Agenda:

• How and where do we host pulse rabbitmq
• What conditions should we impose on clients and add to specification
  • reconnect
  • ack
    
    

----------------- Random Thoughts -----------------

Goal:

• Deliver at-least-once semantics
• No or (minimal) planned downtime
• Forgetabiliity :)
• Robustness over high availability (pause minority)
  
  

Options:

• EC2 managed by IT
• CloudAMQP
  
  

Configuration:

• 3 nodes (possibly multi az)
• Pause minority
  
  

Upgrades:

• head->sand (ignore the 2min downtime, and pretend that didn't just happen)
• remap DNS and move queues over with federation, force clients to reconnect
  
  
  
  
  
  

Scheduling of interim work? (increase of rabbitMQ file descriptors to increase # of sockets avail.)