In sharded configurations, indexes are copied during chunk migrations. If there are no migrations, then there's no automatic mechanism to make indexes consistent across shards.

There's a background thread that we can use to detect missing indexes. If there are, we'd need to re-run the same createIndex command.

This ticket should investigate whether mongos needs to internally retry the createIndexes on NotPrimary-esque errors or whether that should be left to the replicator.