Open Stack

Miguel, Kevin,

Thank you for your comments.

The motivation of this (from our customer) is about
design of network-node reduction procedure.

They prefer stop scheduling for agents on the node
at first then individual services are moved by
*-remove and *-add gradually rather than stopping all
services on the node at once.

As Kevin mentioned, agents must be alive so that 
operations for existing services must be continued. 

Thanks.
Itsuro Oda

On Wed, 7 Jan 2015 15:52:56 -0800
Kevin Benton <blak111 at gmail.com> wrote:

> The problem is that if you just stop the service, it not only removes it
> from scheduling, but it stops it from receiving updates to floating IP
> changes, interface changes, etc. I think it would be nice to have a way to
> explicitly stop it from being scheduled new routers, but still act as a
> functioning L3 agent otherwise.
> 
> On Wed, Jan 7, 2015 at 3:30 PM, Miguel Angel Ajo <majopela at redhat.com>
> wrote:
> 
> > You can stop the neutron-dhcp-agent or neutron-l3-agent,
> > the agents should go not-active after reporting timeout.
> >
> > The actual network services (routers, dhcp, etc) will stay
> > active into the node, but unmanaged. In some cases,
> > if you have automatic rescheduling of the resources
> > configured, those will be spawned on other hosts.
> >
> > Depending on your use case this will be enough or not.
> > It’s intended for upgrades and maintenance. But not
> > for controlling resources in a node.
> >
> > Miguel Angel Ajo
> >
> > On Thursday, 8 de January de 2015 at 00:20, Itsuro ODA wrote:
> >
> > Carl,
> >
> > Thank you for your comment.
> >
> > It seems there is no clear opinion about whether bug report or
> > buleprint is better.
> > So I submitted a bug report for the moment so that the requirememt
> > is not forgotten.
> > https://bugs.launchpad.net/neutron/+bug/1408488
> >
> > Thanks.
> > Itsuro Oda
> >
> > On Tue, 6 Jan 2015 09:05:19 -0700
> > Carl Baldwin <carl at ecbaldwin.net> wrote:
> >
> > Itsuro,
> >
> > It would be desirable to be able to be hide an agent from scheduling
> > but no one has stepped up to make this happen. Come to think of it,
> > I'm not sure that a bug or blueprint has been filed yet to address it
> > though it is something that I've wanted for a little while now.
> >
> > Carl
> >
> > On Mon, Jan 5, 2015 at 4:13 PM, Itsuro ODA <oda at valinux.co.jp> wrote:
> >
> > Neutron experts,
> >
> > I want to stop scheduling to a specific {dhcp|l3}_agent without
> > stopping router/dhcp services on it.
> > I expected setting admin_state_up of the agent to False is met
> > this demand. But this operation stops all services on the agent
> > in actuality. (Is this behavior intended ? It seems there is no
> > document for agent API.)
> >
> > I think admin_state_up of agents should affect only scheduling.
> > If it is accepted I will submit a bug report and make a fix.
> >
> > Or should I propose a blueprint for adding function to stop
> > agent's scheduling without stopping services on it ?
> >
> > I'd like to hear neutron experts' suggestions.
> >
> > Thanks.
> > Itsuro Oda
> > --
> > Itsuro ODA <oda at valinux.co.jp>
> >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >
> >
> > --
> > Itsuro ODA <oda at valinux.co.jp>
> >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >
> >
> >
> > _______________________________________________
> > OpenStack-dev mailing list
> > OpenStack-dev at lists.openstack.org
> > http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
> >
> >
> 
> 
> -- 
> Kevin Benton

-- 
Itsuro ODA <oda at valinux.co.jp>