It is possible for shards to enforce different constraints for inserts and updates due to having different index definitions. Our data migration protocols do not allow inserting documents into a shard which would violate these constraints. (And corrupting an index on another shard to allow removing a shard to succeed is not a tradeoff we're willing to make.) This means removing a shard may leave that shard in the draining state for an indefinite amount of time until a client manually intervenes. Eagerly failing the removeShard command may be preferable to avoid the system from becoming stuck in this state.