← Back to team overview

yahoo-eng-team team mailing list archive

[Bug 1226915] Re: agents should bind to specefic ports

 

** Changed in: neutron
       Status: New => Invalid

** Changed in: neutron
     Assignee: Sean McCully (sean-mccully) => (unassigned)

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1226915

Title:
  agents should bind to specefic ports

Status in OpenStack Neutron (virtual network service):
  Invalid

Bug description:
  Based on comments from review of Change-Id:
  Id3adf4aad03ed4573857776ec92dabb0d24d0514

  There is a need for better state management on the agent side for
  control of ports. Solution from agent, is change agent_id to a value
  that can correctly identify an agent as a unique agent from host to
  host.

  ================Francois Eleouet====================================
  I'll try to elaborate on need to use cfg.CONF.host instead of this id forged with br-int mac address:

  Currently, when a port is migrated, the very last message plugin will
  recieve regarding this port is an update_device_down from the original
  agent (after the update_device_up is reieved from the target agent).
  As a consequence, port status will remain to down in database (this is
  the bug Mathieu was referring to in the previous comment). The
  recently introduced l2-population feature also relies on accurate port
  status management, this is why we'd like to fix these migration
  issues.

  The straightforward way to fix this issue would be to filter
  update_devices_up/down on plugin side to only allow agent on which
  port is bound to update its status (if nova manages to provide plugin
  with accurate port binding information when it migrate a VM, see [1]).
  The issue is that currently plugin has no way to figure out from which
  agent such messages are coming, as agent_id which is being build here
  for RPC messages isn't known (neither used) by plugin.

  For sure, this bugfix could be taken as is, but in that case we'll probably have to change it back in the future, this is why we were proposing to do it here, as using cfg.CONF.host would help solving both issues.
  =====================================================================

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1226915/+subscriptions