Previously, when a virtual machine was paused due to IO error, the only resume policy was "Auto Resume", which resumed the virtual machine. "Auto Resume" was problematic because it could interfere with custom HA solutions. In the current release, the "Kill" and "Leave Paused" resume policies have been added. "Leave Paused" was introduced for users who prefer this option because they have their own HA implementation. "Kill" allows virtual machines with a lease to automatically restart on another host in the event of irrecoverable outage.
The speed of IO Error reporting depends on the underlying storage protocol. On FC storage, IO Errors are generally detected quickly, while on NFS mounts with typical default settings, they may not be detected for several minutes.