Good question Exceptions in close are tricky...
In some places we try hard to keep closing stuff, and then throw the
first exception we hit. Probably that should be our preferred
approach, in general, for readers? (My guess is most apps simply
catch & log the IOException, so, we should try hard to close all that
we can on the first go.)
In other places (SR, DR) we don't.
In general you can retry a close – calling close more than once on
Lucene classes is allowed (or, we aim for that, but likely not all
classes achieve it).
For writers (IndexWriter, and DR with pending changes) we try hard not
to lose state on hitting an exception during close. Meaning, on
hitting an exception in close, either 1) your state already made it
into the index, or 2) if you fix the root cause (free up disk space,
fix permisisions, whatever, which of course hardly any app is going to
be doing) and retry the close, it would succeed, and your index would
be fine and your changes are committed.