On RabbitMQ Cluster - node shutdown - Celery workers hang out

We have configured a rabbitmq cluster with a mirrored queue. All instructions were taken from http://ift.tt/p6Aq5B

All is well and working ok.

The configuration is:

• 
  

  3 nodes

  
  


• 
  

  1 master & 2 slaves

  
  


• 
  

  mirrored queue

  
  


• 
  

  each node on separated EC2 (AWS)

  
  


• 
  

  ELB with the above 3 instances

  
  


• 
  

  Celery connected to the ELB

  
  

The problem:

when we shutdown one of the nodes, The celery workers hang out.

We believe it is something related to the AWS ELB configuration but with no luck.

Anyone?