During cluster startup the Scheduler will return an error until the local backend has started up ("Local backend has not been registered in the cluster membership"). Afterwards, it will assume that the default executor group exists. However, if the coordinator is not also an executor (i.e. it is a dedicated coordinator), then it will not actually create the default executor group in cluster-membership-mgr.cc:256.
Queries are expected to fail in this scenario, but the error message should certainly be improved to indicate that no executors could be found. For this purpose, we should make sure that the default executor group gets created as soon as the local backend has started, but keep it empty if it is not an executor. Then we can warn in the scheduler that no executors have been registered so far.