The generally recommended way to use a Client is to persist one for your entire application and reuse it across requests, in order to take advantage of the shared resources e.g. connection pool and monitoring threads.

We should document this and explicitly note that the Client automatically handles connection pooling, as this has been a point of confusion for users in the past due to being a different model from how e.g. Postgres drivers tend to work, and has led to likely incorrect impressions about poor driver performance.