Ceph

PG::search_for_missing processes a replica's missing map to determine if it has any objects that we need.

If the missing set becomes very large, this could get very time-consuming. Instead, we probably want to just iterate over our unfound set. (Objects we know about but don't know where to find.)

This would involve tracking the unfound set separately (need a fast way to iterate over that). It also involves changing the retry logic in pull so that we can figure out some new locations to pull from when pulling from the first location fails. After all, the whole reason we try to find every location now is because pulling from some of them may not work (in theory.)

This seems fairly low priority but I'm making a note.