Abruptly terminating a VM is quite common in cloud services, can "unapply" a write, because it's equivalent to a power outage. The data lives in the VM guest's

Abruptly terminating a VM is quite common in cloud services, can "unapply" a write, because it's equivalent to a power outage. The data lives in the VM guest's cache until the fsync sends it to the host. With VM termination in the cloud being not exactly uncommon, I would say that's a general enough failure mode in a distributed system that it's an unacceptable risk to run MongoDB in its default configuration (or any database in the equivalent configuration).