Would you please add steps to reproduce?
Steps to reproduce:
1. Create a cluster.
2. In the DB, set cpu_name for this cluster to empty value (update vds_groups set cpu_name='' where name='<cluster name>').
3. Try to create a VM in this cluster.
since its not clear how this is reproducible without changing the db, and the result is a bad message for the user, that is using a bad cluster, i think we should not backport the fix to 3.5.z as well.
closing with wont-fix for this version, but note this already fixed in 3.6.0
please re-open if you consider this a must in 3.5 as well.