In some applications, indexing and searching are important, but not nearly as much as the rest of the application. If Elasticsearch is down, these applications would rather continue operating as usual and just log the errors.
This means in particular that:
Bootstrap must continue even if the Elasticsearch cluster cannot be reached. It should be possible if the Elasticsearch version was provided as part of the configuration.
Exceptions thrown during indexing should not be propagated to the application, but should rather be logged.
We'll probably need some sort of global configuration that affects both the ORM mapper (for item #1) and the Elasticsearch backend (for item #2).