The downtime isn’t the biggest issue, you can plan around that. The real problem is the random lag on some queries and volatile memory being lost (usually when there’s a context change).