# Running on Alibaba Cloud This page primarily explains how to use Spiderpool in an Alibaba Cloud environment and how to implement a complete Underlay solution. ## Background With a multitude of public cloud providers available, such as Alibaba Cloud, Huawei Cloud, Tencent Cloud, AWS, and more, it can be challenging to use mainstream open-source CNI plugins to operate on these platforms using Underlay networks. Instead, one has to rely on proprietary CNI plugins provided by each cloud vendor, leading to a lack of standardized Underlay solutions for public clouds. This page introduces Spiderpool, an Underlay networking solution designed to work seamlessly in any public cloud environment. A unified CNI solution offers easier management across multiple clouds, particularly in hybrid cloud scenarios. ## Features Spiderpool offers features such as node topology, MAC address validation, and integration with services that use `spec.externalTrafficPolicy` set to Local mode. Spiderpool can run on public cloud environments using the IPVlan Underlay CNI. Here's an overview of its implementation: 1. When using Underlay networks in a public cloud environment, each network interface of a cloud server can only be assigned a limited number of IP addresses. To enable communication when an application runs on a specific cloud server, it needs to obtain the valid IP addresses allocated to different network interfaces within the VPC network. To address this IP allocation requirement, Spiderpool introduces a CRD named `SpiderIPPool`. By configuring the nodeName and multusName fields in `SpiderIPPool`, it enables node topology functionality. Spiderpool leverages the affinity between the IP pool and nodes, as well as the affinity between the IP pool and IPvlan Multus, facilitating the utilization and management of available IP addresses on the nodes. This ensures that applications are assigned valid IP addresses, enabling seamless communication within the VPC network, including communication between Pods and also between Pods and cloud servers. 2. In a public cloud VPC network, network security controls and packet forwarding principles dictate that when network data packets contain MAC and IP addresses unknown to the VPC network, correct forwarding becomes unattainable. This issue arises in scenarios where Macvlan or OVS based Underlay CNI plugins generate new MAC addresses for Pod NICs, resulting in communication failures among Pods. To address this challenge, Spiderpool offers a solution in conjunction with [IPVlan CNI](https://www.cni.dev/plugins/current/main/ipvlan/). The IPVlan CNI operates at the L3 of the network, eliminating the reliance on L2 broadcasts and avoiding the generation of new MAC addresses. Instead, it maintains consistency with the parent interface. By incorporating IPVlan, the legitimacy of MAC addresses in a public cloud environment can be effectively resolved. 3. When the `.spec.externalTrafficPolicy` of a service is set to `Local`, the client's source IP can be reserved. However, in self-managed public cloud clusters, using the platform's LoadBalancer component for nodePort forwarding in this mode will lead to access failures. To tackle this problem, Spiderpool offers the coordinator plugin. This plugin utilizes iptables to apply packet marking, ensuring that response packets coming from veth0 are still forwarded through veth0. This resolves the problem of nodePort access in this mode. ## Prerequisites 1. The system kernel version must be greater than 4.2 when using IPVlan as the cluster's CNI. 2. [Helm](https://helm.sh/docs/intro/install/) is installed. ## Steps ### Alibaba Cloud Environment - Prepare an Alibaba Cloud environment with virtual machines that have 2 network interfaces. Assign a set of auxiliary private IP addresses to each network interface, as shown in the picture: ![alicloud-web-network](https://docs.daocloud.io/daocloud-docs-images/docs/en/docs/network/images/alicloud-network-web.png) - Utilize the configured VMs to build a Kubernetes cluster. The available IP addresses for the nodes and the network topology of the cluster are depicted below: ![网络拓扑](https://docs.daocloud.io/daocloud-docs-images/docs/en/docs/network/images/alicloud-k8s-network.png) ### Install Spiderpool Install Spiderpool via helm: ```bash helm repo add spiderpool https://spidernet-io.github.io/spiderpool helm repo update spiderpool helm install spiderpool spiderpool/spiderpool --namespace kube-system --set ipam.enableStatefulSet=false --set multus.multusCNI.defaultCniCRName="ipvlan-eth0" ``` > If you are using a cloud server from a Chinese mainland cloud provider, you can enhance image pulling speed by specifying the parameter `--set global.imageRegistryOverride=ghcr.m.daocloud.io`. > > Spiderpool allows for fixed IP addresses for application replicas with a controller type of `StatefulSet`. However, in the underlay network scenario of public clouds, cloud instances are limited to using specific IP addresses. When StatefulSet replicas migrate to different nodes, the original fixed IP becomes invalid and unavailable on the new node, causing network unavailability for the new Pods. To address this issue, set `ipam.enableStatefulSet` to `false` to disable this feature. > > Specify the Multus clusterNetwork for the cluster using `multus.multusCNI.defaultCniCRName`. `clusterNetwork` is a specific field within the Multus plugin used to define the default network interface for Pods. ### Install CNI To simplify the creation of JSON-formatted Multus CNI configurations, Spiderpool offers the SpiderMultusConfig CR to automatically manage Multus NetworkAttachmentDefinition CRs. Here is an example of creating an IPvlan SpiderMultusConfig configuration: ```shell IPVLAN_MASTER_INTERFACE0="eth0" IPVLAN_MULTUS_NAME0="ipvlan-$IPVLAN_MASTER_INTERFACE0" IPVLAN_MASTER_INTERFACE1="eth1" IPVLAN_MULTUS_NAME1="ipvlan-$IPVLAN_MASTER_INTERFACE1" cat < test-app-1-b7765b8d8-qjgpj 1/1 Running 0 16s 172.31.199.193 worker test-app-2-7c56876fc6-7brhf 1/1 Running 0 12s 192.168.0.160 master test-app-2-7c56876fc6-zlxxt 1/1 Running 0 12s 192.168.0.161 worker ``` Spiderpool automatically assigns IP addresses to the applications, ensuring that the assigned IPs are within the expected IP pool. ```bash $ kubectl get spiderippool NAME VERSION SUBNET ALLOCATED-IP-COUNT TOTAL-IP-COUNT DEFAULT master-172 4 172.31.192.0/20 1 5 true master-192 4 192.168.0.0/24 1 5 true worker-172 4 172.31.192.0/20 1 5 true worker-192 4 192.168.0.0/24 1 5 true ``` ### Test East-West Connectivity - Test communication between Pods and their hosts: ```bash $ kubectl get nodes -owide NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME master Ready control-plane 2d12h v1.27.3 172.31.199.183 CentOS Linux 7 (Core) 6.4.0-1.el7.elrepo.x86_64 containerd://1.7.1 worker Ready 2d12h v1.27.3 172.31.199.184 CentOS Linux 7 (Core) 6.4.0-1.el7.elrepo.x86_64 containerd://1.7.1 $ kubectl exec -ti test-app-1-b7765b8d8-422sb -- ping 172.31.199.183 -c 2 PING 172.31.199.183 (172.31.199.183): 56 data bytes 64 bytes from 172.31.199.183: seq=0 ttl=64 time=0.088 ms 64 bytes from 172.31.199.183: seq=1 ttl=64 time=0.054 ms --- 172.31.199.183 ping statistics --- 2 packets transmitted, 2 packets received, 0% packet loss round-trip min/avg/max = 0.054/0.071/0.088 ms ``` - Test communication between Pods across different nodes and subnets: ```shell $ kubectl exec -ti test-app-1-b7765b8d8-422sb -- ping 172.31.199.193 -c 2 PING 172.31.199.193 (172.31.199.193): 56 data bytes 64 bytes from 172.31.199.193: seq=0 ttl=64 time=0.460 ms 64 bytes from 172.31.199.193: seq=1 ttl=64 time=0.210 ms --- 172.31.199.193 ping statistics --- 2 packets transmitted, 2 packets received, 0% packet loss round-trip min/avg/max = 0.210/0.335/0.460 ms $ kubectl exec -ti test-app-1-b7765b8d8-422sb -- ping 192.168.0.161 -c 2 PING 192.168.0.161 (192.168.0.161): 56 data bytes 64 bytes from 192.168.0.161: seq=0 ttl=64 time=0.408 ms 64 bytes from 192.168.0.161: seq=1 ttl=64 time=0.194 ms --- 192.168.0.161 ping statistics --- 2 packets transmitted, 2 packets received, 0% packet loss round-trip min/avg/max = 0.194/0.301/0.408 ms ``` - Test communication between Pods and ClusterIP services: ```bash $ kubectl get svc test-svc NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE test-svc ClusterIP 10.233.23.194 80/TCP 26s $ kubectl exec -ti test-app-2-7c56876fc6-7brhf -- curl 10.233.23.194 -I HTTP/1.1 200 OK Server: nginx/1.10.1 Date: Fri, 21 Jul 2023 06:45:56 GMT Content-Type: text/html Content-Length: 4086 Last-Modified: Fri, 21 Jul 2023 06:38:41 GMT Connection: keep-alive ETag: "64ba27f1-ff6" Accept-Ranges: bytes ``` ### Test North-South Connectivity #### Test egress traffic from Pods to external destinations - Alibaba Cloud's NAT Gateway provides an ingress and egress gateway for public or private network traffic within a VPC environment. By utilizing NAT Gateway, the cluster can have egress connectivity. Please refer to [the NAT Gateway documentation](https://www.alibabacloud.com/help/en/nat-gateway?spm=a2c63.p38356.0.0.1b111b76Rn9rPa) for creating a NAT Gateway as depicted in the picture: ![alicloud-natgateway](https://docs.daocloud.io/daocloud-docs-images/docs/en/docs/network/images/alicloud-natgateway.png) - Test egress traffic from Pods ```bash $ kubectl exec -ti test-app-2-7c56876fc6-7brhf -- curl www.baidu.com -I HTTP/1.1 200 OK Accept-Ranges: bytes Cache-Control: private, no-cache, no-store, proxy-revalidate, no-transform Connection: keep-alive Content-Length: 277 Content-Type: text/html Date: Fri, 21 Jul 2023 08:42:17 GMT Etag: "575e1f60-115" Last-Modified: Mon, 13 Jun 2016 02:50:08 GMT Pragma: no-cache Server: bfe/1.0.8.18 ``` #### Load Balancer Traffic Ingress Access ##### Deploy Cloud Controller Manager Cloud Controller Manager (CCM) is an Alibaba Cloud's component that enables integration between Kubernetes and Alibaba Cloud services. We will use CCM along with Alibaba Cloud infrastructure to facilitate load balancer traffic ingress access. Follow the steps below and refer to [the CCM documentation](https://github.com/kubernetes/cloud-provider-alibaba-cloud/blob/master/docs/getting-started.md) for deploying CCM. 1. Configure `providerID` on Cluster Nodes On each node in the cluster, run the following command to obtain the `providerID` for each node. is the API entry point provided by Alibaba Cloud CLI for retrieving instance metadata. You don't need to modify it in the provided example. For more information, please refer to [ECS instance metadata](https://www.alibabacloud.com/help/en/ecs/user-guide/overview-of-ecs-instance-metadata?spm=a2c63.p38356.0.0.1c3c592dPUwXMN). ```bash $ META_EP=http://100.100.100.200/latest/meta-data $ provider_id=`curl -s $META_EP/region-id`.`curl -s $META_EP/instance-id` $ echo $provider_id cn-hangzhou.i-bp17345hor9******* ``` On the `master` node of the cluster, use the `kubectl patch` command to add the `providerID` for each node in the cluster. This step is necessary to ensure the proper functioning of the CCM Pod on each corresponding node. Failure to run this step will result in the CCM Pod being unable to run correctly. ```bash kubectl get nodes # Replace and with proper values kubectl patch node ${NODE_NAME} -p '{"spec":{"providerID": "${provider_id}"}}' ``` 2. Create an Alibaba Cloud RAM user and grant authorization. A RAM user is an entity within Alibaba Cloud's Resource Access Management (RAM) that represents individuals or applications requiring access to Alibaba Cloud resources. Refer to [Overview of RAM users](https://www.alibabacloud.com/help/en/ram/user-guide/overview-of-ram-users?spm=a2c63.p38356.0.0.435a7b9fxy619R) to create a RAM user and assign the necessary permissions for accessing resources. To ensure that the RAM user used in the subsequent steps has sufficient privileges, grant the `AdministratorAccess` and `AliyunSLBFullAccess` permissions to the RAM user, following the instructions provided here. 3. Obtain the AccessKey & AccessKeySecret for the RAM user. Log in to the RAM User account and go to [User Center](https://account.alibabacloud.com/login/login.htm?spm=5176.12901015-2.0.0.36cb525cXk2SG0) to retrieve the corresponding AccessKey & AccessKeySecret for the RAM User. 4. Create the Cloud ConfigMap for CCM. Use the following method to write the AccessKey & AccessKeySecret obtained in step 3 as environment variables. ```bash export ACCESS_KEY_ID=LTAI******************** export ACCESS_KEY_SECRET=HAeS************************** ``` Run the following command to create cloud-config: ```bash accessKeyIDBase64=`echo -n "$ACCESS_KEY_ID" |base64 -w 0` accessKeySecretBase64=`echo -n "$ACCESS_KEY_SECRET"|base64 -w 0` cat <>` with the actual cluster CIDR. ```bash wget https://raw.githubusercontent.com/spidernet-io/spiderpool/main/docs/example/alicloud/cloud-controller-manager.yaml kubectl apply -f cloud-controller-manager.yaml ``` 6. Verify if CCM is installed. ```bash $ kubectl get po -n kube-system | grep cloud-controller-manager NAME READY STATUS RESTARTS AGE cloud-controller-manager-72vzr 1/1 Running 0 27s cloud-controller-manager-k7jpn 1/1 Running 0 27s ``` ##### Create Load Balancer Ingress for Applications The following YAML will create two sets of services, one for TCP (layer 4 load balancing) and one for HTTP (layer 7 load balancing), with `spec.type` set to `LoadBalancer`. - `service.beta.kubernetes.io/alibaba-cloud-loadbalancer-protocol-port`: this annotation provided by CCM allows you to customize the exposed ports for layer 7 load balancing. For more information, refer to [the CCM Usage Documentation](https://github.com/kubernetes/cloud-provider-alibaba-cloud/blob/master/docs/usage.md). - `.spec.externalTrafficPolicy`: indicates whether the service prefers to route external traffic to local or cluster-wide endpoints. It has two options: Cluster (default) and Local. Setting `.spec.externalTrafficPolicy` to `Local` preserves the client source IP. ```bash $ cat <