Restarting a service when any hook changes is too extreme and leads to unnecessary restarts

[christophermaier]

Currently, whenever any lifecycle hook file is updated in response to census changes, we trigger a restart of the service.

Based on how the hooks in question are actually run, this is actually far too aggressive, and could lead to services restarting more than is strictly necessary.

For example, the post-stop, suitability, file_updated, and health_check hooks are all run on-demand, and actually outside of the "main loop" of the service hook lifecycle. They cannot, by definition, impact how a service process runs. Yet, if any of them change, we restart the service.

Similarly, the init hook is only run when the service is first loaded, or when it has been updated with a new release from Builder. It does not run each time before the main run hook does. Therefore, any changes to this hook are actually completely irrelevant, because it won't be run again anyway.

The reload hook is a bit of an outlier (see #5306, #5307), and a change to the reconfigure hook seems like it should just trigger a reconfiguration, rather than a restart, since that's its sole purpose.

Changes to the run hook should definitely trigger a restart. Changes to post-run could also trigger a restart (but also see #2364).

[baumanj]

"Second"