+ # Before starting deployment we disorder the group list with the
+ # purpose of speeding up the whole deployment process.
+ # It is likely that the user inserted in the 'group' list closely
+ # resources resources one after another (e.g. all applications
+ # connected to the same node can likely appear one after another).
+ # This can originate a slow down in the deployment since the N
+ # threads the parallel runner uses to processes tasks may all
+ # be taken up by the same family of resources waiting for the
+ # same conditions.
+ # If we disorder the group list, this problem can be mitigated
+ random.shuffle(group)
+