您现在的位置是:首页 >技术教程 >Life of a Packet in Kubernetes - Calico网络进阶(注解版)网站首页技术教程

Life of a Packet in Kubernetes - Calico网络进阶(注解版)

MyySophia 2024-06-17 11:27:59
简介Life of a Packet in Kubernetes - Calico网络进阶(注解版)

目录

Topics — Part 2

CNI Requirements

BIRD (BGP)

ConfD

Felix

Routing Modes

IP-in-IP (Default)

NoEncapMode

VXLAN

Demo — IPIP and UnEncapMode

Demo — VXLAN

Disclaimer

References


As we discussed in Part 1, CNI plugins play an essential role in Kubernetes networking. There are many third-party CNI plugins available today; Calico is one of them. Many engineers prefer Calico; one of the main reasons is its ease of use and how it shapes the network fabric.

Calico supports a broad range of platforms, including Kubernetes, OpenShift, Docker EE, OpenStack, and bare metal services. The Calico node runs in a Docker container on the Kubernetes master node and on each Kubernetes worker node in the cluster. The calico-cni plugin integrates directly with the Kubernetes kubelet process on each node to discover which Pods are created and add them to Calico networking.

We will talk about installation, Calico modules (Felix, BIRD, and Confd), and routing modes.

What is not covered? Network policy — It needs a separate article, therefore skipping that for now.

Topics — Part 2

  1. Requirements
  2. Modules and its functions
  3. Routing modes
  4. Installation (calico and calicoctl)

CNI Requirements

  1. Create veth-pair and move the same inside container
  2. Identify the right POD CIDR
  3. Create a CNI configuration file
  4. Assign and manage IP address
  5. Add default routes inside the container
  6. Advertise the routes to all the peer nodes (Not applicable for VxLan)
  7. Add routes in the HOST server
  8. Enforce Network Policy

There are many other requirements too, but the above ones are the basic. Let’s take a look at the routing table in the Master and Worker node. Each node has a container with an IP address and default container route.

Basic Kubernetes network requirement

By seeing the routing table, it is evident(明显的) that the Pods can talk to each other via the L3 network as the routes are perfect. What module is responsible for adding this route, and how it gets to know the remote routes? Also, why there is a default route with gateway 169.254.1.1? We will talk about that in a moment.

  • What module is responsible for adding this route 

是哪个组件添加这条路由?

  •  how it gets to know the remote routes?

如何获取远程主机的路由?

  • why there is a default route with gateway 169.254.1.1?

pod的默认gateway是169.254.1.1

calico要解决的就是几个问题.

The core components of Calico are Bird, Felix, ConfD, Etcd, and Kubernetes API Server. The data-store is used to store the config information(ip-pools, endpoints info, network policies, etc.). In our example, we will use Kubernetes as a Calico data store.

BIRD (BGP)

The bird is a per-node BGP daemon that exchanges route information with BGP daemons running on other nodes. The common topology could be node-to-node mesh, where each BGP peers with every other.

For large scale deployments, this can get messy. There are Route Reflectors for completing the route propagation (Certain BGP nodes can be configured as Route Reflectors) to reduce the number of BGP-BGP connections. Rather than each BGP system having to peer with every other BGP system with the AS, each BGP speaker instead peers with a router reflector. Routing advertisements sent to the route reflector are then reflected out to all of the other BGP speakers. For more information, please refer to the RFC4456.

The BIRD instance is responsible for propagating the routes to other BIRD instances. The default configuration is ‘BGP Mesh,’ and this can be used for small deployments. In large-scale deployments, it is recommended to use a Route reflector to avoid issues. There can be more than one RR to have high availability. Also, external rack RRs can be used instead of BIRD.

ConfD

ConfD is a simple configuration management tool that runs in the Calico node container. It reads values (BIRD configuration for Calico) from etcd, and writes them to disk files. It loops through pools (networks and subnetworks) to apply configuration data (CIDR keys), and assembles them in a way that BIRD can use. So whenever there is a change in the network, BIRD can detect and propagate routes to other nodes.

Felix

The Calico Felix daemon runs in the Calico node container and brings the solution together by taking several actions:

  • Reads information from the Kubernetes etcd
  • Builds the routing table
  • Configures the IPTables (kube-proxy mode IPTables)
  • Configures IPVS (kube-proxy mode IPVS)

Let’s look at the cluster with all Calico modules,

Deployment with ‘NoSchedule’ Toleration

Something looks different? Yes, the one end of the veth is dangling(悬空的), not connected anywhere; It is in kernel space. veth一端悬空,它位于内核空间不连接到任何地方.

How the packet gets routed to the peer node?

  1. Pod in master tries to ping the IP address 10.0.2.11
  2. Pod sends an ARP request to the gateway.
  3. Get’s the ARP response with the MAC address.
  4. Wait, who sent the ARP response?

What’s going on? How can a container route at an IP that doesn't exist? 

Let’s walk through what’s happening. Some of you reading this might have noticed that 169.254.1.1 is an IPv4 link-local address.

The container has a default route pointing at a link-local address. The container expects this IP address to be reachable on its directly connected interface, in this case, the containers eth0 address. The container will attempt to ARP for that IP address when it wants to route out through the default route.

If we capture the ARP response, it will show the MAC address of the other end of the veth (cali123  这个mac是全e). So you might be wondering how on earth the host is replying to an ARP request for which it doesn’t have an IP interface.(主机究竟是如何响应它没有IP接口的ARP请求的?)

The answer is proxy-arp. If we check the host side VETH interface, we’ll see that proxy-arp is enabled.

master $ cat /proc/sys/net/ipv4/conf/cali123/proxy_arp
1

“Proxy ARP is a technique by which a proxy device on a given network answers the ARP queries for an IP address that is not on that network. The proxy is aware of the location of the traffic’s destination, and offers its own MAC address as the (ostensibly final) destination.[1] The traffic directed to the proxy address is then typically routed by the proxy to the intended destination via another interface or via a tunnel. The process, which results in the node responding with its own MAC address to an ARP request for a different IP address for proxying purposes, is sometimes referred to as publishing”

Let’s take a closer look(细看) at the worker node,

Once the packet reaches the kernel, it routes the packet based on routing table entries.

Incoming traffic

  1. The packet reaches the worker node kernel.
  2. Kernel puts the packet into the cali123.

Routing Modes

Calico supports 3 routing modes; in this section, we will see the pros and cons of each method and where we can use them.

  • IP-in-IP: default; encapsulated
  • Direct/NoEncapMode: unencapsulated (Preferred)
  • VXLAN: encapsulated (No BGP)

IP-in-IP (Default)

IP-in-IP is a simple form of encapsulation achieved by putting an IP packet inside another. A transmitted packet contains an outer header with host source and destination IPs and an inner header with pod source and destination IPs.

inner  header: source Pod ip -> target pod IP

outer header: source node IP of source Pod -> target node IP of target Pod

Azure doesn’t support IP-IP (As far I know); therefore, we can’t use IP-IP in that environment. It’s better to disable IP-IP to get better performance.

NoEncapMode

In this mode, send packets as if they came directly from the pod. Since there is no encapsulation and de-capsulation overhead, direct is highly performant.

Source IP check must be disabled in AWS to use this mode.

VXLAN

VXLAN routing is supported in Calico 3.7+.

VXLAN stands for Virtual Extensible LAN. VXLAN is an encapsulation technique in which layer 2 ethernet frames are encapsulated in UDP packets(二层UDP封包模式,和flanel类似). VXLAN is a network virtualization technology. When devices communicate within a software-defined Datacenter, a VXLAN tunnel is set up between those devices. Those tunnels can be set up on both physical and virtual switches. The switch ports are known as VXLAN Tunnel Endpoints (VTEPs) and are responsible for the encapsulation and de-encapsulation of VXLAN packets(VTEPs 就是一个负责解封包). Devices without VXLAN support are connected to a switch with VTEP functionality. The switch will provide the conversion from and to VXLAN.

VXLAN is great for networks that do not support IP-in-IP, such as Azure or any other DC that doesn’t support BGP.

Demo — IPIP and UnEncapMode

Check the cluster state before the Calico installation.

master $ kubectl get nodes
NAME           STATUS     ROLES    AGE   VERSION
controlplane   NotReady   master   40s   v1.18.0
node01         NotReady   <none>   9s    v1.18.0

master $ kubectl get pods --all-namespaces
NAMESPACE     NAME                                   READY   STATUS    RESTARTS   AGE
kube-system   coredns-66bff467f8-52tkd               0/1     Pending   0          32s
kube-system   coredns-66bff467f8-g5gjb               0/1     Pending   0          32s
kube-system   etcd-controlplane                      1/1     Running   0          34s
kube-system   kube-apiserver-controlplane            1/1     Running   0          34s
kube-system   kube-controller-manager-controlplane   1/1     Running   0          34s
kube-system   kube-proxy-b2j4x                       1/1     Running   0          13s
kube-system   kube-proxy-s46lv                       1/1     Running   0          32s
kube-system   kube-scheduler-controlplane            1/1     Running   0          33s

Check the CNI bin and conf directory. There won’t be any configuration file or the calico binary as the calico installation would populate these via volume mount.

master $ cd /etc/cni
-bash: cd: /etc/cni: No such file or directorymaster $ cd /opt/cni/bin
master $ ls
bridge  dhcp  flannel  host-device  host-local  ipvlan  loopback  macvlan  portmap  ptp  sample  tuning  vlan

Check the IP routes in the master/worker node.

master $ ip route
default via 172.17.0.1 dev ens3
172.17.0.0/16 dev ens3 proto kernel scope link src 172.17.0.32
172.18.0.0/24 dev docker0 proto kernel scope link src 172.18.0.1 linkdown
curl https://docs.projectcalico.org/manifests/calico.yaml -O

Download and apply the calico.yaml based on your environment.

curl https://docs.projectcalico.org/manifests/calico.yaml -O
kubectl apply -f calico.yaml

Let’s take a look at some useful configuration parameters,

cni_network_config: |-
    {
      "name": "k8s-pod-network",
      "cniVersion": "0.3.1",
      "plugins": [
        {
          "type": "calico", >>> Calico's CNI plugin
          "log_level": "info",
          "log_file_path": "/var/log/calico/cni/cni.log",
          "datastore_type": "kubernetes",
          "nodename": "__KUBERNETES_NODE_NAME__",
          "mtu": __CNI_MTU__,
          "ipam": {
              "type": "calico-ipam" >>> Calico's IPAM instaed of default IPAM
          },
          "policy": {
              "type": "k8s"
          },
          "kubernetes": {
              "kubeconfig": "__KUBECONFIG_FILEPATH__"
          }
        },
        {
          "type": "portmap",
          "snat": true,
          "capabilities": {"portMappings": true}
        },
        {
          "type": "bandwidth",
          "capabilities": {"bandwidth": true}
        }
      ]
    }# Enable IPIP
- name: CALICO_IPV4POOL_IPIP
    value: "Always" >> Set this to 'Never' to disable IP-IP
# Enable or Disable VXLAN on the default IP pool.
- name: CALICO_IPV4POOL_VXLAN
    value: "Never"

Check POD and Node status after the calico installation.

master $ kubectl get pods --all-namespaces
NAMESPACE     NAME                                       READY   STATUS              RESTARTS   AGE
kube-system   calico-kube-controllers-799fb94867-6qj77   0/1     ContainerCreating   0          21s
kube-system   calico-node-bzttq                          0/1     PodInitializing     0          21s
kube-system   calico-node-r6bwj                          0/1     PodInitializing     0          21s
kube-system   coredns-66bff467f8-52tkd                   0/1     Pending             0          7m5s
kube-system   coredns-66bff467f8-g5gjb                   0/1     ContainerCreating   0          7m5s
kube-system   etcd-controlplane                          1/1     Running             0          7m7s
kube-system   kube-apiserver-controlplane                1/1     Running             0          7m7s
kube-system   kube-controller-manager-controlplane       1/1     Running             0          7m7s
kube-system   kube-proxy-b2j4x                           1/1     Running             0          6m46s
kube-system   kube-proxy-s46lv                           1/1     Running             0          7m5s
kube-system   kube-scheduler-controlplane                1/1     Running             0          7m6smaster $ kubectl get nodes
NAME           STATUS   ROLES    AGE     VERSION
controlplane   Ready    master   7m30s   v1.18.0
node01         Ready    <none>   6m59s   v1.18.0

Explore the CNI configuration as that’s what Kubelet needs to set up the network.

master $ cd /etc/cni/net.d/
master $ ls
10-calico.conflist  calico-kubeconfig
master $
master $
master $ cat 10-calico.conflist
{
  "name": "k8s-pod-network",
  "cniVersion": "0.3.1",
  "plugins": [
    {
      "type": "calico",
      "log_level": "info",
      "log_file_path": "/var/log/calico/cni/cni.log",
      "datastore_type": "kubernetes",
      "nodename": "controlplane",
      "mtu": 1440,
      "ipam": {
          "type": "calico-ipam"
      },
      "policy": {
          "type": "k8s"
      },
      "kubernetes": {
          "kubeconfig": "/etc/cni/net.d/calico-kubeconfig"
      }
    },
    {
      "type": "portmap",
      "snat": true,
      "capabilities": {"portMappings": true}
    },
    {
      "type": "bandwidth",
      "capabilities": {"bandwidth": true}
    }
  ]
}

Check the CNI binary files,

master $ ls
bandwidth  bridge  calico  calico-ipam dhcp  flannel  host-device  host-local  install  ipvlan  loopback  macvlan  portmap  ptp  sample  tuning  vlan
master $

Let’s install the calicoctl to give good information about the calico and let us modify the Calico configuration.

master $ cd /usr/local/bin/
master $ curl -O -L  https://github.com/projectcalico/calicoctl/releases/download/v3.16.3/calicoctl
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100   633  100   633    0     0   3087      0 --:--:-- --:--:-- --:--:--  3087
100 38.4M  100 38.4M    0     0  5072k      0  0:00:07  0:00:07 --:--:-- 4325k
master $ chmod +x calicoctl
master $ export DATASTORE_TYPE=kubernetes
master $ export KUBECONFIG=~/.kube/config# Check endpoints - it will be empty as we have't deployed any POD
master $ calicoctl get workloadendpoints
WORKLOAD   NODE   NETWORKS   INTERFACEmaster $

Check BGP peer status. This will show the ‘worker’ node as a peer.

master $ calicoctl node status
Calico process is running.IPv4 BGP status
+--------------+-------------------+-------+----------+-------------+
| PEER ADDRESS |     PEER TYPE     | STATE |  SINCE   |    INFO     |
+--------------+-------------------+-------+----------+-------------+
| 172.17.0.40  | node-to-node mesh | up    | 00:24:04 | Established |
+--------------+-------------------+-------+----------+-------------+

Create a busybox POD with two replicas and master node toleration.

cat > busybox.yaml <<"EOF"
apiVersion: apps/v1
kind: Deployment
metadata:
  name: busybox-deployment
spec:
  selector:
    matchLabels:
      app: busybox
  replicas: 2
  template:
    metadata:
      labels:
        app: busybox
    spec:
      tolerations:
      - key: "node-role.kubernetes.io/master"
        operator: "Exists"
        effect: "NoSchedule"
      containers:
      - name: busybox
        image: busybox
        command: ["sleep"]
        args: ["10000"]
EOFmaster $ kubectl apply -f busybox.yaml
deployment.apps/busybox-deployment created

Get Pod and endpoint status,

master $ kubectl get pods -o wide
NAME                                 READY   STATUS    RESTARTS   AGE   IP                NODE           NOMINATED NODE   READINESS GATES
busybox-deployment-8c7dc8548-btnkv   1/1     Running   0          6s    192.168.196.131   node01         <none>           <none>
busybox-deployment-8c7dc8548-x6ljh   1/1     Running   0          6s    192.168.49.66     controlplane   <none>           <none>master $ calicoctl get workloadendpoints
WORKLOAD                             NODE           NETWORKS             INTERFACE
busybox-deployment-8c7dc8548-btnkv   node01         192.168.196.131/32   calib673e730d42
busybox-deployment-8c7dc8548-x6ljh   controlplane   192.168.49.66/32     cali9861acf9f07

Get the details of the host side veth peer of master node busybox POD.

master $ ifconfig cali9861acf9f07
cali9861acf9f07: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 1440
        inet6 fe80::ecee:eeff:feee:eeee  prefixlen 64  scopeid 0x20<link>
        ether ee:ee:ee:ee:ee:ee  txqueuelen 0  (Ethernet)
        RX packets 0  bytes 0 (0.0 B)
        RX errors 0  dropped 0  overruns 0  frame 0
        TX packets 5  bytes 446 (446.0 B)
        TX errors 0  dropped 0 overruns 0  carrier 0  collisions 0

Get the details of the master Pod’s interface,

master $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- ifconfig
eth0      Link encap:Ethernet  HWaddr 92:7E:C4:15:B9:82
          inet addr:192.168.49.66  Bcast:192.168.49.66  Mask:255.255.255.255
          UP BROADCAST RUNNING MULTICAST  MTU:1440  Metric:1
          RX packets:5 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:0
          RX bytes:446 (446.0 B)  TX bytes:0 (0.0 B)lo        Link encap:Local Loopback
          inet addr:127.0.0.1  Mask:255.0.0.0
          UP LOOPBACK RUNNING  MTU:65536  Metric:1
          RX packets:0 errors:0 dropped:0 overruns:0 frame:0
          TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:0 (0.0 B)  TX bytes:0 (0.0 B)master $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- ip route
default via 169.254.1.1 dev eth0
169.254.1.1 dev eth0 scope link
master $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- arp
master $

Get the master node routes,

master $ ip route
default via 172.17.0.1 dev ens3
172.17.0.0/16 dev ens3 proto kernel scope link src 172.17.0.32
172.18.0.0/24 dev docker0 proto kernel scope link src 172.18.0.1 linkdown
blackhole 192.168.49.64/26 proto bird
192.168.49.65 dev calic22dbe57533 scope link
192.168.49.66 dev cali9861acf9f07 scope link
192.168.196.128/26 via 172.17.0.40 dev tunl0 proto bird onlink

Let’s try to ping the worker node Pod to trigger ARP.

master $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- ping 192.168.196.131 -c 1
PING 192.168.196.131 (192.168.196.131): 56 data bytes
64 bytes from 192.168.196.131: seq=0 ttl=62 time=0.823 

msmaster $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- arp
? (169.254.1.1) at ee:ee:ee:ee:ee:ee [ether]  on eth0

The MAC address of the gateway is nothing but the cali9861acf9f07. From now, whenever the traffic goes out, it will directly hit the kernel; And, the kernel knows that it has to write the packet into the tunl0 based on the IP route.

Proxy ARP configuration,

master $ cat /proc/sys/net/ipv4/conf/cali9861acf9f07/proxy_arp
1

How the destination node handles the packet?

node01 $ ip route
default via 172.17.0.1 dev ens3
172.17.0.0/16 dev ens3 proto kernel scope link src 172.17.0.40
172.18.0.0/24 dev docker0 proto kernel scope link src 172.18.0.1 linkdown
192.168.49.64/26 via 172.17.0.32 dev tunl0 proto bird onlink
blackhole 192.168.196.128/26 proto bird
192.168.196.129 dev calid4f00d97cb5 scope link
192.168.196.130 dev cali257578b48b6 scope link
192.168.196.131 dev calib673e730d42 scope link

Upon receiving the packet, the kernel sends the right veth based on the routing table.

We can see the IP-IP protocol on the wire if we capture the packets. Azure doesn’t support IP-IP (As far I know); therefore, we can’t use IP-IP in that environment. It’s better to disable IP-IP to get better performance. Let’s try to disable and see what’s the effect.

Disable IP-IP

Update the ipPool configuration to disable IPIP.

master $ calicoctl get ippool default-ipv4-ippool -o yaml > ippool.yaml
master $ vi ippool.yaml

Open the ippool.yaml and set the IPIP to ‘Never,’ and apply the yaml via calicoctl.

master $ calicoctl apply -f ippool.yaml
Successfully applied 1 'IPPool' resource(s)

Recheck the IP route,

master $ ip route
default via 172.17.0.1 dev ens3
172.17.0.0/16 dev ens3 proto kernel scope link src 172.17.0.32
172.18.0.0/24 dev docker0 proto kernel scope link src 172.18.0.1 linkdown
blackhole 192.168.49.64/26 proto bird
192.168.49.65 dev calic22dbe57533 scope link
192.168.49.66 dev cali9861acf9f07 scope link
192.168.196.128/26 via 172.17.0.40 dev ens3 proto bird

The device is no more tunl0; it is set to the management interface of the master node.

Let’s ping the worker node POD and make sure all works fine. From now, there won’t be any IPIP protocol involved.

master $ kubectl exec busybox-deployment-8c7dc8548-x6ljh -- ping 192.168.196.131 -c 1
PING 192.168.196.131 (192.168.196.131): 56 data bytes
64 bytes from 192.168.196.131: seq=0 ttl=62 time=0.653 ms--- 192.168.196.131 ping statistics ---
1 packets transmitted, 1 packets received, 0% packet loss
round-trip min/avg/max = 0.653/0.653/0.653 ms

Note: Source IP check should be disabled in AWS environment to use this mode.

Demo — VXLAN

Re-initiate the cluster and download the calico.yaml file to apply the following changes,

  1. Remove bird from livenessProbe and readinessProbe
livenessProbe:
            exec:
              command:
              - /bin/calico-node
              - -felix-live
              - -bird-live >> Remove this
            periodSeconds: 10
            initialDelaySeconds: 10
            failureThreshold: 6
          readinessProbe:
            exec:
              command:
              - /bin/calico-node
              - -felix-ready
              - -bird-ready >> Remove this

2. Change the calico_backend to ‘vxlan’ as we don’t need BGP anymore.

kind: ConfigMap
apiVersion: v1
metadata:
  name: calico-config
  namespace: kube-system
data:
  # Typha is disabled.
  typha_service_name: "none"
  # Configure the backend to use.
  calico_backend: "vxlan"

3. Disable IPIP

# Enable IPIP
- name: CALICO_IPV4POOL_IPIP
    value: "Never" >> Set this to 'Never' to disable IP-IP
# Enable or Disable VXLAN on the default IP pool.
- name: CALICO_IPV4POOL_VXLAN
    value: "Never"

Let’s apply this new yaml.

master $ ip route
default via 172.17.0.1 dev ens3
172.17.0.0/16 dev ens3 proto kernel scope link src 172.17.0.15
172.18.0.0/24 dev docker0 proto kernel scope link src 172.18.0.1 linkdown
192.168.49.65 dev calif5cc38277c7 scope link
192.168.49.66 dev cali840c047460a scope link
192.168.196.128/26 via 192.168.196.128 dev vxlan.calico onlinkvxlan.calico: flags=4163<UP,BROADCAST,RUNNING,MULTICAST>  mtu 1440
        inet 192.168.196.128  netmask 255.255.255.255  broadcast 192.168.196.128
        inet6 fe80::64aa:99ff:fe2f:dc24  prefixlen 64  scopeid 0x20<link>
        ether 66:aa:99:2f:dc:24  txqueuelen 0  (Ethernet)
        RX packets 0  bytes 0 (0.0 B)
        RX errors 0  dropped 0  overruns 0  frame 0
        TX packets 0  bytes 0 (0.0 B)
        TX errors 0  dropped 11 overruns 0  carrier 0  collisions 0

Get the POD status,

master $ kubectl get pods -o wide
NAME                                 READY   STATUS    RESTARTS   AGE   IP                NODE           NOMINATED NODE   READINESS GATES
busybox-deployment-8c7dc8548-8bxnw   1/1     Running   0          11s   192.168.49.67     controlplane   <none>           <none>
busybox-deployment-8c7dc8548-kmxst   1/1     Running   0          11s   192.168.196.130   node01         <none>           <none>

Ping the worker node POD from

master $ kubectl exec busybox-deployment-8c7dc8548-8bxnw -- ip route
default via 169.254.1.1 dev eth0
169.254.1.1 dev eth0 scope link

Trigger the ARP request,

master $ kubectl exec busybox-deployment-8c7dc8548-8bxnw -- arp
master $ kubectl exec busybox-deployment-8c7dc8548-8bxnw -- ping 8.8.8.8
PING 8.8.8.8 (8.8.8.8): 56 data bytes
64 bytes from 8.8.8.8: seq=0 ttl=116 time=3.786 ms
^C
master $ kubectl exec busybox-deployment-8c7dc8548-8bxnw -- arp
? (169.254.1.1) at ee:ee:ee:ee:ee:ee [ether]  on eth0
master $

The concept is as the previous modes, but the only difference is that the packet reaches the vxland, and it encapsulates the packet with node IP and its MAC in the inner header and sends it. Also, the UDP port of the vxlan proto will be 4789. The etcd helps here to get the details of available nodes and their supported IP range so that the vxlan-calico can build the packet.

Note: VxLAN mode needs more processing power than the previous modes.

Disclaimer

This article does not provide any technical advice or recommendation; if you feel so, it is my personal view, not the company I work for.

References

Calico Documentation | Calico Documentation
Open Infrastructure Summit videos from past community events, featuring keynotes and sessions from the global network of developers, operators, and supporting organizations.
Kubernetes
IBM Documentation
GitHub - flannel-io/flannel: flannel is a network fabric for containers, designed for Kubernetes
风语者!平时喜欢研究各种技术,目前在从事后端开发工作,热爱生活、热爱工作。