Re: [openstack-dev] [fuel] NodeGroups vs network-templates and static routes

2015-07-20 Thread Sergey Vasilenko
On Thu, Jul 16, 2015 at 10:53 PM, Andrew Woodward xar...@gmail.com wrote:

 Regardless of computing all the routes, we need to compute same role, but
 multi-segement routes. In this case I see that nodegroups becomes
 redundant. It's only value is that it may be a simpler interface then
 templates but it imposes the old network topology which I could see people
 wanting to get away from.


I do not agree with unnecessary node groups.
Yes, move route calculation is a good idea and I think we should implement
it.

But removing node groups... When we will implement multi-rack feature we
will need some abstraction for store rack-specific attributes. Node groups
are seems appropriate for this role.

/sv
__
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


Re: [openstack-dev] [fuel] NodeGroups vs network-templates and static routes

2015-07-20 Thread Andrew Woodward
Ya, that makes sense now that you mention it. We will still need node
groups to act as partitioning for the rack values

On Mon, Jul 20, 2015 at 1:44 AM Sergey Vasilenko svasile...@mirantis.com
wrote:


 On Thu, Jul 16, 2015 at 10:53 PM, Andrew Woodward xar...@gmail.com
 wrote:

 Regardless of computing all the routes, we need to compute same role, but
 multi-segement routes. In this case I see that nodegroups becomes
 redundant. It's only value is that it may be a simpler interface then
 templates but it imposes the old network topology which I could see people
 wanting to get away from.


 I do not agree with unnecessary node groups.
 Yes, move route calculation is a good idea and I think we should implement
 it.

 But removing node groups... When we will implement multi-rack feature we
 will need some abstraction for store rack-specific attributes. Node groups
 are seems appropriate for this role.

 /sv
  __
 OpenStack Development Mailing List (not for usage questions)
 Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
 http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

-- 

--

Andrew Woodward

Mirantis

Fuel Community Ambassador

Ceph Community
__
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev


[openstack-dev] [fuel] NodeGroups vs network-templates and static routes

2015-07-16 Thread Andrew Woodward
In 6.0 we added nodegroups as part of the multiple-cluster-networks
features. With these you can add additional sets of networks with so that
the nodes can exist on different network segments. When these are used you
will also need to set the gateway for each of your networks. When you do
this, you get routes set up between the matching network names across the
nodegroups.

For example network.yaml that looks like (shortened)

networks:
- cidr: 172.16.0.0/24
  gateway: 172.16.0.1
  group_id: 2
  id: 6
- cidr: 172.16.10.0/24
  gateway: 172.16.10.1
  group_id: 3
  id: 9

Will result in mappings like this in a nodes yaml (in nodegroup 2)

network_scheme:
  endpoints:
br-ex:
  IP:
  - 172.16.0.4/24
  routes:
  - net: 172.16.10.0/24
via: 172.16.0.1


With the introduction of templates we may no longer need nodegroups. They
served two functions. 1) They allowed us to create additional networks. 2)
They created additional routes between networks of the same name. Comparing
with what is in templates, #1 is taken care of, but what about #2? I think
that we need the routes configured anyway. Nodes with the same network role
should have a route for it when it crosses network segments.

This would have traditionally been separated by nodegroups. but it now can
be coded with templates. In this case (such as the yaml above) we must have
routes for the nodes to communicate on the correct interface. Since we need
code for routes between segments of the same network role, it might behoove
ourselves to compute (maybe not use when they are the local interface).
This serves two functions, it allows us to visualize the routing topology
instead of just relying on the default route. Secondly when we get to using
a routing protocol it gives us the data necessary to validate the routing
protocol with what we expected.

Regardless of computing all the routes, we need to compute same role, but
multi-segement routes. In this case I see that nodegroups becomes
redundant. It's only value is that it may be a simpler interface then
templates but it imposes the old network topology which I could see people
wanting to get away from.
-- 

--

Andrew Woodward

Mirantis

Fuel Community Ambassador

Ceph Community
__
OpenStack Development Mailing List (not for usage questions)
Unsubscribe: openstack-dev-requ...@lists.openstack.org?subject:unsubscribe
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev