Why is that necessary? One of the things which makes maintaining a search index easier is that typically your search datastore has much weaker consistency and durability requirements than your primary one.
Because in our application, users can add new documents, and will be confused if the document they just added is missing in the search result (which by default show the most recent documents). Considering that Elasticsearch website and documentation emphasize "real-time" features, I think it's not a weird requirement :-)
Why is that necessary? One of the things which makes maintaining a search index easier is that typically your search datastore has much weaker consistency and durability requirements than your primary one.