In the context of live-migrating Virtual Machines (VMs); if taking place over a wide area, there is a chance that the IP address will change between source and destination.
It can be a problem that the user caches the DNS resolution for the original location of the VM, only to find out that the VM is not responding anymore (the original VM was deleted after migration). This problem is mostly true when the user is not connected during the migration, and can therefor not be notified of the change.
While it may be viable to reduce the cache's TTL to very small numbers, I was wondering if there were not a mechanism, e.g. as described in the ipv4/ipv6 RFC, that could make the client automatically evict the DNS cache A/AAAA record entry upon a connection error, and attempt a reconnection.