Snapshots currently first captures the current index metadata and only then writes out the shard data. This leaves room for mapping updates that have taken place after the index metadata was captured but before the shard data was written to not be taken into consideration when the index is restored. Data that was searchable under a certain key, suddenly isn't, and doing a simple update-by-query with no changes can lead to mapping updates. Possibly worse, the same index, when restored can have a mapping update that's incompatible with the existing data in the restored index.