There's actually some precedent with very high volume services taking the minimal downtime approach to avoid the increased amount of development time. Take Airbnb for example [1]. They chose to take two weeks to migrate their Messaging database with a few minutes of downtime rather than take longer to do it with no downtime. In the article, they discuss how they thought about the tradeoffs.

[1] https://medium.com/airbnb-engineering/how-we-partitioned-air...