New and Changed Information

The following table provides an overview of the significant changes up to this current release. The table does not provide an exhaustive list of all changes or of the new features up to this release.

Table 1. New Features and Changed Behavior in Cisco APIC

Cisco APIC Release Version

Feature

Description

4.1(x) Service account hardening

Every cluster is now configured with a dedicated administrative account that enables you to control read and write permissions for the whole fabric.

For more information, see the section Service Account Hardening.

4.0(1)

--

Added how to drain a node.

For more information, see the section Draining a Node.

3.1(1)

Kubernetes nesting in ESX

This release enables you to deploy OpFlex base container solutions such as Kubernetes on VMs that are attached onto an ESX VMM.

3.1(1)

--

Added hardware dependency.

For more information, see the section Hardware Requirements.

3.0(1) and later

Service subnet advertisement

You can configure Cisco Application Centric Infrastructure (ACI) to advertise service subnets externally through a dynamic routing protocol.

For more information, see the section Service Subnet Advertisement.

3.0(1)

Kubernetes on bare-metal servers

The release support integration of Kubernetes on bare-metal servers into the Cisco Application Centric Infrastructure (ACI).

Cisco ACI and Kubernetes Integration

Kubernetes is an open source system that automates the deployment, scaling, and managing containers in a network. Beginning with Cisco APIC Release 3.0(1), you can integrate Kubernetes on bare-metal servers into the Cisco Application Centric Infrastructure (ACI).

To integrate Kubernetes with the Cisco ACI, you perform a series of tasks. Some you perform in the network to set up the Cisco Application Policy Infrastructure Controller (APIC); others you perform on the Kubernetes server. Once you have integrated Kubernetes, you can use the Cisco APIC to view Kubernetes in the Cisco ACI.

This document provides the workflow for integrating Kubernetes and specific instructions for setting up the Cisco APIC. However, it is assumed that you are familiar with Kubernetes and containers and can install Kubernetes. Specific instructions for installing Kubernetes are beyond the scope of this document.

Hardware Requirements

This section provides the hardware requirements:

  • The use of symmetric policy-based routing (PBR) feature for load balancing external services requires the use of Cisco Nexus 9300-EX or FX leaf switches.

Workflow for Kubernetes Integration

This section provides a high-level description of the tasks required to integrate Kubernetes into the Cisco ACI fabric.

Planning for Kubernetes Integration

Various network resources are required to provide capabilities to the Kubernetes cluster, including several subnets, and routers.

You need the following subnets:

  • Node subnet—The subnet used for Kubernetes control traffic. It is where the Kubernetes API services are hosted. Make the node subnet a private subnet, and make sure that it has access to the Cisco APIC management address.

  • Pod subnet—The subnet from which the IP addresses of Kubernetes pods are allocated. Make the pod subnet a private subnet.


    Note

    This subnet specifies the starting address for the IP pool that is used to allocate IP addresses to pods and your Cisco ACI bridge domain IP. For example, if you define it as 192.168.255.254/16, it is a valid configuration from a Cisco ACI perspective. However, your containers will not get an IP address because there are no free IPs after 192.168.255.254 in this subnet. We suggest to always use the first IP address in the POD subnet. In this example: 192.168.0.1/16.
  • Node service subnet—The subnet used for internal routing of load-balanced service traffic. Make the node service subnet a private subnet.


    Note

    Similarly to the Pod subnet note, configure it with the first IP in the subnet.
  • External service subnets—Pools from which load-balanced services are allocated as externally accessible service IPs.

    The externally accessible service IPs could be globally routable. Configure the next-hop router to send traffic to these IPs to the fabric. There are two such pools: One is used for dynamically allocated IPs and the other is available for services to request a specific fixed external IP.

You need the following VLANS for local fabric use:

  • Node VLAN—The VLAN used by the physical domain for Kubernetes nodes.

  • Service VLAN—The VLAN used for delivery of load-balanced service traffic.

  • Infra VLAN—The infra VLAN used by the Cisco ACI fabric.

In addition to providing network resources, read and understand the guidelines in the knowledge base article Cisco ACI and OpFlex Connectivity for Orchestrators.

Prerequisites for Integrating Kubernetes with Cisco ACI

The following are required before you can integrate Kubernetes with the Cisco ACI fabric:

  • A working Cisco ACI installation

  • An attachable entity profile (AEP) set up with interfaces that are desired for the Kubernetes deployment

  • A Layer 3 Outside connection, along with a Layer 3 external network to serve as external access

  • Virtual routing and forwarding (VRF) configured.


    Note

    The VRF and L3Out in Cisco ACI that are used to provide outside connectivity to Kubernetes external services can be in any tenant. The most common usage is to put the VRF and L3Out in the common tenant or in a tenant that is dedicated to the Kubernetes cluster. You can also have separate VRFs, one for the Kubernetes bridge domains and one for the L3Out, and you can configure route leaking between them.
  • Any required route reflector configuration for the Cisco ACI fabric

  • A next-hop router that is connected to the Layer 3 external network capable of doing appropriate external access and configured with the required routes

In addition, the Kubernetes cluster must be up through the fabric-connected interface on all the hosts. The default route should be pointing to the ACI node subnet bridge domain. This is not mandatory, but it simplifies the routing configuration on the hosts and is the recommend configuration. If you choose not to follow this design, all Kubernetes related traffic must go through the fabric.

Provisioning Cisco ACI to Work with Kubernetes

Use the acc_provision tool to provision the fabric for the Kubernetes VMM domain and generate a .yaml file that Kubernetes uses to deploy the required Cisco ACI container components.


Note

We recommended that when using ESXi nested for Kubernetes hosts, you provision one Kubernetes host for each Kubernetes cluster for each ESXi server. Doing so ensures—in case of an ESXi host failure—that a single Kubernetes node is affected for each Kubernetes cluster.

Procedure


Step 1

Download the provisioning tool from cisco.com.

  1. Click APIC OpenStack and Container Plugins.

  2. Choose the package that you want to download.

  3. Click Download.

Step 2

Generate a sample configuration file that you can edit.

Example:

 
terminal$ acc-provision--sample > aci-containers-config.yaml
 
The command generates a configuration file that looks like the following example:


#
# Configuration for ACI Fabric
#
aci_config:
  system_id: mykube             # Every opflex cluster must have a distict ID
  apic_hosts:                   # List of APIC hosts to connect for APIC API
  - 10.1.1.101
  vmm_domain:                   # Kubernetes VMM domain configuration
    encap_type: vxlan           # Encap mode: vxlan or vlan
    mcast_range:                # Every opflex VMM must use a distinct range
      start: 225.20.1.1
      end: 225.20.255.255

  # The following resources must already exist on the APIC,
  # they are used, but not created by the provisioning tool.
  aep: kube-cluster             # The AEP for ports/VPCs used by this cluster
  vrf:                          # This VRF used to create all kubernetes EPs
    name: mykube-vrf
    tenant: common              # This can be system-id or common
  l3out:
    name: mykube_l3out          # Used to provision external IPs
    external_networks:
    - mykube_extepg             # Used for external contracts

#
# Networks used by Kubernetes
#
net_config:
  node_subnet: 10.1.0.1/16      # Subnet to use for nodes
  pod_subnet: 10.2.0.1/16       # Subnet to use for Kubernetes Pods
  extern_dynamic: 10.3.0.1/24   # Subnet to use for dynamic external IPs
  extern_static: 10.4.0.1/24    # Subnet to use for static external IPs
  node_svc_subnet: 10.5.0.1/24  # Subnet to use for service graph--This is not the same as the Kubernetes service-cluster-ip-range: Use different subnets.
  kubeapi_vlan: 4001            # The VLAN used by the physdom for nodes
  service_vlan: 4003            # The VLAN used by LoadBalancer services
  infra_vlan: 4093              # The VLAN used by ACI infra

#
# Configuration for container registry
# Update if a custom container registry has been setup
#
registry:
  image_prefix: noiro                   # e.g: registry.example.com/noiro
  # image_pull_secret: secret_name      # (if needed)
Note 
Do not modify the Cisco ACI bridge domain configuration that is pushed by the acc-provisioning tool. Setting the bridge domain to flood results in a broken environment.
Step 3

Edit and save the sample configuration file, providing information from your network.

Step 4

Provision the Cisco ACI fabric.

Example:

acc-provision -c aci-containers-config.yaml -o aci-containers.yaml -f kubernetes-<version> -a -u [apic username] -p [apic password]
 

This command generates the file aci-containers.yaml that you use after installing Kubernetes. It also creates the files user-[system id].key and user-[system id].crt that contain the certificate used to access Cisco APIC. Save these files in case you change the configuration later and want to avoid disrupting a running cluster because of a key change.

Note 
The file aci-containers.yaml is security sensitive. It contains keys necessary for connecting to the Cisco APIC administration API. 
 
Note 
Currently, the provisioning tool supports only the installation of a single Kubernetes cluster on a single or multipod Cisco ACI fabric. However, you can run the tool as often as needed to install multiple Kubernetes clusters. A single Cisco ACI installation can support more than one Kubernetes cluster.
 
Step 5

(Optional): Advanced optional parameters can be configured to adjust to custom parameters other than the ACI default values or base provisioning assumptions:

  • If your VMM’s multicast address for the fabric is different from 225.1.2.3, you can configure it using the following:
    aci_config:
      vmm_domain:
         mcast_fabric: 225.1.2.3
    
  • If you are using VLAN encapsulation, you can specify vlan-pool for it as follows:
    aci_config:
      vmm_domain:
        encap_type: vlan
        vlan_range:
          start: 10
          end: 25
    
  • If you want to use an existing User, or Key, or Cert, add the following:
    aci_config:
      sync_login:
        username: <name>
        certfile: <pem-file>
        keyfile: <pem-file>
    
  • If provisioning in a system nested inside of virtual machines, enter the name of an existing preconfigured VMM domain in Cisco ACI into the aci_config section under the vmm_domain of the configuration file.

    nested_inside:
        type: vmware
        name: myvmware

Preparing the Kubernetes Nodes

After you provision Cisco Application Centric Infrastructure (ACI), you prepare networking for the Kubernetes nodes.

Procedure


Step 1

Configure your uplink interface with NIC bonding or not, depending on how your AEP is configured.

Set the MTU on this interface to 1600.

Step 2

Create a subinterface on your uplink interface on your infra VLAN.

Configure this subinterface to obtain an IP address using DHCP. Set the MTU on this interface to 1600.

Step 3

Configure a static route for the multicast subnet 224.0.0.0/4 through the uplink interface used for VXLAN traffic.

Step 4

Create a subinterface on your uplink interface on your node VLAN. For example, which is called kubeapi_vlan in the configuration file.

Configure an IP address on this interface in your node subnet. Then set this interface and the corresponding node subnet router as the default route for the node.

Note 
Many Kubernetes installer tools look specifically for the default route to choose interfaces for API server traffic and other traffic. It’s possible to install with the default route on another interface. To accomplish this, you set up several static routes into this interface and override your installer configuration. However, we recommend setting up the default route through the node uplink.
 
Step 5

Create the /etc/dhcp/dhclient-eth0.4093.conf file with the following content, inserting the MAC address of the Ethernet interface for each server on the first line of the file:

Example:

Note 
If you have a single interface, you could name the file just dhclient.conf and not need the interface name, as in dhclient-eth0.4093.conf.

Example:

send dhcp-client-identifier 01:<mac-address of infra VLAN interface>;
request subnet-mask, domain-name, domain-name-servers, host-name;
send host-name <server-host-name>;
 
option rfc3442-classless-static-routes code 121 = array of unsigned integer 8;
option ms-classless-static-routes code 249 = array of unsigned integer 8;
option wpad code 252 = string;
 
also request rfc3442-classless-static-routes;
also request ms-classless-static-routes;
also request static-routes;
also request wpad;
also request ntp-servers;

The network interface on the infra VLAN requests a DHCP address from the Cisco Cisco Application Policy Infrastructure Controller (APIC) infrastructure network for OpFlex communication. Make sure that the server has a client configuration for this interface to receive all the correct DHCP options with the lease.

If you need information on how to configure a VPC interface for the Kubernetes servers, see "Manually Configure the Host vPC" in the Cisco ACI with OpenStack OpFlex Deployment Guide for Red Hat on Cisco.com.

Note 
The infra VLAN interface in your environment may be a basic Linux-level subinterface, such as eth0.4093.
Step 6

If you have a separate management interface for the node being configured, configure any static routes that you need to access your management network on the management interface.

Step 7

Ensure that Open vSwitch (OVS) is not running on the node.

Step 8

Informational: Here is an example of the interface configuration (/etc/network/interfaces):

# Management network interface (not connected to ACI)
auto ens160
iface ens160 inet static
    address  192.168.66.17
    netmask 255.255.255.0
    up route add -net 10.0.0.0/8 gw 192.168.66.1
    dns-nameservers  192.168.66.1

# Interface connected to ACI
auto ens192
iface ens192 inet manual
    mtu 1600

# ACI Infra VLAN
auto ens192.3095
iface ens192.3095 inet dhcp
    mtu 1600
    up route add -net 224.0.0.0/4 dev ens192.3095
    vlan-raw-device ens192

# Node Vlan 
auto ens192.4001
iface ens192.4001 inet static
    address  12.1.0.101
    netmask 255.255.0.0
    mtu 1600
    gateway 12.1.0.1
    vlan-raw-device ens192
Step 9

Tune the igmp_max_memberships kernel parameter.

The Cisco ACI Container Network Interface (CNI) plug-in gives you the flexibility to place Namespaces, Deployment, and PODs into dedicated endpoint groups (EPGs).

To ensure that Broadcast, Unknown Unicast and Multicast traffic (BUM) is not flooded to all the EPGs, Cisco ACI allocates a dedicated multicast address for BUM traffic replication for every EPG that is created.

Recent kernel versions set the default for the igmp_max_memberships parameter to 20, limiting the maximum number of EPGs that can be utilized to 20. To have more than 20 EPGs, you can increase the igmp_max_memberships with the following steps:

  1. Check the current configured maximum with the following command:

    sysctl net.ipv4.igmp_max_memberships
  2. Edit the /etc/sysctl.d/99-sysctl.conf file and add the following line:

    net.ipv4.igmp_max_memberships = Max_Number_Of_EPGs
  3. Restart the node or issue the following command:

    sysctl -p
  4. Verify that the new limit is correctly configured with the following command:

    sysctl net.ipv4.igmp_max_memberships
Note 
For more details about the igmp_max_memberships kernel parameter, see the article "Documentation for /proc/sys" on The Linux Kernel Archives website. Go to the Documentation section and search for "Documentation for pro/sys/net."

Installing Kubernetes and Cisco ACI Containers

After you provision Cisco ACI and prepare the Kubernetes nodes, you can install Kubernetes and ACI containers. You can use any installation method you choose appropriate to your environment. This procedure provides guidance and high-level instruction for installation; for details, consult Kubernetes documentation.

When installing Kubernetes, ensure that the API server is bound to the IP addresses on the node subnet and not to management or other IP addresses. Issues with node routing table configuration and API server advertisement addresses are the most common problems during installation. Should problems occur, check these first.

Install Kubernetes so that it is configured to use a Container Network Interface (CNI) plug-in, but do not install a specific CNI plug-in configuration through your installer. Instead, deploy the CNI plug-in.

Procedure


Install the CNI plug-in using the following command: kubectl apply -f aci-containers.yaml   

  
Note 

You can perform the command wherever you have kubectl set up, generally from a Kubernetes master node. The command installs the following:

  • ACI Containers Host Agent and OpFlex agent in a DaemonSet called aci-containers-host

  • Open vSwitch in a DaemonSet called aci-containers-openvswitch

  • ACI Containers Controller in a deployment called aci-containers-controller.

  • Other required configurations, including service accounts, roles, and security context


Verifying the Kubernetes Integration

After you have performed the previous steps, you can verify the integration in the Cisco APIC GUI. The integration creates a tenant, three EPGs, and a VMM domain.

Procedure


Step 1

Log in to the Cisco APIC.

Step 2

Go to Tenants > tenant.

The tenant should have the name that you specified in the configuration file that you edited and used in installing Kubernetes and the ACI containers.

Step 3

In the tenant navigation pane, expand the following: tenant > Application Profiles > application profile > Application EPGs.

You should see three folders inside the Application EPGs folder:

  • kube-default—The default EPG for containers that are otherwise not mapped to any specific EPG.

  • kube-nodes—The EPG for the Kubernetes nodes.

  • kube-system—The EPG for the kube-system Kubernetes namespace. This typically contains the kube-dns pods that provide DNS services for a Kubernetes cluster.

Step 4

In the tenant navigation pane, expand the Networking and Bridge Domains folders.

You should see two bridge domains:

  • node-bd—The bridge domain used by the node EPG.

  • pod-bd—The bridge domain used by all pods.

Step 5

If you deploy Kubernetes with a load balancer, go to Tenants > common, expand L4-L7 Services, and perform the following steps:

  1. Open the L4-L7 Service Graph Templates folder; you should see a template for Kubernetes.

  2. Open the L4-L7 Devices folder; you should see a device for Kubernetes.

  3. Open the Deployed Graph Instances folder; you should see an instance for Kubernetes.

Step 6

Go to VM Networking > Inventory.

Step 7

In the Inventory navigation pane, expand the Kubernetes folder.

You should see that a VMM domain, with the name that you provided in the configuration file, is created and that the domain contains a folder called Nodes and a folder called Namespaces.


Unprovisioning Kubernetes from the ACI Fabric

This section describes how to uprovision Kubernetes from the ACI fabric.

Procedure


To unprovision, enter the following command:

Example:


acc-provision -c aci-containers-config.yaml -o aci-containers.yaml \
-a -d -u [apic username] -f kubernetes-<version> -p [apic password]

This command unprovisions the resources that have been allocated for this Kubernetes.

This also deletes the tenant. If you are using a shared tenant, this is very dangerous.


Uninstalling the CNI Plug-In

This section describes how to uninstall the CNI plug-in.

Procedure


Uninstall the CNI plug-in using the following command:

Example:


kubectl delete -f aci-containers.yaml

Using Policy

Network Policy and EPGs

The Cisco ACI and Kubernetes integration was designed to offer a highly flexible approach to policy. It was based on two premises: that Kubernetes templates not need to change when they run on Cisco ACI, and that developers not be forced to implement any APIC configuration to launch new applications. At the same time, the solution optionally exposes Cisco ACI EPGs and contacts to Kubernetes users if they choose to leverage them for application isolation.

By default, Cisco plug-ins create an EPG and a bridge domain in APIC for the entire Kubernetes cluster. All pods by default are attached to the new EPG, has no special properties. The container team or the network team do not need to take any further action for a fully functional Kubernetes cluster—as one might find in a public cloud environment. Also, security enforcement can occur based on usage of the Kubernetes NetworkPolicy API. NetworkPolicy objects are transparently mapped into Cisco ACI and enforced for containers within the same EPG and between EPGs.

The following is an example of NetworkPolicy in Kubernetes:

apiVersion: networking.k8s.io/v1
kind: NetworkPolicy
metadata:
  name: test-network-policy
  namespace: default
spec:
  podSelector:
    matchLabels:
      role: db
  ingress:
  - from:
    - namespaceSelector:
        matchLabels:
          project: myproject
    - podSelector:
        matchLabels:
          role: frontend
    ports:
    - protocol: TCP
      port: 6379
 

However, in many cases, you may want to leverage EPGs and contracts in a more flexible way to define policy. You can create additional EPGs and contracts either directly or through the Cisco APIC.

To move namespaces, deployments, replication controllers, or pods into these EPGs, a Kubernetes user simply applies an annotation to any of these objects specifying the application profile and EPG. The running pods automatically shift to the new EPG, and any configured contracts are applied. In this model, it is still possible to use Kubernetes NetworkPolicy, which is honored regardless of how pods are mapped to EPGs.

Mapping to Cisco APIC

Each Kubernetes cluster is represented by a tenant within Cisco APIC. By default, all pods are placed in a single EPG created automatically by the plug-ins. However, it is possible to map a namespace, deployment, or pod to a specific application profile and EPG in Kubernetes through Kubernetes annotations.

While this is a highly flexible model, there are three typical ways to use it:

  • EPG=Kubernetes Cluster—This is the default behavior and provides the simplest solution. All pods are placed in a single EPG, kube-default.

  • EPG=Namespace—This approach can be used to add namespace isolation to Kubernetes. While Kubernetes does not dictate network isolation between namespaces, many users may find this desirable. Mapping EPGs to namespaces accomplishes this isolation.

  • EPG=Deployment—A Kubernetes deployment represents a replicated set of pods for a microservice. You can put that set of pods in its EPG as a means of isolating specific microservices and then use contracts between them.

Creating and Mapping an EPG

Use this procedure to create and EPG, using annotations to map namespaces or deployments into it.

For information about EPGs and bridge domains, see the Cisco APIC Basic Configuration Guide.

Procedure


Step 1

Log in to Cisco APIC.

Step 2

Create the EPG and add it to the bridge domain kube-pod-bd.

Step 3

Attach the EPG to the VMM domain.

Step 4

Configure the EPG to consume contracts in the Kubernetes tenant:

Step 5

Configure the EPG to consume contracts in the common tenant:

Consume: kube-l3out-allow-all (optional)

Step 6

Create any contracts you need for your application and provide and consume them as needed.

Step 7

Apply annotations to the namespaces or deployments.

You can apply annotations in three ways:

  • Through kubectl:

    • Deployment example:
      kubectl --namespace=namespace annotate deployment deployment opflex.cisco.com/endpoint-group='{"tenant":"tenant","app-profile":"app-profile","name":"EPG"}'
    • Namespace example:

      kubectl annotate namespace namespace opflex.cisco.com/endpoint-group='{"tenant":"tenant","app-profile":"app-profile","name":"EPG"}'
  • Through acikubectl:

    • Deployment example:
      acikubectl set default-eg deployment deployment -n namespace -t kube -a app-profile -g EPG
      
    • Namespace example:
      acikubectl set default-eg namespace namespace -t kube -a app-profile -g EPG
      
  • Through the .yaml file:
    annotations:                                          
    opflex.cisco.com/endpoint-group: {
    "tenant":"tenant",
    "app-profile":"app-profile",
    "name":"EPG"
    }

The acikubectl Command

The acikubectl command is a command-line utility that provides an abbreviated way to manage Cisco ACI policies for Kubernetes objects and annotations. It also enables you to debug the system and collect logs.

The acikubectl command includes a --help option that displays descriptions of the command's supported syntax and options, as seen in the following example:

acikubectl -–help
 
Available Commands:
  debug       Commands to help diagnose problems with ACI containers
  get         Get a value
  help        Help about any command
  set         Set a value

Load Balancing External Services

For Kubernetes services that are exposed externally and need to be load balanced, Kubernetes does not handle the provisioning of the load balancing. It is expected that the load balancing network function is implemented separately. For these services, Cisco ACI takes advantage of the symmetric policy-based routing (PBR) feature available in the Cisco Nexus 9300-EX or FX leaf switches in ACI mode.

On ingress, incoming traffic to an externally exposed service is redirected by PBR to one of the Kubernetes nodes that hosts at least one pod for that particular service. Each node hosts a special service endpoint that handles the traffic for all external services hosted for that endpoint. Traffic that reaches the service endpoint is not rewritten by the fabric, so it retains its original destination IP address. It is not expected that the Kubernetes pods handle traffic that is sent to the service IP address, so Cisco ACI performs the necessary network address translation (NAT).

If a Kubernetes worker node contains more than one IP pod for a particular service, the traffic is load balanced a second time across all the local pods for that service.

A simplified view of the Cisco ACI policy model required for the north-south load balancer is shown in the following illustration.

Figure 1. Load Balancer Logical Path

load balancer logical path

In the example, service IP 10.4.0.1 is exposed as an external service. It is presented as a /32 Layer 3 external network. This network provides a contract that is consumed by the default /0 Layer 3 external network. Traffic that comes into the fabric from the outside hits this contract and is redirected by the service graph to the correct set of Kubernetes service endpoints.

The Cisco ACI Kubernetes integration components are responsible for automatically writing the Cisco ACI policy that implements the external service into Cisco ACI.

Draining a Node

This section describes how to drain the pods on a node.

Procedure


Step 1

Drain the node:

Example:


kubectl drain --ignore-daemonsets <nodename>
Step 2

After the drain is finished, scale the controller down to 0 replicas:

Example:


kubectl scale deployment aci-containers-controller --replicas=0 -n aci-containers-system
Step 3

Remove the old annotations from the node by editing the node description and remove the opflex.cisco.com annotations and save the changes:

Example:


kubectl edit node <nodename>
Step 4

Find and delete the hostagent pod for the drained node:

Example:


kubectl delete pod aci-containers-host-
xxxx -n aci-containers-system
Note 

The pod remains in Terminating state until you uncordon it.

Step 5

Bring the controller up again:

Example:


kubectl scale deployment aci-containers-controller --replicas=1 -n aci-containers-system
Step 6

Bring back the node:

Example:


kubectl uncordon <nodename>
Step 7

Verify annotations are present and that the hostagent pod is running:

Example:


kubectl describe node <nodename> | grep cisco
kubectl get pods -n aci-containers-system -o wide | grep <nodename>

Moving a Node Between ESXi Hosts in Nested Mode

Follow the procedure in this section to move a node from one ESXI host to another when you have nested ESXi hosts.


Note

Using VMware vMotion to move nodes is not supported for nested ESXi hosts. Using VMware vMotion under these circumstances results in a loss of connectivity on the node that you are trying to move.

Procedure


Step 1

Create a new VM and add it to the cluster.

Step 2

Drain the old VM.

Step 3

Delete the old VM.


Service Subnet Advertisement

By default, service subnets are not advertised externally, requiring that external routers be configured with static routes. However, you can configure Cisco Application Centric Infrastructure (ACI) to advertise the service subnets through a dynamic routing protocol.

To configure and verify service subnet advertisement, complete the procedures in this section.

Configuring Service Subnet Advertisement

Complete the following steps to configure service subnet advertisement.


Note

Perform this procedure for all border leafs that are required to advertise the external subnets.

Procedure


Step 1

Add routes to null to the subnets:

  1. Log in to Cisco Application Policy Infrastructure Controller (APIC).

  2. Go to Tenant > Networking > External Routed Networks > your_L3-Out > Logical Node Profile > your_node_profile .

  3. In the node profile work pane, double-click a node.

  4. In the Node Association dialog box, in the Static Routes area, click the + (plus) icon.

  5. In the Create Static Route dialog box, in the Prefix field, enter the static route IP address and network mask that is assigned to the outside network.

    Note 
    Add a static route for external dynamic and external static subnets. Leave the Next Hop Addresses field empty. A null interface is automatically created.
  6. Click Submit.

    In the Node Association dialog box, click Close.

  7. In the Node Association dialog box, in the Static Routes area, repeat steps 1d through 1e for each node.

Step 2

Create match rules for the route map and add the subnet.

  1. Go to Tenant > Networking > External Routed Networks > Match Rules for Route Map.

  2. Right-click Match Rules for Route Map and choose Create Match Rule for a Route Map.

  3. In the Create Match Rule dialog box, in the Name field, enter a name for the match rule.

  4. In the Match Prefix area, click the + (plus) icon.

  5. In the IP field, enter the static route IP address and network mask that you entered in step 1e.

  6. Click Update and then click Submit.

Step 3

Create a route map for the L3 Out, add a context to it, and choose the match rules.

  1. Go to Tenant > Networking > External Routed Networks > your_L3-Out > Route Maps/Profiles.

  2. Right-click Route Maps/Profiles, and choose Create Route Maps/Profiles.

  3. In the Create Route Map dialog box, from the Name drop-down list, choose default-export.

  4. In the Type area, make sure that Match Prefix AND Routing Policy is chosen.

  5. In the Contexts area, click the + (plus) icon.

  6. In the Create Route Control Context dialog box, with the Order selector, choose 0.

  7. In the Name field, enter a name for the policy context.

  8. In the Action area, make sure that Permit is chosen.

  9. From the Match Rule drop-down list, choose the rule that you created in Step 2c.

  10. Click OK and then click Submit.


What to do next

Verify that the external routers have been configured with the external routes. See the section Verifying Service Subnet Advertisement.

Verifying Service Subnet Advertisement

Use NX-OS style CLI to verify that the external routers have been configured with the external routes. Perform the commands for each for the border leafs.

Before you begin

Ensure that your border leafs are configured with routes to Null0 for extern_static and extern_dynamic subnets (10.3.0.1/24 and 10.4.0.1/24 in the following example):
fab2-apic1# fabric 203 show ip route vrf common:k8s | grep null0 -B1
10.3.0.1/24, ubest/mbest: 1/0
    *via , null0, [1/0], 04:31:23, static
10.4.0.1/24, ubest/mbest: 1/0
    *via , null0, [1/0], 04:31:23, static

Procedure


Step 1

Check the route maps applied to your dynamic routing protocol that permits the advertisement of the subnets.

  1. Find your route map for static routes:

    Example:

    fabric 203 show ip ospf vrf common:k8s  | grep route-map
     Table-map using route-map exp-ctx-2981889-deny-external-tag
       static route-map exp-ctx-st-2981889
       direct route-map exp-ctx-st-2981889
       bgp route-map exp-ctx-proto-2981889
       eigrp route-map exp-ctx-proto-2981889
       coop route-map exp-ctx-st-2981889
    
Step 2

Find the specific route for each of the nodes, looking for entries that match the name of the match rule:

Example:

In the example, k8s-svc-export is the name of the match rule in Cisco Application Policy Infrastructure Controller (APIC).

fabric 203 show route-map exp-ctx-st-2981889 | grep k8s-svc-export
    ip address prefix-lists: IPv4-st19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst

fabric 204 show route-map  exp-ctx-proto-2981889 | grep k8s-svc-export
    ip address prefix-lists: IPv4-proto19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst
Step 3

Verify that the IP addresses are correct for each of the nodes:

Example:

fab2-apic1# fabric 203 show ip prefix-list IPv4-st19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst
----------------------------------------------------------------
 Node 203 (Leaf203)
----------------------------------------------------------------
ip prefix-list IPv4-st19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst: 2 entries
   seq 1 permit 10.3.0.1/24
   seq 2 permit 10.4.0.1/24

fab2-apic1# fabric 204 show ip prefix-list IPv4-proto19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst
----------------------------------------------------------------
 Node 204 (Leaf204)
----------------------------------------------------------------
ip prefix-list IPv4-proto19-2981889-exc-ext-out-k8s-svc-export2k8s-svc-export0k8s-svc-export-dst: 2 entries
   seq 1 permit 10.3.0.1/24
   seq 2 permit 10.4.0.1/24

Service Account Hardening

Every time that you create a cluster, a dedicated user account is automatically created in Cisco Application Policy Infrastructure Controller (APIC). This account is an administrative account with read and write permissions for the whole fabric.

Read and write permissions at the fabric level could be a security concern in case of multitenant fabrics where you do not want the cluster administrator to have administrative access to the Cisco Application Centric Infrastructure (ACI) fabric.

You can modify the dedicated user account limits and permissions. The level and scope of permission that is required for the cluster account depend on the location of the networking resources:

(The networking resources include the bridge domain, virtual routing and forwarding (VRF), and Layer 3 outside (L3Out).)

  • When cluster resources are in the cluster dedicated tenant, the account needs read and write access to the cluster tenant and the cluster container domain.

  • When cluster resources are in the common tenant, the account needs read and write access to the common tenant, the cluster tenant, and the cluster container domain.

Checking the Current Administrative Privileges

Complete the following procedure to see the current administrator privileges for the Cisco Application Centric Infrastructure (ACI) fabric.

Procedure


Step 1

Log in to Cisco Application Policy Infrastructure Controller (APIC).

Step 2

Go to Admin > AAA > Users.

Step 3

Click the username associated with your cluster.

Step 4

Scroll to the security domain and verify the following:

  • That the security domain "all" has role admin "writePriv" and "readPriv"

  • That the security domain "common" has role read-all "readPriv"


Modifying Administrative Account Permissions

After you configure the fabric, you can see a new tenant and Cisco Application Centric Infrastructure (ACI) user. Its name is equal to the system_id parameter specified in the Cisco ACI Container Network Interface (CNI) configuration file. Complete the following procedure to modify administrative permissions:


Note

This procedure works whether cluster resources are in the same tenant or when virtual routing and forwarding (VRF) and Layer 3 Outside (L3Out) are in the common tenant. However, if VRF and L3Out are in the common tenant, you must give write permission to the common tenant In Step 3.

Procedure


Step 1

Log in to the Cisco Application Policy Infrastructure Controller (APIC).

Step 2

Create a new security domain by completing the following steps:

  1. Go to Admin > AAA > Security.

  2. Right-click Security and choose Create Security Domain.

  3. In the Create Security Domain dialog box, enter a name for the security name and click Submit.

    We recommend that you use the Kubernetes system_id name. Entering a description is optional.

Step 3

Go to Admin > AAA > Users and complete the following steps:

  1. Double-click the system_id username.

  2. In the Local User dialog box, scroll to the Security Domains area.

  3. Expand Security Domain all, right-click Role Admin, and then click Delete from the drop-down list.

  4. Click the + (plus) icon.

  5. In the Add User Domain dialog box, choose the container system_id domain.

    You now add two new roles.

  6. Click the + (plus) icon.

  7. From the Name drop-down list, choose Admin, from the Access Type drop-down list, choose Write Privilege, and then click Update.

    Note 
    Complete the following three steps only if the cluster resources—such as the bridge domain, virtual routing and forwarding (VRF), and Layer 3 outside (L3Out)—are in the common tenant.
  8. From the drop-down list, choose common.

  9. Click the + (plus) icon.

  10. From the Name drop-down list, choose Admin, from the Access Type drop-down list, choose Write Privilege, and then click Update

  11. Click Submit.

Step 4

Create a custom role-based access control (RBAC) rule to allow the container account to write information into the container domain by completing the following steps:

  1. Go to Admin > Security and in the User Management - Security central pane, choose RBAC Rules and Explicit Rules.

  2. Click the tools icon and choose Create RBAC Rule from the drop-down list.

  3. In the Create RBAC Rule dialog box, in the DN field, enter comp/prov-Kubernetes/ctrlr-[system_id]-system_id.

    Example:

    If your system_id is "k8s," the DN is comp/prov-Kubernetes/ctrlr-[k8s]-k8s
  4. From the Domain drop-down list, choose the security domain that you created in Step 2.

  5. Set the Allow Writes option to Yes.

  6. Click Submit.

Step 5

Map the security domain to the cluster tenant by completing the following steps:

  1. Go to Tenants, choose the tenant for the Kuber netes system_id, and then in the Tenant central pane, choose Policy.

  2. In the Security Domains area, click the + (plus) icon.

  3. From the Name drop-down list, choose the newly created security name, click Update, and then click Submit.


Troubleshooting Kubernetes Integration

This section contains instructions for troubleshooting the Kubernetes integration with Cisco ACI.

Troubleshooting Checklist

This section contains a checklist to troubleshoot problems that occur after you integrate Kubernetes with Cisco ACI.

Procedure


Step 1

Check for faults on the fabric and resolve any that are relevant.

Step 2

Check that the API server advertisement addresses use the node subnet, and that the nodes are configured to route all Kubernetes subnets over the node uplink.

Typically, the API server advertisement address is pulled from the default route on the node during installation. If you are putting the default route on a different network than the node uplink interfaces, you should do so—in addition to configuring the subnets from the planning process and the cluster IP subnet used internally for Kubernetes.

Step 3

Check the logs for the container aci-containers-controller for errors using the following command on the Kubernetes master node: acikubectl debug logs controller acc

Step 4

Check these node-specific logs for errors using the following commands on the Kubernetes master node:

  1. Host agent: acikubectl debug logs node -n [nodename] host-agent

  2. OpFlex agent: acikubectl debug logs node -n [nodename] opflex-agent

  3. Open vSwitch: acikubectl debug logs node -n [nodename] openvswitch


Troubleshooting Specific Problems

Collecting and Exporting Logs

Collecting and exporting logs can help you and Cisco Support troubleshoot problems.

Procedure


Enter the following command to collect and export Kubernetes logs: acikubectl debug cluster-report -o cluster-report.tar.gz


Troubleshooting External Connectivity

Follow the instructions in this section if external connectivity is not working.

Procedure


Check configuration of the next-hop router.

Note 
You cannot access external services from the next-hop router will not work because contracts are not enforced in this case by the fabric. Instead, access external services from an IP address that is not in the subnet configured on the next-hop router interface.

Troubleshooting POD EPG Communication

Follow the instructions in this section if communication between two pod EPGs is not working.

Procedure


Check the contracts between the pod EPGs.

Ensure that you verify that there is a contract that allows ARP traffic. All pods are in the same subnet so Address Resolution Protocol (ARP) is required.


Troubleshooting Endpoint Discovery

If an endpoint is not automatically discovered, either EPG does not exist or mapping of the annotation to EPG is not in place. Follow the instructions in this section to troubleshoot the problem.

Procedure


Step 1

Ensure that the EPG name, tenant, and application are spelled correctly.

Step 2

Make sure that the VMM domain is mapped to an EPG.


Troubleshooting Pod Health Check

Follow the instructions in this section if the pod health check doesn't work.

Procedure


Ensure that the health check contract exists between the pod EPG and the node EPG.


Troubleshooting aci-containers-host

Follow the instructions in this section if the mcast-daemon inside aci-containers-host fails to start.

Procedure


Check the mcast-daemon log messages:

Example:

kubectl -n kube-system logs aci-containers-host-[xxxxx] mcast-daemon

If the following error message is present, Fatal error: open: Address family not supported by protocol, ensure that IPv6 support is enabled in the kernel. IPv6 must be enabled in the kernel for the mcast-daemon to start.


Contacting Support

If you need help with troubleshooting problems, generate a cluster report file and contact Cisco TAC for support.