The logic implemented on client is (for started instances):

Notify instances are available to master
run promises if fail report error
Notify instances are started to master

The change on this commit prevent us to consider the notification of availability as "no error", as promise can still fail, so we just ignore and let the decision to suspend for the next run.