Except people end up running systems on them that have their own replication built in like kafka, cockroach etc. In all those cases you’d be better off with you

Except people end up running systems on them that have their own replication built in like kafka, cockroach etc. In all those cases you’d be better off with your node just hard crashing if disk goes bad rather than the latency penalty that inevitably follows the failover