rpc-on-ironic

# rpc-on-ironic network notes.


# Generate the networks for your cluster.

The ./scripts/gen-onironic-nets.py generates a yaml var file to use for the network setup.
The vxlan ids at the top of each section is randomly generated.  The vxlan_group for 
multicast traffic is also randomly generated. Here is what it looks like:

  vxlan_group: 239.51.50.107

  mgmt_vxlan: 3095351
  mgmt_network: 172.22.0.0/20
  mgmt_gateway: 172.22.0.1
  mgmt_netmask: 255.255.240.0

  storage_vxlan: 3095352
  storage_network: 172.22.16.0/20
  storage_gateway: 172.22.16.1
  storage_netmask: 255.255.240.0

  flat_vxlan: 3095353
  flat_network: 172.22.32.0/20
  flat_gateway: 172.22.32.1
  flat_netmask: 255.255.240.0

  vlan_vxlan: 3095354
  vlan_network: 172.22.48.0/20
  vlan_gateway: 172.22.48.1
  vlan_netmask: 255.255.240.0

  tunnel_vxlan: 3095355
  tunnel_network: 172.22.64.0/20
  tunnel_gateway: 172.22.64.1
  tunnel_netmask: 255.255.240.0

  repl_vxlan: 3095356
  repl_network: 172.22.80.0/20
  repl_gateway: 172.22.80.1
  repl_netmask: 255.255.240.0


- The mgmt(or container) network is used for most controle plane communications.
- The storage network is for contact between hypervisors and the storage devices.
- The flat network is going to be use for the public/gateway neutron network. It will be masqueraded through the deploy box.
- The vlan network is used for neutron 'vlan' type networks.  
- The tunnel network is used for project private networks(vxlan)
- The repl network is used for swift/ceph cluster communications(if needed)



# Configure the networks for your cluster

This is done through the prep-onironic-network.yml playlbook which uses the roles/rpc-on-ironic-netconf role.

For each network(mgmt, storage, flat, vlan, tunnel), it uses the templates/vxlan_interfaces/debian-<network>.cfg.j2
template to create an /etc/network/interfaces.d/<network>.cfg file.  The layout before osa or rpco will end up as below.

eno49(Physical Nic) =>
eno50(Physical Nic) => bond0(bond) => mgmt-mesh(vxlan int)    => br-mgmt(linux bridge with a mgmt net ip)
                                   => flat-mesh(vxlan int)    => br-flat(linux bridge with a flat net ip)
                                   => storage-mesh(vxlan int) => br-storage(linux bridge with a storage net ip)
                                   => tunnel-mesh(vxlan int)  => br-tunnel(linux bridge with a tunnel net ip)
                                   => vlan-mesh(vxlan int)    => br-vlan(linux bridge with a vlan net ip)
 





# OSA changes to network:


The scripts/scripts/create_rpc_configs.sh sets up the openstack_user_config.yml that defines the
provider networks.  The changes after may look like:


eno49(Physical Nic) =>
eno50(Physical Nic) => bond0(bond) => mgmt-mesh(vxlan int)    => br-mgmt(linux bridge with a mgmt net ip)         => veth pair => eth1 on all containers

                                   => flat-mesh(vxlan int)    => br-flat(linux bridge with a flat net ip)         => veth pair => eth12 on neutron containers & compute (flat provider type) 
                                                              => neutron created bridge using flat-mesh           => veth pairs => vms using flat privider.
                                                                 Note: Used for flat network types as neutron can't use bridges with flat.(set with host_bind_override)

                                   => storage-mesh(vxlan int) => br-storage(linux bridge with a storage net ip)   => veth pair => eth2 on glance_api, cinder_api, cinder_volume and nova_compute containers

                                   => tunnel-mesh(vxlan int)  => br-tunnel(linux bridge with a tunnel net ip)     => veth pair => eth10 on neutron containers & compute (vxlan provider type range 1:1000)
                                                                                                                  => veth pairs => nics on vms using private vxlan networks

                                   => vlan-mesh(vxlan int)    => br-vlan(linux bridge with a vlan net ip)         => veth pair => eth11 on neutron containers & compute (vlan provider type range 100:200)
                                                                                                                  => veth pairs => nics on vms using public/private vlan networks



# Deploy Box

- Should have all of the same networks as the ironic devices.  The bond is replaced by the single primary virtual nic. 
- ip forwarding set up
- An iptables masquerade is set up. It takes anything from the flat network, not going to the flat network and masquerades it externally with the deploy box ip.
- The flat network gateway ip address(172.22.32.1) is assigned as an additional ip to the br-flat bridge.