Instead of supporting implicit collection creation (which is risky in any scenario) createIndex on an existing collection's cluster key with 'clustered' = true should be a no-op. Otherwise, throw if the collection does not already exist or already exists as clustered.