Rudder

Some more precisions:

- this a pre-hook / post-hook general mecanism, need to factor out common actions that need to be done only one time, for a run, for all the directives from the same technique (OR for the whole run for all techniques ?)

- examples are:
- restart a service if it was modified during the run
- visudo edit and replace in one go (NOTE: it may be more logic to have only atomic visudo)
- get all package repository keys (because extremelly long)
- update package list (or any other long running process that need to be done only time per run, and even must be done only one time to have some consistency along the run)

Thought iteration +1:

• in fact, we just need a to had a pre-trigger and post-trigger. The triggers come from a new technique standard file, named "action-triggers.json".
• the action-triggers.json contains pre- and post- hook registration
• a hook line is composed of a hook type (service restart, command execution, etc), a condition (a class expression), list of reportIds, and some parameters,
• on node policy generation, we merge all pre- and post-hooks for a node, with some merging logic:

• for a some hook type, with same parameters, we OR class condition and |+| reportId,

• we add the pre- and post- code logic in the standard system bundle list

In fact, it is more logical to put the information into metadata.xml in place of new ad-hoc json files, since we won't use the same format agent-side. See updated ticket description for details.

We forgot to manage the case where we have two directive with the same hook/parameters (say, "service-restart"), one in "enforce" and the other in "audit".

In that case, we need to "enforce" the restart, but the reporting for the audit directive will be broken (not expecting a change).

So, we need to not only give the reportIds but also the corresponding policy mode (and the method implementation will have to do the case matching when writting reports).