Without doing this a graceful loss of connection with the control host is ignored and if the whole cluster is to be restarted after that, cql-io does not get any more events. When the last node in the cluster is restarted, cql-io gets into an unrecoverable state where it thinks the control host is connected, but also thinks that no hosts are reachable.
The CI is failing due to https://github.com/TomMD/entropy/issues/75, hopefully it will get fixed soon.