Open Stack

    Hi!,
   
    I want to ask, specifically, about the intended purpose of the
host=... parameter in the neutron-agents (undocumented AFAIK).

    I've been checking a reference HA setup, where one of the issues
was having an HA l3-agent in active/passive or active/active.

    One of the configurations we came up with, was setting two network
nodes, in active/passive, and using the "host=...." config field in
agents. 
    
   Network Node 1:  /etc/neutron/l3_agent.ini   host=ha-nnode
   Network Node 2:  /etc/neutron/l3_agent.ini   host=ha-nnode


      This allows that in the event that the active node goes down,
the passive l3-agent can be started, and it will take all virtual routers
associated to the first node. Otherwise (if they had different host 
identifications), the virtual routers would stick scheduled to the 
first network node they were scheduled to... as, when an l3-agent
first connects to neutron server, it gets routers scheduled to itself.


Thinking of active/active

      One interesting feature, regarding this, could be automatically
rescheduling virtual routers from down l3-agents to other available l3-agents
which have the same external network connectivity capabilities. But ok,
that's content for a blueprint, and I'm sure that has many situations to
address: for example, we wouldn't want several nodes running the same
virtual-router configurations at the same time without VRRP or equivalents...,
which in an active/passive setup like the one explained above, can be
handled with fencing/power cycling the failing node.

   That kind of behavior is already handled externally by this tool:
https://github.com/stackforge/cookbook-openstack-network/blob/master/files/default/neutron-ha-tool.py



Cheers,
    Miguel Ángel Ajo