It looks like the Elasticsearch backend doesn't propagate exceptions to the completable future returned by IndexingPlans, in particular. As far as I can tell, the Lucene backend does.
I think the correct behavior would be to propagate the exception, so that if users configured an automatic indexing strategy that requires commit or more (i.e. everything except "queued"), they will get an exception in their code. If they don't want that, they can always set the indexing strategy to "queued".
We should add tests for this, preferably integration tests (for the backends only, for now) and adjust existing unit tests.