==================================== Multi-pod Installation with DevStack ==================================== Introduction ^^^^^^^^^^^^ In the single pod installation guide, we discuss how to deploy the Tricircle in one single pod with DevStack. Besides the Tricircle API and the central Neutron server, only one pod(one pod means one OpenStack instance) is running. Network is created with the default network type: local. Local type network will be only presented in one pod. If a local type network is already hosting virtual machines in one pod, you can not use it to boot virtual machine in another pod. That is to say, local type network doesn't support cross-Neutron l2 networking. With multi-pod installation of the Tricircle, you can try out cross-Neutron l2 networking and cross-Neutron l3 networking features. To support cross-Neutron l2 networking, we have added both VLAN and VxLAN network type to the Tricircle. When a VLAN type network created via the central Neutron server is used to boot virtual machines in different pods, local Neutron server in each pod will create a VLAN type network with the same VLAN ID and physical network as the central network, so each pod should be configured with the same VLAN allocation pool and physical network. Then virtual machines in different pods can communicate with each other in the same physical network with the same VLAN tag. Similarly, for VxLAN network type, each pod should be configured with the same VxLAN allocation pool, so local Neutron server in each pod can create a VxLAN type network with the same VxLAN ID as is allocated by the central Neutron server. Cross-Neutron l3 networking is supported in two ways in the Tricircle. If two networks connected to the router are of local type, we utilize a shared VLAN or VxLAN network to achieve cross-Neutron l3 networking. When a subnet is attached to a router via the central Neutron server, the Tricircle not only creates corresponding subnet and router in the pod, but also creates a "bridge" network. Both tenant network and "bridge" network are attached to the router. Each tenant will have one allocated VLAN or VxLAN ID, which is shared by the tenant's "bridge" networks across Neutron servers. The CIDRs of "bridge" networks for one tenant are also the same, so the router interfaces in "bridge" networks across different Neutron servers can communicate with each other. By adding an extra route as following:: destination: CIDR of tenant network in another pod nexthop: "bridge" network interface ip in another pod When a virtual machine sends a packet whose receiver is in another network and in another pod, the packet first goes to router, then is forwarded to the router in another pod according to the extra route, at last the packet is sent to the target virtual machine. This route configuration job is triggered when user attaches a subnet to a router via the central Neutron server and the job is finished asynchronously. If one of the network connected to the router is not local type, meaning that cross-Neutron l2 networking is supported in this network(like VLAN type), and the l2 network can be stretched into current pod, packets sent to the virtual machine in this network will not pass through the "bridge" network. Instead, packets first go to router, then are directly forwarded to the target virtual machine via the l2 network. A l2 network's presence scope is determined by the network's availability zone hint. If the l2 network is not able to be stretched into the current pod, the packets will still pass through the "bridge network". For example, let's say we have two pods, pod1 and pod2, and two availability zones, az1 and az2. Pod1 belongs to az1 and pod2 belongs to az2. If the availability zone hint of one VLAN type network is set to az1, this network can not be stretched to pod2. So packets sent from pod2 to virtual machines in this network still need to pass through the "bridge network". Prerequisite ^^^^^^^^^^^^ In this guide we take two nodes deployment as an example. One node to run the Tricircle API, the central Neutron server and one pod, the other one node to run another pod. For VLAN network, both nodes should have two network interfaces, which are connected to the management network and provider VLAN network. The physical network infrastructure should support VLAN tagging. For VxLAN network, you can combine the management plane and data plane, in this case, only one network interface is needed. If you would like to try north-south networking, too, you should prepare one more network interface in the second node for the external network. In this guide, the external network is also VLAN type, so the local.conf sample is based on VLAN type external network setup. For the resource requirements to setup each node, please refer to `All-In-One Single Machine `_ for installing DevStack in bare metal server and `All-In-One Single VM `_ for installing DevStack in virtual machine. If you want to experience cross Neutron VxLAN network, please make sure compute nodes are routable to each other on data plane, and enable L2 population mechanism driver in OpenStack RegionOne and OpenStack RegionTwo. Setup ^^^^^ In pod1 in node1 for Tricircle service, central Neutron and OpenStack RegionOne, - 1 Install DevStack. Please refer to `DevStack document `_ on how to install DevStack into single VM or bare metal server. - 2 In DevStack folder, create a file local.conf, and copy the content of `local.conf node1 sample `_ to local.conf, change password in the file if needed. - 3 Change the following options according to your environment - change HOST_IP to your management interface ip:: HOST_IP=10.250.201.24 - the format of Q_ML2_PLUGIN_VLAN_TYPE_OPTIONS is (network_vlan_ranges=::), you can change physical network name, but remember to adapt your change to the commands showed in this guide; also, change min VLAN and max vlan to adapt the VLAN range your physical network supports. You need to additionally specify the physical network "extern" to ensure the central neutron can create "extern" physical network which located in other pods:: Q_ML2_PLUGIN_VLAN_TYPE_OPTIONS=(network_vlan_ranges=bridge:2001:3000,extern:3001:4000) - if you would like to also configure vxlan network, you can set Q_ML2_PLUGIN_VXLAN_TYPE_OPTIONS. the format of it is (vni_ranges=:):: Q_ML2_PLUGIN_VXLAN_TYPE_OPTIONS=(vni_ranges=1001:2000) - the format of OVS_BRIDGE_MAPPINGS is :, you can change these names, but remember to adapt your change to the commands showed in this guide. You do not need specify the bridge mapping for "extern", because this physical network is located in other pods:: OVS_BRIDGE_MAPPINGS=bridge:br-vlan this option can be omitted if only VxLAN networks are needed - if you would like to also configure flat network, you can set Q_ML2_PLUGIN_FLAT_TYPE_OPTIONS, the format of it is (flat_networks=phy_net1,phy_net2,...). Besides specifying a list of physical network names, you can also use '*' to allow flat networks with arbitrary physical network names; or use an empty list to disable flat networks. For simplicity, we use the same physical networks and bridge mappings for vlan and flat network configuration. Similar to vlan network, You need to additionally specify the physical network "extern" to ensure the central neutron can create "extern" physical network which located in other pods:: Q_ML2_PLUGIN_FLAT_TYPE_OPTIONS=(flat_networks=bridge,extern) - set TRICIRCLE_START_SERVICES to True to install the Tricircle service and central Neutron in node1:: TRICIRCLE_START_SERVICES=True - 4 Create OVS bridge and attach the VLAN network interface to it :: sudo ovs-vsctl add-br br-vlan sudo ovs-vsctl add-port br-vlan eth1 br-vlan is the OVS bridge name you configure on OVS_PHYSICAL_BRIDGE, eth1 is the device name of your VLAN network interface, this step can be omitted if only VxLAN networks are provided to tenants. - 5 Run DevStack. In DevStack folder, run :: ./stack.sh - 6 After DevStack successfully starts, begin to setup node2. In pod2 in node2 for OpenStack RegionTwo, - 1 Install DevStack. Please refer to `DevStack document `_ on how to install DevStack into single VM or bare metal server. - 2 In DevStack folder, create a file local.conf, and copy the content of `local.conf node2 sample `_ to local.conf, change password in the file if needed. - 3 Change the following options according to your environment - change HOST_IP to your management interface ip:: HOST_IP=10.250.201.25 - change KEYSTONE_SERVICE_HOST to management interface ip of node1:: KEYSTONE_SERVICE_HOST=10.250.201.24 - change KEYSTONE_AUTH_HOST to management interface ip of node1:: KEYSTONE_AUTH_HOST=10.250.201.24 - the format of Q_ML2_PLUGIN_VLAN_TYPE_OPTIONS is (network_vlan_ranges=::), you can change physical network name, but remember to adapt your change to the commands showed in this guide; also, change min vlan and max vlan to adapt the vlan range your physical network supports:: Q_ML2_PLUGIN_VLAN_TYPE_OPTIONS=(network_vlan_ranges=bridge:2001:3000,extern:3001:4000) - if you would like to also configure vxlan network, you can set Q_ML2_PLUGIN_VXLAN_TYPE_OPTIONS. the format of it is (vni_ranges=:):: Q_ML2_PLUGIN_VXLAN_TYPE_OPTIONS=(vni_ranges=1001:2000) - the format of OVS_BRIDGE_MAPPINGS is :, you can change these names, but remember to adapt your change to the commands showed in this guide:: OVS_BRIDGE_MAPPINGS=bridge:br-vlan,extern:br-ext if you only use vlan network for external network, it can be configured like:: OVS_BRIDGE_MAPPINGS=extern:br-ext - if you would like to also configure flat network, you can set Q_ML2_PLUGIN_FLAT_TYPE_OPTIONS, the format of it is (flat_networks=phy_net1,phy_net2,...). Besides specifying a list of physical network names, you can also use '*' to allow flat networks with arbitrary physical network names; or use an empty list to disable flat networks. For simplicity, we use the same physical networks and bridge mappings for vlan and flat network configuration:: Q_ML2_PLUGIN_FLAT_TYPE_OPTIONS=(flat_networks=bridge,extern) - set TRICIRCLE_START_SERVICES to False(it's True by default) so Tricircle services and central Neutron will not be started in node2:: TRICIRCLE_START_SERVICES=False In this guide, we define two physical networks in node2, one is "bridge" for bridge network, the other one is "extern" for external network. If you do not want to try l3 north-south networking, you can simply remove the "extern" part. The external network type we use in the guide is VLAN, if you want to use other network type like flat, please refer to `DevStack document `_. - 4 Create OVS bridge and attach the VLAN network interface to it :: sudo ovs-vsctl add-br br-vlan sudo ovs-vsctl add-port br-vlan eth1 sudo ovs-vsctl add-br br-ext sudo ovs-vsctl add-port br-ext eth2 br-vlan and br-ext are the OVS bridge names you configure on OVS_PHYSICAL_BRIDGE, eth1 and eth2 are the device names of your VLAN network interfaces, for the "bridge" network and the external network. Omit br-vlan if you only use vxlan network as tenant network. - 5 Run DevStack. In DevStack folder, run :: ./stack.sh - 6 After DevStack successfully starts, the setup is finished. .. note:: In the newest version of codes, we may fail to boot an instance in node2. The reason is that Apache configuration file of Nova placement API doesn't grant access right to the placement API bin folder. You can use "screen -r" to check placement API is working well or not. If placement API is in stuck status, manually update "/etc/apache2/sites-enabled/placement-api.conf" placement API configuration file in node2 to add the following section:: Require all granted After update, restart Apache service first, and then placement API. **This problem no longer exists after this patch:** https://github.com/openstack-dev/devstack/commit/6ed53156b6198e69d59d1cf3a3497e96f5b7a870 How to play ^^^^^^^^^^^ - 1 After DevStack successfully starts, we need to create environment variables for the user (admin user as example in this guide). In DevStack folder :: source openrc admin demo - 2 Unset the region name environment variable, so that the command can be issued to specified region in following commands as needed :: unset OS_REGION_NAME - 3 Check if services have been correctly registered. Run :: openstack --os-region-name=RegionOne endpoint list you should get output looks like as following :: +----------------------------------+---------------+--------------+----------------+ | ID | Region | Service Name | Service Type | +----------------------------------+---------------+--------------+----------------+ | 4adaab1426d94959be46314b4bd277c2 | RegionOne | glance | image | | 5314a11d168042ed85a1f32d40030b31 | RegionTwo | nova_legacy | compute_legacy | | ea43c53a8ab7493dacc4db079525c9b1 | RegionOne | keystone | identity | | a1f263473edf4749853150178be1328d | RegionOne | neutron | network | | ebea16ec07d94ed2b5356fb0a2a3223d | RegionTwo | neutron | network | | 8d374672c09845f297755117ec868e11 | CentralRegion | tricircle | Tricircle | | e62e543bb9cf45f593641b2d00d72700 | RegionOne | nova_legacy | compute_legacy | | 540bdedfc449403b9befef3c2bfe3510 | RegionOne | nova | compute | | d533429712954b29b9f37debb4f07605 | RegionTwo | glance | image | | c8bdae9506cd443995ee3c89e811fb45 | CentralRegion | neutron | network | | 991d304dfcc14ccf8de4f00271fbfa22 | RegionTwo | nova | compute | +----------------------------------+---------------+--------------+----------------+ "CentralRegion" is the region you set in local.conf via CENTRAL_REGION_NAME, whose default value is "CentralRegion", we use it as the region for the Tricircle API and central Neutron server. "RegionOne" and "RegionTwo" are the normal OpenStack regions which includes Nova, Neutron and Glance. Shared Keystone service is registered in "RegionOne". - 4 Create pod instances for the Tricircle to manage the mapping between availability zones and OpenStack instances :: openstack multiregion networking pod create --region-name CentralRegion openstack multiregion networking pod create --region-name RegionOne --availability-zone az1 openstack multiregion networking pod create --region-name RegionTwo --availability-zone az2 Pay attention to "region_name" parameter we specify when creating pod. Pod name should exactly match the region name registered in Keystone. In the above commands, we create pods named "CentralRegion", "RegionOne" and "RegionTwo". - 5 Create necessary resources in central Neutron server :: neutron --os-region-name=CentralRegion net-create --availability-zone-hint RegionOne net1 neutron --os-region-name=CentralRegion subnet-create net1 10.0.1.0/24 neutron --os-region-name=CentralRegion net-create --availability-zone-hint RegionTwo net2 neutron --os-region-name=CentralRegion subnet-create net2 10.0.2.0/24 Please note that the net1 and net2 ID will be used in later step to boot VM. - 6 Get image ID and flavor ID which will be used in VM booting :: glance --os-region-name=RegionOne image-list nova --os-region-name=RegionOne flavor-list glance --os-region-name=RegionTwo image-list nova --os-region-name=RegionTwo flavor-list - 7 Boot virtual machines :: nova --os-region-name=RegionOne boot --flavor 1 --image $image1_id --nic net-id=$net1_id vm1 nova --os-region-name=RegionTwo boot --flavor 1 --image $image2_id --nic net-id=$net2_id vm2 - 8 Verify the VMs are connected to the networks :: neutron --os-region-name=CentralRegion port-list neutron --os-region-name=RegionOne port-list nova --os-region-name=RegionOne list neutron --os-region-name=RegionTwo port-list nova --os-region-name=RegionTwo list The ip address of each VM could be found in local Neutron server and central Neutron server. The port has same uuid in local Neutron server and central Neutron Server. - 9 Create external network and subnet :: neutron --os-region-name=CentralRegion net-create --router:external --provider:network_type vlan --provider:physical_network extern --availability-zone-hint RegionTwo ext-net neutron --os-region-name=CentralRegion subnet-create --name ext-subnet --disable-dhcp ext-net 163.3.124.0/24 Pay attention that when creating external network, we need to pass "availability_zone_hints" parameter, which is the name of the pod that will host external network. *Currently external network needs to be created before attaching subnet to the router, because plugin needs to utilize external network information to setup bridge network when handling interface adding operation. This limitation will be removed later.* - 10 Create router and attach subnets in central Neutron server :: neutron --os-region-name=CentralRegion router-create router neutron --os-region-name=CentralRegion router-interface-add router $subnet1_id neutron --os-region-name=CentralRegion router-interface-add router $subnet2_id - 11 Set router external gateway in central Neutron server :: neutron --os-region-name=CentralRegion router-gateway-set router ext-net Now virtual machine in the subnet attached to the router should be able to ping machines in the external network. In our test, we use hypervisor tool to directly start a virtual machine in the external network to check the network connectivity. - 12 Launch VNC console and test connection :: nova --os-region-name=RegionOne get-vnc-console vm1 novnc nova --os-region-name=RegionTwo get-vnc-console vm2 novnc You should be able to ping vm1 from vm2 and vice versa. - 13 Create floating ip in central Neutron server :: neutron --os-region-name=CentralRegion floatingip-create ext-net - 14 Associate floating ip :: neutron --os-region-name=CentralRegion floatingip-list neutron --os-region-name=CentralRegion port-list neutron --os-region-name=CentralRegion floatingip-associate $floatingip_id $port_id Now you should be able to access virtual machine with floating ip bound from the external network.