[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Rdo-list] Attempt to reproduce https://github.com/beekhof/osp-ha-deploy/blob/master/HA-keepalived.md


cutting some html, responses inline:

Actually, using bondings and external network provider of VLAN type,
controller nodes may have 3 VLAN's : bond0.100(management network),
bond0.101 (tunnel network) and bond0.102(external-network).
Then tune haproxy with keepalived for a virtual ip setup on bond0.100

In my case, I even collapse tunnel network with the general/management/provisioning network (where my foreman smart proxy lives and provision bare metal); it’s a routed but secure network. My nodes thus have two interfaces:

-bond0 (untagged native vlan for provisioniong)
-bond0.112 (external traffic, no nodes have an IP)

My haproxys have two interfaces:

-bond0.111 (external API access, routed)

This way I isolate and/or allow traffic from instances (and the rest of the organisation) to the openstack API’s, secured by terminating SSL at the loadbalancers (see diagram)

This excellent approach, but question which is my concern
is a bit different.


On controller and compute, define the external vlan interface and its bridge:

cat <<EOF > /etc/sysconfig/network-scripts/ifcfg-bond0.102

cat <<EOF > /etc/sysconfig/network-scripts/ifcfg-br-bond0.102

then, make sure that everywhere exists /etc/neutron/plugins/ml2/openvswitch_agent.ini as:

enable_tunneling = True
tunnel_id_ranges = 1:1000
tenant_network_type = vxlan
integration_bridge = br-int
tunnel_bridge = br-tun
local_ip = <IP on bond0>
bridge_mappings = physnet1:br-bond0.102
network_vlan_ranges = physnet1
tunnel_types = vxlan
vxlan_udp_port = 4789
l2_population = False
firewall_driver = neutron.agent.linux.iptables_firewall.OVSHybridIptablesFirewallDriver

restart neutron-openvswitch-agent everywhere to make it work:

Hope it helps

I would guess , that external net was created similar to this way

controller# neutron net-create --router:external=True \
--provider:network_type=vlan --provider:segmentation_id=102  ext-network

No, external provider network is a flat network (because it uses an already-tagged interface, bond0.112)

controller# neutron subnet-create --name ext-subnet --disable-dhcp \
   --allocation-pool start,end= \

I do use neutron DHCP: some instances have only one interface in the provider network, thus they won’t be accessible if I don’t provide them an IP+metadata

of VLAN type not FLAT

It would be a VLAN-provider network if you would add bond0 interface to the bridge, and let OVS tag packets. In this case it’s “flat”, as OVS is not aware of any VLAN tagging, which happens downstream at the interface. But there’s dozens of more skilled network dudes&gals on the list that may correct me :)

Could you be so kind to share yours ml2_conf.ini on controllers in cluster.

a very simple one:

type_drivers = flat,vxlan,vlan
tenant_network_types = vxlan
mechanism_drivers = openvswitch
flat_networks = *
vni_ranges = 10:10000
vxlan_group =
enable_security_group = True
firewall_driver = neutron.agent.linux.iptables_firewall.OVSHybridIptablesFirewallDriver

I also believe that tuning your switch contains a kind of :-
switchport trunk allowed vlan 100,102,104

Indeed, switch ports are trunked.

Thank you


14 Nov 2015, at 09:35, Boris Derzhavets <bderzhavets hotmail com> wrote:

From: Dan Sneddon <dsneddon redhat com>
Sent: Friday, November 13, 2015 4:10 PM
To: Boris Derzhavets; rdo-list redhat com
Subject: Re: [Rdo-list] Attempt to reproduce https://github.com/beekhof/osp-ha-deploy/blob/master/HA-keepalived.md

On 11/13/2015 12:56 PM, Dan Sneddon wrote:
Hi Boris,

Let's keep this on-list, there may be others who are having similar
issues who could find this discussion useful.

Answers inline...

On 11/13/2015 12:17 PM, Boris Derzhavets wrote:

From: Dan Sneddon <dsneddon redhat com>
Sent: Friday, November 13, 2015 2:46 PM
To: Boris Derzhavets; Javier Pena
Cc: rdo-list redhat com
Subject: Re: [Rdo-list] Attempt to reproduce https://github.com/beekhof/osp-ha-deploy/blob/master/HA-keepalived.md

On 11/13/2015 11:38 AM, Boris Derzhavets wrote:
I understand that in usual situation , creating ifcfg-br-ex and ifcfg-eth2 ( as OVS bridge and OVS port) ,
`service network restart` should be run to make eth2 (no IP) OVS port of br-ex (any IP which belongs ext net and is available)
What bad does NetworkManager when external network provider is used ?
Disabling it,  I break routing via eth0's interfaces of cluster nodes to ( ext net),
so nothing is supposed to work :-
Either I am missing something here.
From: rdo-list-bounces redhat com <rdo-list-bounces redhat com> on behalf of Boris Derzhavets <bderzhavets hotmail com>
Sent: Friday, November 13, 2015 1:09 PM
To: Javier Pena
Cc: rdo-list redhat com
Subject: [Rdo-list] Attempt to reproduce https://github.com/beekhof/osp-ha-deploy/blob/master/HA-keepalived.md

Working on this task I was able to build 3 node HAProxy/Keepalived  Controller's cluster , create compute node , launch CirrOS VM,
However, I cannot ping floating IP of VM running on compute ( total 4 CentOS 7.1 VMs, nested kvm enabled )
Looks like provider external networks  doesn't work for me.

But , to have eth0 without IP (due to `ovs-vsctl add-port br-eth0 eth0 ) still allowing to ping,
I need NetworkManager active, rather then network.service

[root hacontroller1 network-scripts]# systemctl status NetworkManager
NetworkManager.service - Network Manager
  Loaded: loaded (/usr/lib/systemd/system/NetworkManager.service; enabled)
  Active: active (running) since Fri 2015-11-13 20:39:21 MSK; 12min ago
Main PID: 808 (NetworkManager)
  CGroup: /system.slice/NetworkManager.service
          ├─ 808 /usr/sbin/NetworkManager --no-daemon
          └─2325 /sbin/dhclient -d -q -sf /usr/libexec/nm-dhcp-helper -pf /var/run/dhclient-eth0...

Nov 13 20:39:22 hacontroller1.example.com NetworkManager[808]: <info>  NetworkManager state is n...L
Nov 13 20:39:22 hacontroller1.example.com dhclient[2325]: bound to -- renewal in 1...s.
Nov 13 20:39:22 hacontroller1.example.com NetworkManager[808]: <info>  (eth0): Activation: succe....
Nov 13 20:39:25 hacontroller1.example.com NetworkManager[808]: <info>  startup complete

[root hacontroller1 network-scripts]# systemctl status network.service
network.service - LSB: Bring up/down networking
  Loaded: loaded (/etc/rc.d/init.d/network)
  Active: inactive (dead)

[root hacontroller1 network-scripts]# cat ifcfg-eth0

[root hacontroller1 network-scripts]# ping -c 3
PING ( 56(84) bytes of data.
64 bytes from icmp_seq=1 ttl=64 time=0.087 ms
64 bytes from icmp_seq=2 ttl=64 time=0.128 ms
64 bytes from icmp_seq=3 ttl=64 time=0.117 ms

--- ping statistics ---
3 packets transmitted, 3 received, 0% packet loss, time 1999ms
rtt min/avg/max/mdev = 0.087/0.110/0.128/0.021 ms

If I disable NetworkManager and enable network this feature will be lost. Eth0 would have to have static IP or dhcp lease,
to provide route to

Thank you.

Rdo-list mailing list
Rdo-list redhat com

To unsubscribe: rdo-list-unsubscribe redhat com

OK, a few things here. First of all, you don't actually need to have an
IP address on the host system to use a VLAN or interface as an external
provider network. The Neutron router will have an IP on the right
network, and within its namespace will be able to reach the 10.10.10.x

It looks to me like NetworkManager is running dhclient for eth0, even
though you have BOOTPROTO="static". This is causing an IP address to be
added to eth0, so you are able to ping 10.10.10.x from the host. When
you turn off NetworkManager, this unexpected behavior goes away, *but
you should still be able to use provider networks*.

    Here I am quoting Lars Kellogg Stedman
    The bottom statement in blog post above states :-
    "This assumes that eth1 is connected to a network using and eth2 is connected to a network using, and that each network has a gateway sitting at the corresponding .1 address."

Right, what Lars means is that eth1 is physically connected to a
network with the subnet, and eth2 is physically connected
to a network with the subnet.

You might notice that in Lars's instructions, he never puts a host IP
on either interface.

Try creating a Neutron router with an IP on 10.10.10.x, and then you
should be able to ping that network from the router namespace.

  " When I issue `neutron router-creater --ha True --tenant-id xxxxxx RouterHA`  , i cannot specify router's
   IP "

Let me refer you to this page, which explains the basics of creating
and managing Neutron networks:


You will have to create an external network, which you will associate
with a physical network via a bridge mapping. The default bridge
mapping for br-ex is datacentre:br-ex.

Using the name of the physical network "datacentre", we can create an

1.  Javier is using external network provider ( and so  did I ,  following  him)

#.   /root/keystonerc_admin
# neutron net-create public --provider:network_type flat --provider:physical_network physnet1 --router:external
# neutron subnet-create --gateway --allocation-pool start=,end=  --disable-dhcp --name public_subnet public

HA Neutron router and tenant's subnet have been created.
Then interface to tenant's network was activated as well as gateway to public.
Security rules were implemented as usual.
Cloud VM was launched,  it obtained private IP and committed cloud-init OK.
Then I assigned FIP from public to cloud VM , it should be ping able from from F23 Visualization

2.  All  traffic to/from external network flows through br-int when provider external networks has been involved. No br-ex is needed.
When in Javier does  `ovs-vsctl add-port br-eth0 eth0` , eth0 (which is inside VM ,running  Controller node)
should be on 10.10.10.X/24. It doesn't happen when service network is active (and NM disabled) .
In this case  eth0 doesn't have any kind of IP assigned to provide route to  Libvirt's  subnet 10.10.10.X/24 ( pre created by myself)

In meantime I am under impression that ovs bridge br-eth0 and OVS port eth0 
would work when IP is assigned to port eth0, not to bridge. OVS release =>2.3.1 seems to allow that.
Tested here (VM's case ) :- http://dbaxps.blogspot.com/2015/10/multiple-external-networks-with-single.html
If neither one of br-eth0 and eth0 would have IP then packets won't be forwarded to external net

external network:

[If the external network is on VLAN 104]
neutron net-create ext-net --router:external \
--provider:physical_network datacentre \
--provider:network_type vlan \
--provider:segmentation_id 104

[If the external net is on the native VLAN (flat)]
neutron net-create ext-net --router:external \
--provider:physical_network datacentre \
--provider:network_type flat

Next, you must create a subnet for the network, including the range of
floating IPs (allocation pool):

neutron subnet-create --name ext-subnet \
--enable_dhcp=False \
--allocation-pool start=,end= \
--gateway \

Next, you have to create a router:

neutron router-create ext-router

You then add an interface to the router. Since Neutron will assign the
first address in the subnet to the router by default (, you
will want to first create a port with a specific IP, then assign that
port to the router.

neutron port-create ext-net --fixed-ip ip_address=

You will need to note the UUID of the newly created port. You can also
see this with "neutron port-list". Now, create the router interface
with the port you just created:

neutron router-interface-add ext-router port=<UUID>

If you want to be able to ping 10.10.10.x from the host, then you
should put either a static IP or DHCP on the bridge, not on eth0. This
should work whether you are running NetworkManager or network.service.

  "I do can ping 10.0.0.x from F23 KVM Server (running cluster's VMs as Controller's nodes),
    it's  just usual non-default libvirt subnet,matching exactly external network creating in Javier's  "Howto".
    It was  created via `virsh net-define openstackvms.xml`, but I cannot ping FIPs belong to
    cloud VM on this subnet."

I think you will have better luck once you create the external network
and router. You can then use namespaces to ping the network from the

First, obtain the qrouter-<UUID> from the list of namespaces:

sudo ip netns list

Then, find the qrouter-<UUID> and ping from there:

ip netns exec qrouter-XXXX-XXXX-XXX-XXX ping

One more quick thing to note:

In order to use floating IPs, you will also have to attach the external
router to the tenant networks where floating IPs will be used.

When you go through the steps to create a tenant network, also attach
it to the router:

1) Create the network:

neutron net-create tenant-net-1

2) Create the subnet:

neutron subnet-create --name tenant-subnet-1 tenant-net-1

3) Attach the external router to the network:

neutron router-interface-add tenant-router-1 subnet=tenant-subnet-1

(since no specific port was given in the router-interface-add command,
Neutron will automatically choose the first address in the given
subnet, so in this example)

Dan Sneddon         |  Principal OpenStack Engineer
dsneddon redhat com |  redhat.com/openstack
650.254.4025        |  dsneddon:irc   @dxs:twitter

Rdo-list mailing list
Rdo-list redhat com

To unsubscribe: rdo-list-unsubscribe redhat com

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]