Skip to content

Latest commit

 

History

History
317 lines (265 loc) · 13.2 KB

README.md

File metadata and controls

317 lines (265 loc) · 13.2 KB

1. Building connection between two hosts

Suppose you have two host, Host-Alice(ip:192.168.1.10), Host-Bob(ip:192.168.1.11)

1.1 install etcd, ovs on Host-Alice, and run the etcd and ovs

  • suppose the etcd listening port is 2379, and the etcd should support v3 api
  • you should use a ovs above(include) version 2.8.0
  • each tuplenet node should install and start ovs, and the ovsdb must has has system-id
    • you can use ovs-vsctl get open_vswitch . external-ids:system-id to see if the ovsdb has system-id
    • you can use ovs-vsctl set open_vswitch . external-ids:system-id=${your-host-system-id}

1.2 create logical switch and logical router(you can create them on any host which had been installed tuplenet)

tpctl ls add LS-A   **(you may need to specified the etcd  e.g. tpctl --endpoints 10.199.132.54:2379  ls add LS-A)**
tpctl ls add LS-B
tpctl lr add LR-central
tpctl lsp add LS-A lsp-alice 10.20.1.10 f2:01:00:00:00:01
tpctl lsp add LS-B lsp-bob 10.20.2.10 f2:01:00:00:00:02
tpctl lr link LR-central LS-A 10.20.1.1/24
tpctl lr link LR-central LS-B 10.20.2.1/24

you can use tpctl lsp show LS-A or tpctl lrp show to see what you had configured. The above commands try to build a virtual networking like

             +------------------------+
             |                        |
        +----+       LR-central       +-----+
        |    |                        |     |
        |    +------------------------+     |
        |                                   |
  +----------+                        +----------+
  |          |                        |          |
  |   LS-A   |                        |   LS-B   |
  |          |                        |          |
  +----------+                        +----------+

1.3 install tuplenet on both Host-Alice and Host-Bob

pip install tuplenet
running tuplenet on each hosts by using command: tuplenet --interface eth0 --host 192.168.1.10:2379

1.4 create namespaces on Host-Alice and Host-Bob on Host-Alice:

ip netns add ns-alice
ip link add nseth0 type veth peer name nseth0-peer
ip link set nseth0 netns ns-alice
ovs-vsctl add-port br-int nseth0-peer -- set Interface  nseth0-peer external-ids:iface-id=lsp-alice
ip netns exec ns-alice ip addr add 10.20.1.10/24 dev nseth0
ip netns exec ns-alice ip link set dev nseth0 address f2:01:00:00:00:01
ip netns exec ns-alice ip link set dev nseth0 up
ip netns exec ns-alice ip route add default via 10.20.1.1
ifconfig nseth0-peer up

on Host-Bob:

ip netns add ns-bob
ip link add nseth0 type veth peer name nseth0-peer
ip link set nseth0 netns ns-bob
ovs-vsctl add-port br-int nseth0-peer -- set Interface  nseth0-peer external-ids:iface-id=lsp-bob
ip netns exec ns-bob ip addr add 10.20.2.10/24 dev nseth0
ip netns exec ns-bob ip link set dev nseth0 address f2:01:00:00:00:02
ip netns exec ns-bob ip link set dev nseth0 up
ip netns exec ns-bob ip route add default via 10.20.2.1
ifconfig nseth0-peer up

1.5 now try to ping namespace !

on Host-Alice: ip netns exec ns-alice ping 10.20.2.10

on Host-Bob: ip netns exec ns-bob ping 10.20.1.10

2. Building connection between virtual networking and physical networking

suppose you had already build connection between Host-Alice and Host-Bob which mention above. And now you have a another host name host-edge1

2.1 install tuplenet on host-edge1 and run the tuplenet as a gateway

pip install tuplenet

we should add a nic(which link to physical networking) into a ovs bridge, if you have two nics(eth1 is the nic links to the physical networking, eth0 rx/tx the tunnel frames)

ovs-vsctl add-br br0
ovs-vsctl add-port br0 eth1 (eth1 is the nic links to the physical networking, eth0 rx/tx the tunnel frames)
ONDEMAND=0 GATEWAY=1 tuplenet --interface eth0 --host 192.168.1.10:2379  (ONDEMAND=0 means this tuplenet instance generate all ovs-flow, GATEWAY=1 means the tuplenet should generate some specified ovs-flows which work on gateway node only)

if you only have one nic(eth0 link to physical networking and eth0 was used to rx/tx tunnel frames)

ovs-vsctl add-br br0
ifconfig br0 up; ifconfig br0 \${eth0_ip}; ovs-vsctl add-port br0 eth0; ifconfig eth0 0; route add default gw \${default_gw}
ONDEMAND=0 GATEWAY=1 tuplenet --interface br0 --host 192.168.1.10:2379  (ONDEMAND=0 means this tuplenet instance generate all ovs-flow, GATEWAY=1 means the tuplenet should generate some specified ovs-flows which work on gateway node only)

2.2 adding new virtual networking device

python tools/edge-operate.py --endpoint=192.168.1.10:2379 --op=init --vip=192.168.1.20/24 --virt=10.20.0.0/16 --ext_gw=192.168.1.1 --phy_br=br0  (execute this cmd on a edge node. 192.168.1.1 is the physical gateway. vip is an ip assgin to the edge1, physical switch can ping this vip to detect if the edge node is alive or not. So please give a vip which can be reach in physcial networking)
tpctl lnat add tp_LR_edge1 snat_rule1 10.20.0.0/16 snat 192.168.1.30   (create snat on tp_LR_edge1)

              +-----------------+
              |    outside1     |
              +-----------------+
                       |
                       |
                 +------------+
                 |            |
                 |   edge1    |
                 +------------+
                        |
                        |
                   +---------+
                   |   m1    |
                   +---------+
                        |
                        |
            +------------------------+
            |                        |
       +----+       LR-central       +-----+
       |    |                        |     |
       |    +------------------------+     |
       |                                   |
 +----------+                        +----------+
 |          |                        |          |
 |   LS-A   |                        |   LS-B   |
 |          |                        |          |
 +----------+                        +----------+

2.3 config mtu and try to ping/wget

on Host-Alice: ip netns exec ns-alice ip link set dev nseth0 mtu 1400 (if the physical networking mtu is 1500, the mtu of endpoints in overlay networking should minus 100, tuplenet utilizes Geneve to construct frames)
on Host-Bob: ip netns exec ns-bob ip link set dev nseth0 mtu 1400

Now you can try to ping each other

ip netns exec ns-alice ping 192.168.X.X
ip netns exec ns-alice wget X.X.X.X

3. ECMP and HA-HA edge

tuplenet support ecmp(tuplenet node deliver traffic to edge node in ecmp way), tuplenet node consume bfd to detect status of tuplenet edge nodes and deliver to other edge node once it found no edge node is dead.

3.1 install tuplenet on host-edge2 and run the tuplenet as a gateway

  • reference the commands in 2.1

3.2 adding new virtual edge node

python tools/edge-operate.py --endpoint=192.168.1.10:2379 --op=add --vip=192.168.1.21/24 --phy_br=br0  (it search the current logical network-view and try to create a new ECMP path to ex-network. edge2 was assigned an ip 192.168.1.21)
tpctl lnat add tp_LR_edge2 snat_rule2 10.20.0.0/16 snat 192.168.1.31  (create snat on tp_LR_edge2)

      +-----------------+            +-----------------+
      |    outside1     |            |    outside2     |
      +-----------------+            +-----------------+
               |                              |
               |                              |
         +------------+                 +------------+
         |            |                 |            |
         |   edge1    |                 |   edge2    |
         +------------+                 +------------+
                |                              |
                |                              |
           +---------+                    +---------+
           |   m1    |                    |   m2    |
           +---------+                    +---------+
                |                              |
                +------------------------------+
                              |
                  +------------------------+
                  |                        |
             +----+       LR-central       +-----+
             |    |                        |     |
             |    +------------------------+     |
             |                                   |
       +----------+                        +----------+
       |          |                        |          |
       |   LS-A   |                        |   LS-B   |
       |          |                        |          |
       +----------+                        +----------+


4. CNM in tuplenet

tuplenet implement cnm interface which can be consumed by docker

run tpcnm, it communicates with docker, tuplenet etcd and ovs side

  • create a config.json file
  • input essential info into config.json: e.g
{
  "etcd_cluster": "192.168.5.50:2379,192.168.5.51:2379,192.168.5.53:2379",
  "data_store_prefix": "/tuplenet",
  "docker_unix_sock": "/var/run/docker.sock",
  "egress_router_name": "LR-central"
}

data_store_prefix: it is the default tuplenet's etcd prefix datapath egress_route_name: it is a optional parameter, if it was set then it means

suppose all docker in tuplenet node were connected to a single cluster.

4.1 run tpcnm

tpcnm -config /tmp/config.json

NOTE: please exec mkdir -p /run/docker/plugins/ if it show err msg "listen unix /run/docker/plugins/tuplenet.sock: bind: no such file or directory"

4.2 create tuplenet network

- docker network create -d tuplenet --subnet=10.20.3.0/24 --gateway=10.20.3.1  tp-docker-net1 (this networking is global you just need to create it once. tp-docker-net1 will link to LR-central if config.json has ***"egress_router_name": "LR-central"*** )

NOTE: the name of tp-docker-net1 in tuplenet's etcd may be a just a uuid.

NOTE: please confirm whether your docker has connected to a cluster before running this command to create a global networking

4.3 create container link to tp-docker-net1

docker run  --privileged --net=tp-docker-net1 -ti -d centos-tool /bin/bash (tpcnm will create lsp on switch tp-docker-net1 automatically )

NOTE: please decrease the mtu of container's ethX as well, the mtu of endpoints in overlay networking should minus 100, tuplenet utilizes Geneve to construct frames.

   +-----------------+            +-----------------+
   |    outside1     |            |    outside2     |
   +-----------------+            +-----------------+
            |                              |
            |                              |
      +------------+                 +------------+
      |            |                 |            |
      |   edge1    |                 |   edge2    |
      +------------+                 +------------+
             |                              |
             |                              |
        +---------+                    +---------+
        |   m1    |                    |   m2    |
        +---------+                    +---------+
             |                              |
             +------------------------------+
                           |
               +------------------------+
               |                        |
          +----+       LR+central       +-----+
          |    |                        |     |
          |    +------------------------+     |
          |                 |                 |
    +----------+            |           +----------+
    |          |            |           |          |
    |   LS_A   |            |           |   LS_B   |
    |          |            |           |          |
    +----------+            |           +----------+
                            |
                  +-------------------+
                  |  tp-docker-net1   |
                  |                   |
                  +-------------------+

5. CNI in TupleNet

TupleNet implement CNI interface which can be consumed by kubelet

5.1 config tpcni.conf

cat <<EOF  > /etc/cni/net.d/tpcni.conf
{
        "cniVersion": "0.3.0",
        "name": "tpcni-network",
        "type": "tpcni",
        "mtu": 1400,
        "switchname": "LS_A",
        "subnet": "10.20.1.1/24",
        "etcd_cluster": "YOUR_ETCD_IP:PORT",
        "data_store_prefix": "/tuplenet"
}
EOF

This config file tell tpcni that it should allocate ip from 10.20.1.1/24, and the created lsp was pinned on LS-A.(the IP 10.20.1.1 is default gw)

5.1 build link for tpcni

Once tpcni.conf was built the kubelet searchs tpcni in /opt/cni/bin. User must set link for tpcni in /opt/cni/bin

ln -s /usr/bin/tpcni /opt/cni/bin/tpcni

Now please try to restart kubelet and create a new POD.

6. Metrics and monitoring

6.1 how to enable IPFIX in tuplenet node

Append IPFIX_COLLECTOR=IP:port and optional IPFIX_SAMPLING_RATE=x to environment variable. Domain ID will be the uint32 representation of the tuplenet node IP

7. Enable UNTUNNEL mode

Once enable untunnel mode, the regular tuplenet's out-traffic(should forward to physical network) would not deliver to edge node. Instead, those traffic will be deliver to host tcpip stack through br-int port which is an internal port. It helps to improve latency and throughput.

Append ENABLE_UNTUNNEL=1 to environment variable, restart tuplenet then it enable the untunnel mode.