Networking
Configuring and managing cluster networking
Abstract
Chapter 1. About the OVN-Kubernetes network plugin
The OVN-Kubernetes Container Network Interface (CNI) plugin is the default networking solution for MicroShift clusters. OVN-Kubernetes is a virtualized network for pods and services that is based on Open Virtual Network (OVN).
-
Default network configuration and connections are applied automatically in MicroShift with the
microshift-networking
RPM during installation. - A cluster that uses the OVN-Kubernetes network plugin also runs Open vSwitch (OVS) on the node.
- OVN-K configures OVS on the node to implement the declared network configuration.
-
Host physical interfaces are not bound by default to the OVN-K gateway bridge,
br-ex
. You can use standard tools on the host for managing the default gateway, such as the Network Manager CLI (nmcli
). - Changing the CNI is not supported on MicroShift.
Using configuration files or custom scripts, you can configure the following networking settings:
- You can use subnet CIDR ranges to allocate IP addresses to pods.
- You can change the maximum transmission unit (MTU) value.
- You can configure firewall ingress and egress.
- You can define network policies in the MicroShift cluster, including ingress and egress rules.
1.1. MicroShift networking customization matrix
The following table summarizes the status of networking features and capabilities that are either present as defaults, supported for configuration, or not available with the MicroShift service:
Network feature | Availability | Customization supported |
---|---|---|
Advertise address | Yes | Yes [1] |
Kubernetes network policy | Yes | Yes |
Kubernetes network policy logs | Not available | N/A |
Load balancing | Yes | Yes |
Multicast DNS | Yes | Yes [2] |
Network proxies | Yes [3] | CRI-O |
Network performance | Yes | MTU configuration |
Egress IPs | Not available | N/A |
Egress firewall | Not available | N/A |
Egress router | Not available | N/A |
Firewall | No [4] | Yes |
Hardware offloading | Not available | N/A |
Hybrid networking | Not available | N/A |
IPsec encryption for intra-cluster communication | Not available | N/A |
IPv6 | Not available [5] | N/A |
-
If unset, the default value is set to the next immediate subnet after the service network. For example, when the service network is
10.43.0.0/16
, theadvertiseAddress
is set to10.44.0.0/32
. -
You can use the multicast DNS protocol (mDNS) to allow name resolution and service discovery within a Local Area Network (LAN) using multicast exposed on the
5353/UDP
port. - There is no built-in transparent proxying of egress traffic in MicroShift. Egress must be manually configured.
- Setting up the firewalld service is supported by RHEL for Edge.
- IPv6 is not available in any configuration.
1.1.1. Default settings
If you do not create a config.yaml
file, default values are used. The following example shows the default configuration settings.
To see the default values, run the following command:
$ microshift show-config
Default values example output in YAML form
dns: baseDomain: microshift.example.com 1 network: clusterNetwork: - 10.42.0.0/16 2 serviceNetwork: - 10.43.0.0/16 3 serviceNodePortRange: 30000-32767 4 node: hostnameOverride: "" 5 nodeIP: "" 6 apiServer: advertiseAddress: 10.44.0.0/32 7 subjectAltNames: [] 8 debugging: logLevel: "Normal" 9
- 1
- Base domain of the cluster. All managed DNS records will be subdomains of this base.
- 2
- A block of IP addresses from which Pod IP addresses are allocated.
- 3
- A block of virtual IP addresses for Kubernetes services.
- 4
- The port range allowed for Kubernetes services of type NodePort.
- 5
- The name of the node. The default value is the hostname.
- 6
- The IP address of the node. The default value is the IP address of the default route.
- 7
- A string that specifies the IP address from which the API server is advertised to members of the cluster. The default value is calculated based on the address of the service network.
- 8
- Subject Alternative Names for API server certificates.
- 9
- Log verbosity. Valid values for this field are
Normal
,Debug
,Trace
, orTraceAll
.
1.2. Network features
Networking features available with MicroShift 4.14 include:
- Kubernetes network policy
- Dynamic node IP
- Custom gateway interface
- Second gateway interface
- Cluster network on specified host interface
- Blocking external access to NodePort service on specific host interfaces
Networking features not available with MicroShift 4.14:
- Egress IP/firewall/QoS: disabled
- Hybrid networking: not supported
- IPsec: not supported
- Hardware offload: not supported
Additional resources
1.3. IP forward
The host network sysctl net.ipv4.ip_forward
kernel parameter is automatically enabled by the ovnkube-master
container when started. This is required to forward incoming traffic to the CNI. For example, accessing the NodePort service from outside of a cluster fails if ip_forward
is disabled.
1.4. Network performance optimizations
By default, three performance optimizations are applied to OVS services to minimize resource consumption:
-
CPU affinity to
ovs-vswitchd.service
andovsdb-server.service
-
no-mlockall
toopenvswitch.service
-
Limit handler and
revalidator
threads toovs-vswitchd.service
1.5. MicroShift networking components and services
This brief overview describes networking components and their operation in MicroShift. The microshift-networking
RPM is a package that automatically pulls in any networking-related dependencies and systemd services to initialize networking, for example, the microshift-ovs-init
systemd service.
- NetworkManager
-
NetworkManager is required to set up the initial gateway bridge on the MicroShift node. The NetworkManager and
NetworkManager-ovs
RPM packages are installed as dependencies to themicroshift-networking
RPM package, which contains the necessary configuration files. NetworkManager in MicroShift uses thekeyfile
plugin and is restarted after installation of themicroshift-networking
RPM package. - microshift-ovs-init
-
The
microshift-ovs-init.service
is installed by themicroshift-networking
RPM package as a dependent systemd service tomicroshift.service
. It is responsible for setting up the OVS gateway bridge. - OVN containers
Two OVN-Kubernetes daemon sets are rendered and applied by MicroShift.
-
ovnkube-master Includes the
northd
,nbdb
,sbdb
andovnkube-master
containers. ovnkube-node The ovnkube-node includes the OVN-Controller container.
After MicroShift starts, the OVN-Kubernetes daemon sets are deployed in the
openshift-ovn-kubernetes
namespace.
-
ovnkube-master Includes the
- Packaging
OVN-Kubernetes manifests and startup logic are built into MicroShift. The systemd services and configurations included in the
microshift-networking
RPM are:-
/etc/NetworkManager/conf.d/microshift-nm.conf
forNetworkManager.service
-
/etc/systemd/system/ovs-vswitchd.service.d/microshift-cpuaffinity.conf
forovs-vswitchd.service
-
/etc/systemd/system/ovsdb-server.service.d/microshift-cpuaffinity.conf
forovs-server.service
-
/usr/bin/configure-ovs-microshift.sh
formicroshift-ovs-init.service
-
/usr/bin/configure-ovs.sh
formicroshift-ovs-init.service
-
/etc/crio/crio.conf.d/microshift-ovn.conf
for the CRI-O service
-
1.6. Bridge mappings
Bridge mappings allow provider network traffic to reach the physical network. Traffic leaves the provider network and arrives at the br-int
bridge. A patch port between br-int
and br-ex
then allows the traffic to traverse to and from the provider network and the edge network. Kubernetes pods are connected to the br-int
bridge through virtual ethernet pair: one end of the virtual ethernet pair is attached to the pod namespace, and the other end is attached to the br-int
bridge.
1.7. Network topology
OVN-Kubernetes provides an overlay-based networking implementation. This overlay includes an OVS-based implementation of Service and NetworkPolicy. The overlay network uses the Geneve (Generic Network Virtualization Encapsulation) tunnel protocol. The pod maximum transmission unit (MTU) for the Geneve tunnel is set to the default route MTU if it is not configured.
To configure the MTU, you must set an equal-to or less-than value than the MTU of the physical interface on the host. A less-than value for the MTU makes room for the required information that is added to the tunnel header before it is transmitted.
OVS runs as a systemd service on the MicroShift node. The OVS RPM package is installed as a dependency to the microshift-networking
RPM package. OVS is started immediately when the microshift-networking
RPM is installed.
MicroShift network topology
1.7.1. Description of the OVN logical components of the virtualized network
- OVN node switch
A virtual switch named
<node-name>
. The OVN node switch is named according to the hostname of the node.-
In this example, the
node-name
ismicroshift-dev
.
-
In this example, the
- OVN cluster router
A virtual router named
ovn_cluster_router
, also known as the distributed router.-
In this example, the cluster network is
10.42.0.0/16
.
-
In this example, the cluster network is
- OVN join switch
-
A virtual switch named
join
. - OVN gateway router
-
A virtual router named
GR_<node-name>
, also known as the external gateway router. - OVN external switch
-
A virtual switch named
ext_<node-name>.
1.7.2. Description of the connections in the network topology figure
-
The north-south traffic between the network service and the OVN external switch
ext_microshift-dev
is provided through the host kernel by the gateway bridgebr-ex
. -
The OVN gateway router
GR_microshift-dev
is connected to the external network switchext_microshift-dev
through the logical router port 4. Port 4 is attached with the node IP address 192.168.122.14. The join switch
join
connects the OVN gateway routerGR_microshift-dev
to the OVN cluster routerovn_cluster_router
. The IP address range is 100.62.0.0/16.-
The OVN gateway router
GR_microshift-dev
connects to the OVN join switchjoin
through the logical router port 3. Port 3 attaches with the internal IP address 100.64.0.2. -
The OVN cluster router
ovn_cluster_router
connects to the join switchjoin
through the logical router port 2. Port 2 attaches with the internal IP address 100.64.0.1.
-
The OVN gateway router
-
The OVN cluster router
ovn_cluster_router
connects to the node switchmicroshift-dev
through the logical router port 1. Port 1 is attached with the OVN cluster network IP address 10.42.0.1. -
The east-west traffic between the pods and the network service is provided by the OVN cluster router
ovn_cluster_router
and the node switchmicroshift-dev
. The IP address range is 10.42.0.0/24. -
The east-west traffic between pods is provided by the node switch
microshift-dev
without network address translation (NAT). -
The north-south traffic between the pods and the external network is provided by the OVN cluster router
ovn_cluster_router
and the host network. This router is connected through theovn-kubernetes
management portovn-k8s-mp0
, with the IP address 10.42.0.2. All the pods are connected to the OVN node switch through their interfaces.
-
In this example, Pod 1 and Pod 2 are connected to the node switch through
Interface 1
andInterface 2
.
-
In this example, Pod 1 and Pod 2 are connected to the node switch through
Chapter 2. Understanding networking settings
Learn how to apply networking customization and default settings to MicroShift deployments. Each node is contained to a single machine and single MicroShift, so each deployment requires individual configuration, pods, and settings.
Cluster Administrators have several options for exposing applications that run inside a cluster to external traffic and securing network connections:
- A service such as NodePort
-
API resources, such as
Ingress
andRoute
By default, Kubernetes allocates each pod an internal IP address for applications running within the pod. Pods and their containers can have traffic between them, but clients outside the cluster do not have direct network access to pods except when exposed with a service such as NodePort.
To troubleshoot connection problems with the NodePort service, read about the known issue in the Release Notes.
2.1. Creating an OVN-Kubernetes configuration file
MicroShift uses built-in default OVN-Kubernetes values if an OVN-Kubernetes configuration file is not created. You can write an OVN-Kubernetes configuration file to /etc/microshift/ovn.yaml
. An example file is provided for your configuration.
Procedure
To create your
ovn.yaml
file, run the following command:$ sudo cp /etc/microshift/ovn.yaml.default /etc/microshift/ovn.yaml
To list the contents of the configuration file you created, run the following command:
$ cat /etc/microshift/ovn.yaml.default
Example YAML file with default maximum transmission unit (MTU) value
mtu: 1400
To customize your configuration, you can change the MTU value. The table that follows provides details:
Table 2.1. Supported optional OVN-Kubernetes configurations for MicroShift Field Type Default Description Example mtu
uint32
auto
MTU value used for the pods
1300
ImportantIf you change the
mtu
configuration value in theovn.yaml
file, you must restart the host that MicroShift is running on to apply the updated setting.Example custom
ovn.yaml
configuration filemtu: 1300
2.2. Restarting the ovnkube-master pod
The following procedure restarts the ovnkube-master
pod.
Prerequisites
-
The OpenShift CLI (
oc
) is installed. -
Access to the cluster as a user with the
cluster-admin
role. - A cluster installed on infrastructure configured with the OVN-Kubernetes network plugin.
- The KUBECONFIG environment variable is set.
Procedure
Use the following steps to restart the ovnkube-master
pod.
Access the remote cluster by running the following command:
$ export KUBECONFIG=$PWD/kubeconfig
Find the name of the
ovnkube-master
pod that you want to restart by running the following command:$ pod=$(oc get pods -n openshift-ovn-kubernetes | awk -F " " '/ovnkube-master/{print $1}')
Delete the
ovnkube-master
pod by running the following command:$ oc -n openshift-ovn-kubernetes delete pod $pod
Confirm that a new
ovnkube-master
pod is running by using the following command:$ oc get pods -n openshift-ovn-kubernetes
The listing of the running pods shows a new
ovnkube-master
pod name and age.
2.3. Deploying MicroShift behind an HTTP(S) proxy
Deploy a MicroShift cluster behind an HTTP(S) proxy when you want to add basic anonymity and security measures to your pods.
You must configure the host operating system to use the proxy service with all components initiating HTTP(S) requests when deploying MicroShift behind a proxy.
All the user-specific workloads or pods with egress traffic, such as accessing cloud services, must be configured to use the proxy. There is no built-in transparent proxying of egress traffic in MicroShift.
2.4. Using the RPM-OStree HTTP(S) proxy
To use the HTTP(S) proxy in RPM-OStree, you must add a Service
section to the configuration file and set the http_proxy environment
variable for the rpm-ostreed
service.
Procedure
Add this setting to the
/etc/systemd/system/rpm-ostreed.service.d/00-proxy.conf
file:[Service] Environment="http_proxy=http://$PROXY_USER:$PROXY_PASSWORD@$PROXY_SERVER:$PROXY_PORT/"
Next, reload the configuration settings and restart the service to apply your changes.
Reload the configuration settings by running the following command:
$ sudo systemctl daemon-reload
Restart the
rpm-ostreed
service by running the following command:$ sudo systemctl restart rpm-ostreed.service
2.5. Using a proxy in the CRI-O container runtime
To use an HTTP(S) proxy in CRI-O
, you must add a Service
section to the configuration file and set the HTTP_PROXY
and HTTPS_PROXY
environment variables. You can also set the NO_PROXY
variable to exclude a list of hosts from being proxied.
Procedure
Create the directory for the configuration file if it does not exist:
$ sudo mkdir /etc/systemd/system/crio.service.d/
Add the following settings to the
/etc/systemd/system/crio.service.d/00-proxy.conf
file:[Service] Environment=NO_PROXY="localhost,127.0.0.1" Environment=HTTP_PROXY="http://$PROXY_USER:$PROXY_PASSWORD@$PROXY_SERVER:$PROXY_PORT/" Environment=HTTPS_PROXY="http://$PROXY_USER:$PROXY_PASSWORD@$PROXY_SERVER:$PROXY_PORT/"
ImportantYou must define the
Service
section of the configuration file for the environment variables or the proxy settings fail to apply.Reload the configuration settings:
$ sudo systemctl daemon-reload
Restart the CRI-O service:
$ sudo systemctl restart crio
Restart the MicroShift service to apply the settings:
$ sudo systemctl restart microshift
Verification
Verify that pods are started by running the following command and examining the output:
$ oc get all -A
Verify that MicroShift is able to pull container images by running the following command and examining the output:
$ sudo crictl images
2.6. Getting a snapshot of OVS interfaces from a running cluster
A snapshot represents the state and data of OVS interfaces at a specific point in time.
Procedure
To see a snapshot of OVS interfaces from a running MicroShift cluster, use the following command:
$ sudo ovs-vsctl show
Example OVS interfaces in a running cluster
9d9f5ea2-9d9d-4e34-bbd2-dbac154fdc93 Bridge br-ex Port br-ex Interface br-ex type: internal Port patch-br-ex_localhost.localdomain-to-br-int 1 Interface patch-br-ex_localhost.localdomain-to-br-int type: patch options: {peer=patch-br-int-to-br-ex_localhost.localdomain} 2 Bridge br-int fail_mode: secure datapath_type: system Port patch-br-int-to-br-ex_localhost.localdomain Interface patch-br-int-to-br-ex_localhost.localdomain type: patch options: {peer=patch-br-ex_localhost.localdomain-to-br-int} Port eebee1ce5568761 Interface eebee1ce5568761 3 Port b47b1995ada84f4 Interface b47b1995ada84f4 4 Port "3031f43d67c167f" Interface "3031f43d67c167f" 5 Port br-int Interface br-int type: internal Port ovn-k8s-mp0 6 Interface ovn-k8s-mp0 type: internal ovs_version: "2.17.3"
- 1
- The
patch-br-ex_localhost.localdomain-to-br-int
andpatch-br-int-to-br-ex_localhost.localdomain
are OVS patch ports that connectbr-ex
andbr-int
. - 2
- The
patch-br-ex_localhost.localdomain-to-br-int
andpatch-br-int-to-br-ex_localhost.localdomain
are OVS patch ports that connectbr-ex
andbr-int
. - 3
- The pod interface
eebee1ce5568761
is named with the first 15 bits of the pod sandbox ID and is plugged into thebr-int
bridge. - 4
- The pod interface
b47b1995ada84f4
is named with the first 15 bits of the pod sandbox ID and is plugged into thebr-int
bridge. - 5
- The pod interface
3031f43d67c167f
is named with the first 15 bits of the pod sandbox ID and is plugged into thebr-int
bridge. - 6
- The OVS internal port for hairpin traffic,
ovn-k8s-mp0
is created by theovnkube-master
container.
2.7. Deploying a load balancer for a workload
MicroShift has a built-in implementation of network load balancers. The following example procedure uses the node IP address as the external IP address for the LoadBalancer
service configuration file. You can use this example as guidance for how to deploy load balancers for your workloads.
Prerequisites
-
The OpenShift CLI (
oc
) is installed. - You have access to the cluster as a user with the cluster administration role.
- You installed a cluster on an infrastructure configured with the OVN-Kubernetes network plugin.
-
The
KUBECONFIG
environment variable is set.
Procedure
Verify that your pods are running by running the following command:
$ oc get pods -A
Create the example namespace by running the following commands:
$ NAMESPACE=nginx-lb-test
$ oc create ns $NAMESPACE
The following example deploys three replicas of the test
nginx
application in your namespace:$ oc apply -n $NAMESPACE -f - <<EOF apiVersion: v1 kind: ConfigMap metadata: name: nginx data: headers.conf: | add_header X-Server-IP \$server_addr always; --- apiVersion: apps/v1 kind: Deployment metadata: name: nginx spec: replicas: 3 selector: matchLabels: app: nginx template: metadata: labels: app: nginx spec: containers: - image: quay.io/packit/nginx-unprivileged imagePullPolicy: Always name: nginx ports: - containerPort: 8080 volumeMounts: - name: nginx-configs subPath: headers.conf mountPath: /etc/nginx/conf.d/headers.conf securityContext: allowPrivilegeEscalation: false seccompProfile: type: RuntimeDefault capabilities: drop: ["ALL"] runAsNonRoot: true volumes: - name: nginx-configs configMap: name: nginx items: - key: headers.conf path: headers.conf EOF
You can verify that the three sample replicas started successfully by running the following command:
$ oc get pods -n $NAMESPACE
Create a
LoadBalancer
service for thenginx
test application with the following sample commands:$ oc create -n $NAMESPACE -f - <<EOF apiVersion: v1 kind: Service metadata: name: nginx spec: ports: - port: 81 targetPort: 8080 selector: app: nginx type: LoadBalancer EOF
NoteYou must ensure that the
port
parameter is a host port that is not occupied by otherLoadBalancer
services or MicroShift components.Verify that the service file exists, that the external IP address is properly assigned, and that the external IP is identical to the node IP by running the following command:
$ oc get svc -n $NAMESPACE
Example output
NAME TYPE CLUSTER-IP EXTERNAL-IP PORT(S) AGE nginx LoadBalancer 10.43.183.104 192.168.1.241 81:32434/TCP 2m
Verification
The following command forms five connections to the example
nginx
application using the external IP address of theLoadBalancer
service configuration. The result of the command is a list of those server IP addresses. Verify that the load balancer sends requests to all the running applications with the following command:EXTERNAL_IP=192.168.1.241 seq 5 | xargs -Iz curl -s -I http://$EXTERNAL_IP:81 | grep X-Server-IP
The output of the previous command contains different IP addresses if the load balancer is successfully distributing the traffic to the applications, for example:
Example output
X-Server-IP: 10.42.0.41 X-Server-IP: 10.42.0.41 X-Server-IP: 10.42.0.43 X-Server-IP: 10.42.0.41 X-Server-IP: 10.42.0.43
2.8. Blocking external access to the NodePort service on a specific host interface
OVN-Kubernetes does not restrict the host interface where a NodePort service can be accessed from outside a MicroShift node. The following procedure explains how to block the NodePort service on a specific host interface and restrict external access.
Prerequisites
- You must have an account with root privileges.
Procedure
Change the
NODEPORT
variable to the host port number assigned to your Kubernetes NodePort service by running the following command:# export NODEPORT=30700
Change the
INTERFACE_IP
value to the IP address from the host interface that you want to block. For example:# export INTERFACE_IP=192.168.150.33
Insert a new rule in the
nat
table PREROUTING chain to drop all packets that match the destination port and IP address. For example:$ sudo nft -a insert rule ip nat PREROUTING tcp dport $NODEPORT ip daddr $INTERFACE_IP drop
List the new rule by running the following command:
$ sudo nft -a list chain ip nat PREROUTING table ip nat { chain PREROUTING { # handle 1 type nat hook prerouting priority dstnat; policy accept; tcp dport 30700 ip daddr 192.168.150.33 drop # handle 134 counter packets 108 bytes 18074 jump OVN-KUBE-ETP # handle 116 counter packets 108 bytes 18074 jump OVN-KUBE-EXTERNALIP # handle 114 counter packets 108 bytes 18074 jump OVN-KUBE-NODEPORT # handle 112 } }
NoteNote the
handle
number of the newly added rule. You need to remove thehandle
number in the following step.Remove the custom rule with the following sample command:
$ sudo nft -a delete rule ip nat PREROUTING handle 134
2.9. The multicast DNS protocol
You can use the multicast DNS protocol (mDNS) to allow name resolution and service discovery within a Local Area Network (LAN) using multicast exposed on the 5353/UDP
port.
MicroShift includes an embedded mDNS server for deployment scenarios in which the authoritative DNS server cannot be reconfigured to point clients to services on MicroShift. The embedded DNS server allows .local
domains exposed by MicroShift to be discovered by other elements on the LAN.
2.10. Additional resources
Chapter 3. Network policies
3.1. About network policies
Learn how network policies work for MicroShift to restrict or allow network traffic to pods in your cluster.
3.1.1. How network policy works in MicroShift
In a cluster using the default OVN-Kubernetes Container Network Interface (CNI) plugin for MicroShift, network isolation is controlled by both firewalld, which is configured on the host, and by NetworkPolicy
objects created within MicroShift. Simultaneous use of firewalld and NetworkPolicy
is supported.
-
Network policies work only within boundaries of OVN-Kubernetes-controlled traffic, so they can apply to every situation except for
hostPort/hostNetwork
enabled pods. -
Firewalld settings also do not apply to
hostPort/hostNetwork
enabled pods.
Firewalld rules run before any NetworkPolicy
is enforced.
Network policy does not apply to the host network namespace. Pods with host networking enabled are unaffected by network policy rules. However, pods connecting to the host-networked pods might be affected by the network policy rules.
Network policies cannot block traffic from localhost.
By default, all pods in a MicroShift node are accessible from other pods and network endpoints. To isolate one or more pods in a cluster, you can create NetworkPolicy
objects to indicate allowed incoming connections. You can create and delete NetworkPolicy
objects.
If a pod is matched by selectors in one or more NetworkPolicy
objects, then the pod accepts only connections that are allowed by at least one of those NetworkPolicy
objects. A pod that is not selected by any NetworkPolicy
objects is fully accessible.
A network policy applies to only the TCP, UDP, ICMP, and SCTP protocols. Other protocols are not affected.
The following example NetworkPolicy
objects demonstrate supporting different scenarios:
Deny all traffic:
To make a project deny by default, add a
NetworkPolicy
object that matches all pods but accepts no traffic:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: deny-by-default spec: podSelector: {} ingress: []
Allow connections from the default router, which is the ingress in MicroShift:
To allow connections from the MicroShift default router, add the following
NetworkPolicy
object:apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: allow-from-openshift-ingress spec: ingress: - from: - namespaceSelector: matchLabels: ingresscontroller.operator.openshift.io/deployment-ingresscontroller: default podSelector: {} policyTypes: - Ingress
Only accept connections from pods within the same namespace:
To make pods accept connections from other pods in the same namespace, but reject all other connections from pods in other namespaces, add the following
NetworkPolicy
object:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-same-namespace spec: podSelector: {} ingress: - from: - podSelector: {}
Only allow HTTP and HTTPS traffic based on pod labels:
To enable only HTTP and HTTPS access to the pods with a specific label (
role=frontend
in following example), add aNetworkPolicy
object similar to the following:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-http-and-https spec: podSelector: matchLabels: role: frontend ingress: - ports: - protocol: TCP port: 80 - protocol: TCP port: 443
Accept connections by using both namespace and pod selectors:
To match network traffic by combining namespace and pod selectors, you can use a
NetworkPolicy
object similar to the following:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-pod-and-namespace-both spec: podSelector: matchLabels: name: test-pods ingress: - from: - namespaceSelector: matchLabels: project: project_name podSelector: matchLabels: name: test-pods
NetworkPolicy
objects are additive, which means you can combine multiple NetworkPolicy
objects together to satisfy complex network requirements.
For example, for the NetworkPolicy
objects defined in previous examples, you can define both allow-same-namespace
and allow-http-and-https
policies. That configuration allows the pods with the label role=frontend
to accept any connection allowed by each policy. That is, connections on any port from pods in the same namespace, and connections on ports 80
and 443
from pods in any namespace.
3.1.2. Optimizations for network policy with OVN-Kubernetes network plugin
When designing your network policy, refer to the following guidelines:
-
For network policies with the same
spec.podSelector
spec, it is more efficient to use one network policy with multipleingress
oregress
rules, than multiple network policies with subsets ofingress
oregress
rules. Every
ingress
oregress
rule based on thepodSelector
ornamespaceSelector
spec generates the number of OVS flows proportional tonumber of pods selected by network policy + number of pods selected by ingress or egress rule
. Therefore, it is preferable to use thepodSelector
ornamespaceSelector
spec that can select as many pods as you need in one rule, instead of creating individual rules for every pod.For example, the following policy contains two rules:
apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: test-network-policy spec: podSelector: {} ingress: - from: - podSelector: matchLabels: role: frontend - from: - podSelector: matchLabels: role: backend
The following policy expresses those same two rules as one:
apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: test-network-policy spec: podSelector: {} ingress: - from: - podSelector: matchExpressions: - {key: role, operator: In, values: [frontend, backend]}
The same guideline applies to the
spec.podSelector
spec. If you have the sameingress
oregress
rules for different network policies, it might be more efficient to create one network policy with a commonspec.podSelector
spec. For example, the following two policies have different rules:apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: policy1 spec: podSelector: matchLabels: role: db ingress: - from: - podSelector: matchLabels: role: frontend --- apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: policy2 spec: podSelector: matchLabels: role: client ingress: - from: - podSelector: matchLabels: role: frontend
The following network policy expresses those same two rules as one:
apiVersion: networking.k8s.io/v1 kind: NetworkPolicy metadata: name: policy3 spec: podSelector: matchExpressions: - {key: role, operator: In, values: [db, client]} ingress: - from: - podSelector: matchLabels: role: frontend
You can apply this optimization when only multiple selectors are expressed as one. In cases where selectors are based on different labels, it may not be possible to apply this optimization. In those cases, consider applying some new labels for network policy optimization specifically.
3.2. Creating network policies
You can create a network policy for a namespace.
3.2.1. Example NetworkPolicy object
The following annotates an example NetworkPolicy object:
kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-27107 1 spec: podSelector: 2 matchLabels: app: mongodb ingress: - from: - podSelector: 3 matchLabels: app: app ports: 4 - protocol: TCP port: 27017
- 1
- The name of the NetworkPolicy object.
- 2
- A selector that describes the pods to which the policy applies.
- 3
- A selector that matches the pods from which the policy object allows ingress traffic. The selector matches pods in the same namespace as the NetworkPolicy.
- 4
- A list of one or more destination ports on which to accept traffic.
3.2.2. Creating a network policy using the CLI
To define granular rules describing ingress or egress network traffic allowed for namespaces in your cluster, you can create a network policy.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace that the network policy applies to.
Procedure
Create a policy rule:
Create a
<policy_name>.yaml
file:$ touch <policy_name>.yaml
where:
<policy_name>
- Specifies the network policy file name.
Define a network policy in the file that you just created, such as in the following examples:
Deny ingress from all pods in all namespaces
This is a fundamental policy, blocking all cross-pod networking other than cross-pod traffic allowed by the configuration of other Network Policies.
kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: deny-by-default spec: podSelector: {} policyTypes: - Ingress ingress: []
Allow ingress from all pods in the same namespace
kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-same-namespace spec: podSelector: ingress: - from: - podSelector: {}
Allow ingress traffic to one pod from a particular namespace
This policy allows traffic to pods labelled
pod-a
from pods running innamespace-y
.kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-traffic-pod spec: podSelector: matchLabels: pod: pod-a policyTypes: - Ingress ingress: - from: - namespaceSelector: matchLabels: kubernetes.io/metadata.name: namespace-y
To create the network policy object, enter the following command:
$ oc apply -f <policy_name>.yaml -n <namespace>
where:
<policy_name>
- Specifies the network policy file name.
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
Example output
networkpolicy.networking.k8s.io/deny-by-default created
3.2.3. Creating a default deny all network policy
This is a fundamental policy, blocking all cross-pod networking other than network traffic allowed by the configuration of other deployed network policies. This procedure enforces a default deny-by-default
policy.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace that the network policy applies to.
Procedure
Create the following YAML that defines a
deny-by-default
policy to deny ingress from all pods in all namespaces. Save the YAML in thedeny-by-default.yaml
file:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: deny-by-default namespace: default 1 spec: podSelector: {} 2 ingress: [] 3
Apply the policy by entering the following command:
$ oc apply -f deny-by-default.yaml
Example output
networkpolicy.networking.k8s.io/deny-by-default created
3.2.4. Creating a network policy to allow traffic from external clients
With the deny-by-default
policy in place you can proceed to configure a policy that allows traffic from external clients to a pod with the label app=web
.
Firewalld rules run before any NetworkPolicy
is enforced.
Follow this procedure to configure a policy that allows external service from the public Internet directly or by using a Load Balancer to access the pod. Traffic is only allowed to a pod with the label app=web
.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace that the network policy applies to.
Procedure
Create a policy that allows traffic from the public Internet directly or by using a load balancer to access the pod. Save the YAML in the
web-allow-external.yaml
file:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: web-allow-external namespace: default spec: policyTypes: - Ingress podSelector: matchLabels: app: web ingress: - {}
Apply the policy by entering the following command:
$ oc apply -f web-allow-external.yaml
Example output
networkpolicy.networking.k8s.io/web-allow-external created
3.2.5. Creating a network policy allowing traffic to an application from all namespaces
Follow this procedure to configure a policy that allows traffic from all pods in all namespaces to a particular application.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace that the network policy applies to.
Procedure
Create a policy that allows traffic from all pods in all namespaces to a particular application. Save the YAML in the
web-allow-all-namespaces.yaml
file:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: web-allow-all-namespaces namespace: default spec: podSelector: matchLabels: app: web 1 policyTypes: - Ingress ingress: - from: - namespaceSelector: {} 2
NoteBy default, if you omit specifying a
namespaceSelector
it does not select any namespaces, which means the policy allows traffic only from the namespace the network policy is deployed to.Apply the policy by entering the following command:
$ oc apply -f web-allow-all-namespaces.yaml
Example output
networkpolicy.networking.k8s.io/web-allow-all-namespaces created
Verification
Start a web service in the
default
namespace by entering the following command:$ oc run web --namespace=default --image=nginx --labels="app=web" --expose --port=80
Run the following command to deploy an
alpine
image in thesecondary
namespace and to start a shell:$ oc run test-$RANDOM --namespace=secondary --rm -i -t --image=alpine -- sh
Run the following command in the shell and observe that the request is allowed:
# wget -qO- --timeout=2 http://web.default
Expected output
<!DOCTYPE html> <html> <head> <title>Welcome to nginx!</title> <style> html { color-scheme: light dark; } body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } </style> </head> <body> <h1>Welcome to nginx!</h1> <p>If you see this page, the nginx web server is successfully installed and working. Further configuration is required.</p> <p>For online documentation and support please refer to <a href="http://nginx.org/">nginx.org</a>.<br/> Commercial support is available at <a href="http://nginx.com/">nginx.com</a>.</p> <p><em>Thank you for using nginx.</em></p> </body> </html>
3.2.6. Creating a network policy allowing traffic to an application from a namespace
Follow this procedure to configure a policy that allows traffic to a pod with the label app=web
from a particular namespace. You might want to do this to:
- Restrict traffic to a production database only to namespaces where production workloads are deployed.
- Enable monitoring tools deployed to a particular namespace to scrape metrics from the current namespace.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace that the network policy applies to.
Procedure
Create a policy that allows traffic from all pods in a particular namespaces with a label
purpose=production
. Save the YAML in theweb-allow-prod.yaml
file:kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: web-allow-prod namespace: default spec: podSelector: matchLabels: app: web 1 policyTypes: - Ingress ingress: - from: - namespaceSelector: matchLabels: purpose: production 2
Apply the policy by entering the following command:
$ oc apply -f web-allow-prod.yaml
Example output
networkpolicy.networking.k8s.io/web-allow-prod created
Verification
Start a web service in the
default
namespace by entering the following command:$ oc run web --namespace=default --image=nginx --labels="app=web" --expose --port=80
Run the following command to create the
prod
namespace:$ oc create namespace prod
Run the following command to label the
prod
namespace:$ oc label namespace/prod purpose=production
Run the following command to create the
dev
namespace:$ oc create namespace dev
Run the following command to label the
dev
namespace:$ oc label namespace/dev purpose=testing
Run the following command to deploy an
alpine
image in thedev
namespace and to start a shell:$ oc run test-$RANDOM --namespace=dev --rm -i -t --image=alpine -- sh
Run the following command in the shell and observe that the request is blocked:
# wget -qO- --timeout=2 http://web.default
Expected output
wget: download timed out
Run the following command to deploy an
alpine
image in theprod
namespace and start a shell:$ oc run test-$RANDOM --namespace=prod --rm -i -t --image=alpine -- sh
Run the following command in the shell and observe that the request is allowed:
# wget -qO- --timeout=2 http://web.default
Expected output
<!DOCTYPE html> <html> <head> <title>Welcome to nginx!</title> <style> html { color-scheme: light dark; } body { width: 35em; margin: 0 auto; font-family: Tahoma, Verdana, Arial, sans-serif; } </style> </head> <body> <h1>Welcome to nginx!</h1> <p>If you see this page, the nginx web server is successfully installed and working. Further configuration is required.</p> <p>For online documentation and support please refer to <a href="http://nginx.org/">nginx.org</a>.<br/> Commercial support is available at <a href="http://nginx.com/">nginx.com</a>.</p> <p><em>Thank you for using nginx.</em></p> </body> </html>
3.3. Editing a network policy
You can edit an existing network policy for a namespace. Typical edits might include changes to the pods to which the policy applies, allowed ingress traffic, and the destination ports on which to accept traffic. The apiVersion
, kind
, and name
fields must not be changed when editing NetworkPolicy
objects, as these define the resource itself.
3.3.1. Editing a network policy
You can edit a network policy in a namespace.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace where the network policy exists.
Procedure
Optional: To list the network policy objects in a namespace, enter the following command:
$ oc get networkpolicy
where:
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
Edit the network policy object.
If you saved the network policy definition in a file, edit the file and make any necessary changes, and then enter the following command.
$ oc apply -n <namespace> -f <policy_file>.yaml
where:
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
<policy_file>
- Specifies the name of the file containing the network policy.
If you need to update the network policy object directly, enter the following command:
$ oc edit networkpolicy <policy_name> -n <namespace>
where:
<policy_name>
- Specifies the name of the network policy.
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
Confirm that the network policy object is updated.
$ oc describe networkpolicy <policy_name> -n <namespace>
where:
<policy_name>
- Specifies the name of the network policy.
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
3.3.2. Example NetworkPolicy object
The following annotates an example NetworkPolicy object:
kind: NetworkPolicy apiVersion: networking.k8s.io/v1 metadata: name: allow-27107 1 spec: podSelector: 2 matchLabels: app: mongodb ingress: - from: - podSelector: 3 matchLabels: app: app ports: 4 - protocol: TCP port: 27017
- 1
- The name of the NetworkPolicy object.
- 2
- A selector that describes the pods to which the policy applies.
- 3
- A selector that matches the pods from which the policy object allows ingress traffic. The selector matches pods in the same namespace as the NetworkPolicy.
- 4
- A list of one or more destination ports on which to accept traffic.
3.4. Deleting a network policy
You can delete a network policy from a namespace.
3.4.1. Deleting a network policy using the CLI
You can delete a network policy in a namespace.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace where the network policy exists.
Procedure
To delete a network policy object, enter the following command:
$ oc delete networkpolicy <policy_name> -n <namespace>
where:
<policy_name>
- Specifies the name of the network policy.
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
Example output
networkpolicy.networking.k8s.io/default-deny deleted
3.5. Viewing a network policy
Use the following procedure to view a network policy for a namespace.
3.5.1. Viewing network policies using the CLI
You can examine the network policies in a namespace.
Prerequisites
-
You installed the OpenShift CLI (
oc
). - You are working in the namespace where the network policy exists.
Procedure
List network policies in a namespace:
To view network policy objects defined in a namespace, enter the following command:
$ oc get networkpolicy
Optional: To examine a specific network policy, enter the following command:
$ oc describe networkpolicy <policy_name> -n <namespace>
where:
<policy_name>
- Specifies the name of the network policy to inspect.
<namespace>
- Optional: Specifies the namespace if the object is defined in a different namespace than the current namespace.
For example:
$ oc describe networkpolicy allow-same-namespace
Output for
oc describe
commandName: allow-same-namespace Namespace: ns1 Created on: 2021-05-24 22:28:56 -0400 EDT Labels: <none> Annotations: <none> Spec: PodSelector: <none> (Allowing the specific traffic to all pods in this namespace) Allowing ingress traffic: To Port: <any> (traffic allowed to all ports) From: PodSelector: <none> Not affecting egress traffic Policy Types: Ingress
Chapter 4. Using a firewall
Firewalls are not required in MicroShift, but using a firewall can prevent undesired access to the MicroShift API.
4.1. About network traffic through the firewall
Firewalld is a networking service that runs in the background and responds to connection requests, creating a dynamic customizable host-based firewall. If you are using Red Hat Enterprise Linux for Edge (RHEL for Edge) with MicroShift, firewalld should already be installed and you just need to configure it. Details are provided in procedures that follow. Overall, you must explicitly allow the following OVN-Kubernetes traffic when the firewalld
service is running:
- CNI pod to CNI pod
- CNI pod to Host-Network pod Host-Network pod to Host-Network pod
- CNI pod
- The Kubernetes pod that uses the CNI network
- Host-Network pod
-
The Kubernetes pod that uses host network You can configure the
firewalld
service by using the following procedures. In most cases, firewalld is part of RHEL for Edge installations. If you do not have firewalld, you can install it with the simple procedure in this section.
MicroShift pods must have access to the internal CoreDNS component and API servers.
Additional resources
4.2. Installing the firewalld service
If you are using RHEL for Edge, firewalld should be installed. To use the service, you can simply configure it. The following procedure can be used if you do not have firewalld, but want to use it.
Install and run the firewalld
service for MicroShift by using the following steps.
Procedure
Optional: Check for firewalld on your system by running the following command:
$ rpm -q firewalld
If the
firewalld
service is not installed, run the following command:$ sudo dnf install -y firewalld
To start the firewall, run the following command:
$ sudo systemctl enable firewalld --now
4.3. Required firewall settings
An IP address range for the cluster network must be enabled during firewall configuration. You can use the default values or customize the IP address range. If you choose to customize the cluster network IP address range from the default 10.42.0.0/16
setting, you must also use the same custom range in the firewall configuration.
IP Range | Firewall rule required | Description |
---|---|---|
10.42.0.0/16 | No | Host network pod access to other pods |
169.254.169.1 | Yes | Host network pod access to MicroShift API server |
The following are examples of commands for settings that are mandatory for firewall configuration:
Example commands
Configure host network pod access to other pods:
$ sudo firewall-cmd --permanent --zone=trusted --add-source=10.42.0.0/16
Configure host network pod access to services backed by Host endpoints, such as the MicroShift API:
$ sudo firewall-cmd --permanent --zone=trusted --add-source=169.254.169.1
4.4. Using optional port settings
The MicroShift firewall service allows optional port settings.
Procedure
To add customized ports to your firewall configuration, use the following command syntax:
$ sudo firewall-cmd --permanent --zone=public --add-port=<port number>/<port protocol>
Table 4.2. Optional ports Port(s) Protocol(s) Description 80
TCP
HTTP port used to serve applications through the OpenShift Container Platform router.
443
TCP
HTTPS port used to serve applications through the OpenShift Container Platform router.
5353
UDP
mDNS service to respond for OpenShift Container Platform route mDNS hosts.
30000-32767
TCP
Port range reserved for NodePort services; can be used to expose applications on the LAN.
30000-32767
UDP
Port range reserved for NodePort services; can be used to expose applications on the LAN.
6443
TCP
HTTPS API port for the MicroShift API.
The following are examples of commands used when requiring external access through the firewall to services running on MicroShift, such as port 6443 for the API server, for example, ports 80 and 443 for applications exposed through the router.
Example command
Configuring a port for the MicroShift API server:
$ sudo firewall-cmd --permanent --zone=public --add-port=6443/tcp
To close unnecessary ports in your MicroShift instance, follow the procedure in "Closing unused or unnecessary ports to enhance network security".
Additional resources
4.5. Adding services to open ports
On a MicroShift instance, you can open services on ports by using the firewall-cmd
command.
Procedure
Optional: You can view all predefined services in firewalld by running the following command
$ sudo firewall-cmd --get-services
To open a service that you want on a default port, run the following example command:
$ sudo firewall-cmd --add-service=mdns
4.6. Allowing network traffic through the firewall
You can allow network traffic through the firewall by configuring the IP address range and inserting the DNS server to allow internal traffic from pods through the network gateway.
Procedure
Use one of the following commands to set the IP address range:
Configure the IP address range with default values by running the following command:
$ sudo firewall-offline-cmd --permanent --zone=trusted --add-source=10.42.0.0/16
Configure the IP address range with custom values by running the following command:
$ sudo firewall-offline-cmd --permanent --zone=trusted --add-source=<custom IP range>
To allow internal traffic from pods through the network gateway, run the following command:
$ sudo firewall-offline-cmd --permanent --zone=trusted --add-source=169.254.169.1
4.6.1. Applying firewall settings
To apply firewall settings, use the following one-step procedure:
Procedure
- After you have finished configuring network access through the firewall, run the following command to restart the firewall and apply the settings:
$ sudo firewall-cmd --reload
4.7. Verifying firewall settings
After you have restarted the firewall, you can verify your settings by listing them.
Procedure
To verify rules added in the default public zone, such as ports-related rules, run the following command:
$ sudo firewall-cmd --list-all
To verify rules added in the trusted zone, such as IP-range related rules, run the following command:
$ sudo firewall-cmd --zone=trusted --list-all
4.8. Overview of firewall ports when a service is exposed
Firewalld is often active when you run services on MicroShift. This can disrupt certain services on MicroShift because traffic to the ports might be blocked by the firewall. You must ensure that the necessary firewall ports are open if you want certain services to be accessible from outside the host. There are several options for opening your ports:
Services of the
NodePort
andLoadBalancer
type are automatically available with OVN-Kubernetes.In these cases, OVN-Kubernetes adds iptables rules so the traffic to the node IP address is delivered to the relevant ports. This is done using the PREROUTING rule chain and is then forwarded to the OVN-K to bypass the firewalld rules for local host ports and services. Iptables and firewalld are backed by nftables in RHEL 9. The nftables rules, which the iptables generates, always have priority over the rules that the firewalld generates.
Pods with the
HostPort
parameter settings are automatically available. This also includes therouter-default
pod, which uses ports 80 and 443.For
HostPort
pods, the CRI-O config sets up iptables DNAT (Destination Network Address Translation) to the pod’s IP address and port.
These methods function for clients whether they are on the same host or on a remote host. The iptables rules, which are added by OVN-Kubernetes and CRI-O, attach to the PREROUTING and OUTPUT chains. The local traffic goes through the OUTPUT chain with the interface set to the lo
type. The DNAT runs before it hits filler rules in the INPUT chain.
Because the MicroShift API server does not run in CRI-O, it is subject to the firewall configurations. You can open port 6443 in the firewall to access the API server in your MicroShift cluster.
4.9. Additional resources
4.10. Known firewall issue
-
To avoid breaking traffic flows with a firewall reload or restart, execute firewall commands before starting RHEL. The CNI driver in MicroShift makes use of iptable rules for some traffic flows, such as those using the NodePort service. The iptable rules are generated and inserted by the CNI driver, but are deleted when the firewall reloads or restarts. The absence of the iptable rules breaks traffic flows. If firewall commands have to be executed after MicroShift is running, manually restart
ovnkube-master
pod in theopenshift-ovn-kubernetes
namespace to reset the rules controlled by the CNI driver.
Chapter 5. Configuring network settings for fully disconnected hosts
Learn how to apply networking customization and settings to run MicroShift on fully disconnected hosts. A disconnected host should be the Red Hat Enterprise Linux (RHEL) operating system, versions 9.0+, whether real or virtual, that runs without network connectivity.
5.1. Preparing networking for fully disconnected hosts
Use the procedure that follows to start and run MicroShift clusters on devices running fully disconnected operating systems. A MicroShift host is considered fully disconnected if it has no external network connectivity.
Typically this means that the device does not have an attached network interface controller (NIC) to provide a subnet. These steps can also be completed on a host with a NIC that is removed after setup. You can also automate these steps on a host that does not have a NIC by using the %post
phase of a Kickstart file.
Configuring networking settings for disconnected environments is necessary because MicroShift requires a network device to support cluster communication. To meet this requirement, you must configure MicroShift networking settings to use the "fake" IP address you assign to the system loopback device during setup.
5.1.1. Procedure summary
To run MicroShift on a disconnected host, the following steps are required:
- Prepare the host
- Stop MicroShift if it is currently running and clean up changes the service has made to the network.
- Set a persistent hostname.
- Add a “fake” IP address on the loopback interface.
- Configure DNS to use the fake IP as local name server.
-
Add an entry for the hostname to
/etc/hosts
.
- Update the MicroShift configuration
-
Define the
nodeIP
parameter as the new loopback IP address. -
Set the
.node.hostnameOverride
parameter to the persistent hostname.
-
Define the
- For the changes to take effect
- Disable the default NIC if attached.
- Restart the host or device.
After starting, MicroShift runs using the loopback device for within-cluster communication.
5.2. Restoring MicroShift networking settings to default
You can remove networking customizations and return the network to default settings by stopping MicroShift and running a clean-up script.
Prerequisites
- RHEL 9 or newer.
- MicroShift 4.14 or newer.
- Access to the host CLI.
Procedure
Stop the MicroShift service by running the following command:
$ sudo systemctl stop microshift
Stop the
kubepods.slice
systemd unit by running the following command:$ sudo systemctl stop kubepods.slice
MicroShift installs a helper script to undo network changes made by OVN-K. Run the cleanup script by entering the following command:
$ sudo /usr/bin/microshift-cleanup-data --ovn
5.3. Configuring the networking settings for fully disconnected hosts
To configure the networking settings for running MicroShift on a fully disconnected host, you must prepare the host, update the networking configuration, then restart to apply the new settings. All commands are executed from the host CLI.
Prerequisites
- RHEL 9 or newer.
- MicroShift 4.14 or newer.
- Access to the host CLI.
- A valid IP address chosen to avoid both internal and potential future external IP conflicts when running MicroShift.
- MicroShift networking settings are set to defaults.
The following procedure is for use cases in which access to the MicroShift cluster is not required after devices are deployed in the field. There is no remote cluster access after the network connection is removed.
Procedure
Add a fake IP address to the loopback interface by running the following command:
$ IP="10.44.0.1" 1 $ sudo nmcli con add type loopback con-name stable-microshift ifname lo ip4 ${IP}/32
- 1
- The fake IP address used in this example is “10.44.0.1”.
NoteAny valid IP works if it avoids both internal MicroShift and potential future external IP conflicts. This can be any subnet that does not collide with the MicroShift node subnet or is be accessed by other services on the device.
Configure the DNS interface to use the local name server by setting modifying the settings to ignore automatic DNS and reset it to the local name server:
Bypass the automatic DNS by running the following command:
$ sudo nmcli conn modify stable-microshift ipv4.ignore-auto-dns yes
Point the DNS interface to use the local name server:
$ sudo nmcli conn modify stable-microshift ipv4.dns "10.44.1.1"
Get the hostname of the device by running the following command:
$ NAME="$(hostnamectl hostname)"
Add an entry for the hostname of the node in the
/etc/hosts
file by running the following command:$ echo "$IP $NAME" | sudo tee -a /etc/hosts >/dev/null
Update the MicroShift configuration file by adding the following YAML snippet to
/etc/microshift/config.yaml
:sudo tee /etc/microshift/config.yaml > /dev/null <<EOF node: hostnameOverride: $(echo $NAME) nodeIP: $(echo $IP) EOF
MicroShift is now ready to use the loopback device for cluster communications. Finish preparing the device for offline use.
- If the device currently has a NIC attached, disconnect the device from the network.
- Shut down the device and disconnect the NIC.
- Restart the device for the offline configuration to take effect.
Restart the MicroShift host to apply the configuration changes by running the following command:
$ sudo systemctl reboot 1
- 1
- This step restarts the cluster. Wait for the greenboot health check to report the system healthy before implementing verification.
Verification
At this point, network access to the MicroShift host has been severed. If you have access to the host terminal, you can use the host CLI to verify that the cluster has started in a stable state.
Verify that the MicroShift cluster is running by entering the following command:
$ export KUBECONFIG=/var/lib/microshift/resources/kubeadmin/kubeconfig $ sudo -E oc get pods -A
Example output
NAMESPACE NAME READY STATUS RESTARTS AGE kube-system csi-snapshot-controller-74d566564f-66n2f 1/1 Running 0 1m kube-system csi-snapshot-webhook-69bdff8879-xs6mb 1/1 Running 0 1m openshift-dns dns-default-dxglm 2/2 Running 0 1m openshift-dns node-resolver-dbf5v 1/1 Running 0 1m openshift-ingress router-default-8575d888d8-xmq9p 1/1 Running 0 1m openshift-ovn-kubernetes ovnkube-master-gcsx8 4/4 Running 1 1m openshift-ovn-kubernetes ovnkube-node-757mf 1/1 Running 1 1m openshift-service-ca service-ca-7d7c579f54-68jt4 1/1 Running 0 1m openshift-storage topolvm-controller-6d777f795b-bx22r 5/5 Running 0 1m openshift-storage topolvm-node-fcf8l 4/4 Running 0 1m