s hell. Luckily, the defaults are well factored, so converting to module-data should not be complex. Only tedious. There’s only a single occurrence of a none parameter-default reference into mysql::params, so there’re no other dependencies.

  • Core mysql parameters are managed via a generic $override_options hash, that is merged into defaults. These options are partly supplied by other params variables as they are distro-specific.

  • some parameters are deprecated, requiring a $real_ variable. I can think of some hacks to workaround this by breaking into the databinding code and fixing things up there. I’m not sure that is implementable (or sensible) though.

    TODO: It’s a general problem and thus should be addressed somehow.

  • mysql::server::install optionally creates the initial datadir. On the one hand this is laudable, since some distros choose to deliver mysql in a non-functioning state. On the other hand, this functionality is buried in a secondary class under a very strange conditional, mixing checks of the merged options with checks on the user-specified options. The former is annoying as it is not usable for multi-instance installations, the latter looks like a bug to me. Perhaps the original author wanted to avoid creating the datadir when the default location was used?

PS: Now Playing: New Electro & House 2013 Chart Mix