Could handle API rate limit errors better

[Evan]

I ran into a rate limit on HP Cloud:

  "12":
    agent-state-info: "(error: failed to get list of flavour details\ncaused by: request
      (https://region-a.geo-1.compute.hpcloudsvc.com/v2/11597203075020/flavors/detail\)
      returned unexpected status: 500; error info: <fault>\n <faultstring>Unknown
      error occurred while executing command in a step</faultstring>\n <detail>\n
      \ <errorcode>31303</errorcode>\n <trace>\n
      \ Fault Name: \n Error Type: MessageRouter\n
      \ Description: Unknown error occurred while executing
      command in a step\n Message: null\n Service:
      nova_openstack\n Endpoint: nova_openstack_client\n Operation
      (Client):defaultOperation\n FlowTransitionState : Client_Request_User_Error\n
      \ Policy : TokenRateLimit\n clientIdentification\n
      \ </trace>\n </detail>\n</fault>\n)"
    instance-id: pending
    series: precise

I seemingly have no way out of this. I can't nova delete a machine (that doesn't exit) to get juju to try again. The current rate is queryable, so juju could back off its API calls based on this, or at least know when it can retry.

[Curtis Hovey (sinzui)]

This bug might be a dupe of bug 1277397 which also wants to solve the rate limit issue. If the solution is provider-specific, then this bug is unique.