The following topics are discussed:
- Changing Configuration Options
- Securing The Setup
Before installing Weave Net, you should make sure the following ports are not blocked by your firewall: TCP 6783 and UDP 6783/6784. For more details, see the FAQ.
Weave Net can be installed onto your CNI-enabled Kubernetes cluster with a single command:
$ kubectl apply -f https://github.com/weaveworks/weave/releases/download/v2.8.1/weave-daemonset-k8s.yaml
Important: this configuration won’t enable encryption by default. If your data plane traffic isn’t secured that could allow malicious actors to access your pod network. Read on to see the alternatives.
After a few seconds, a Weave Net pod should be running on each Node and any further pods you create will be automatically attached to the Weave network.
Note: This command requires Kubernetes 1.4 or later, and we recommend your master node has at least two CPU cores.
CNI, the Container Network Interface, is a proposed standard for configuring network interfaces for Linux containers.
If you do not already have a CNI-enabled cluster, you can bootstrap one easily with kubeadm.
Alternatively, you can configure CNI yourself
Weave net depends on the portmap standard CNI plugin to support hostport functionality. Please ensure that portmap CNI plugin is installed (either by cluster installers like kubeadm or manually if you have configured CNI yourself) in
Note: If using the Weave CNI Plugin from a prior full install of Weave Net with your cluster, you must first uninstall it before applying the Weave-kube addon. Shut down Kubernetes, and on all nodes perform the following:
- Remove any separate provisions you may have made to run Weave at
Then relaunch Kubernetes and install the addon as described above.
Installing on GKE
Please note that you must grant the user the ability to create roles in Kubernetes before launching Weave Net. This is a prerequisite to use use role-based access control on GKE. Please see the GKE instructions.
Installing on EKS
EKS by default installs
amazon-vpc-cni-k8s CNI. Please follow below steps to use Weave-net as CNI
- create EKS cluster in any of prescribed way
amazon-vpc-cni-k8sdaemonset by running
kubectl delete ds aws-node -n kube-systemcommand
/etc/cni/net.d/10-aws.confliston each of the node
- edit instance security group to allow TCP 6783 and UDP 6783/6784 ports
- flush iptables nat, mangle, filter tables to clear any iptables configurations done by
- restart kube-proxy pods to reconfigure iptables
- apply weave-net daemoset by following above installation steps
- delete existing pods so they get recreated in Weave pod CIDR’s address-space.
Please note that while pods can connect to the Kubernetes API server for your cluser, API server will not be able to connect to the pods as API server nodes are not connected to Weave Net (they run on network managed by EKS).
Upgrading the Daemon Sets
The DaemonSet definition specifies Rolling Updates, so when you apply a new version Kubernetes will automatically restart the Weave Net pods one by one.
CPU and Memory Requirements
Kubernetes manages resources on each node, and only schedules pods to run on nodes that have enough free resources.
In the example manifests we request 10% of a CPU, which will be enough for small installations, but you should monitor how much it uses in your production clusters and adjust the requests to suit.
We do not recommend that you set a CPU or memory limit unless you are very experienced in such matters, because the implementation of limits in the Linux kernel sometimes behaves in a surprising way.
On a 1-node single-CPU cluster you may find Weave Net does not install at all, because other Kubernetes components already take 95% of the CPU. The best way to resolve this issue is to use machines with at least two CPU cores.
If a node runs out of CPU, memory or disk, Kubernetes may decide to evict one or more pods. It may choose to evict the Weave Net pod, which will disrupt pod network operations.
You can reduce the chance of eviction by changing the DaemonSet to have a much bigger request, and a limit of the same value.
This causes Kubernetes to apply a “guaranteed” rather than a “burstable” policy. However a similar request for disk space can not be made, and so please be aware of this issue and monitor your resources to ensure that they stay below 100%.
You can see when pods have been evicted via the
kubectl get events command
LASTSEEN COUNT NAME KIND TYPE REASON SOURCE MESSAGE 1m 1 mypod-09vkd Pod Warning Evicted kubelet, node-1 The node was low on resource: memory.
kubectl get pods
NAME READY STATUS RESTARTS AGE IP NODE mypod-09vkd 0/1 Evicted 0 1h <none> node-1
If you see this in your cluster, consider some of the above steps to reduce disruption.
Weave Net provides a network to connect all pods together, implementing the Kubernetes model.
Kubernetes uses the Container Network Interface (CNI) to join pods onto Weave Net.
Kubernetes implements many network features itself on top of the pod network. This includes Services, Service Discovery via DNS and Ingress into the cluster. WeaveDNS is disabled when using the Kubernetes addon.
Kubernetes Network Policies let you securely isolate pods from each other based on namespaces and labels. For more information on configuring network policies in Kubernetes see the walkthrough and the NetworkPolicy API object definition
Note: as of version 1.9 of Weave Net, the Network Policy
Controller allows all multicast traffic. Since a single multicast
address may be used by multiple pods, we cannot implement rules to
isolate them individually. You can turn this behaviour off (block
all multicast traffic) by adding
--allow-mcast=false as an
weave-npc in the YAML configuration.
Note: Since ingress traffic is masqueraded, it makes sense to use
ipBlock selector in an ingress rule only when limiting access to a Service
externalTrafficPolicy=Local or between Pods when
podIP is used
to access a Pod.
The first thing to check is whether Weave Net is up and
kubectl apply command you used to install it only
requests that it be downloaded and started; if anything goes wrong
at startup, those details will only be visible in the logs of
To check what is running:
$ kubectl get pods -n kube-system -l name=weave-net NAME READY STATUS RESTARTS AGE weave-net-1jkl6 2/2 Running 0 1d weave-net-bskbv 2/2 Running 0 1d weave-net-m4x1b 2/2 Running 0 1d
You should see one line for each node in your cluster; each line should have STATUS “Running”, and READY should be 2 out of 2. If you see a STATUS like “Error” or “CrashLoopBackoff”, look in the logs of the container with that status.
Reading the logs
Pick one of the pods from the list output by
kubectl get pods and
ask for the logs like this:
$ kubectl logs -n kube-system weave-net-1jkl6 weave
For easier viewing, pipe the output into a file, especially if it is long.
By default log level of
weave container is set to
info level. If you wish to see more detailed logs you can set the desired log level for the
--log-level flag through the
EXTRA_ARGS environment variable for the
weave container in the weave-net daemon set. Add environment variable as below.
containers: - command: - /home/weave/launch.sh name: weave env: - name: EXTRA_ARGS value: --log-level=debug
You may also set the
--log-level flag to
error if you
prefer to only log exceptional conditions.
Many Kubernetes network issues occur at a higher level than Weave Net. The Kubernetes Service Debugging Guide has a detailed step-by-step guide.
Once it is up and running, the status of Weave Net can be checked by running its CLI commands. This can be done in various ways:
1. Install the
weave script and run:
$ weave status Version: 2.0.1 (up to date; next check at 2017/07/10 13:49:29) Service: router Protocol: weave 1..2 Name: 42:8e:e8:c4:52:1b(host-0) Encryption: disabled PeerDiscovery: enabled Targets: 3 Connections: 3 (2 established, 1 failed) Peers: 3 (with 6 established connections) TrustedSubnets: none Service: ipam Status: ready Range: 10.32.0.0/12 DefaultSubnet: 10.32.0.0/12
2. If you don’t want to install additional software onto your hosts, run via
kubectl commands, which produce the exact same outcome as the previous example:
### Identify the Weave Net pods: $ kubectl get pods -n kube-system -l name=weave-net -o wide NAME READY STATUS RESTARTS AGE IP NODE weave-net-1jkl6 2/2 Running 0 1d 10.128.0.4 host-0 weave-net-bskbv 2/2 Running 0 1d 10.128.0.5 host-1 weave-net-m4x1b 2/2 Running 0 1d 10.128.0.6 host-2
The above shows all Weave Net pods available in your cluster. You can see Kubernetes has deployed one Weave Net pod per host, in order to interconnect all hosts.
You then need to:
- choose which pod you want to run your command from (in most cases it doesn’t matter
which one you pick so just pick the first one, e.g. pod
kubectl execto run the
- specify the absolute path
--localbecause it’s running inside a container
$ kubectl exec -n kube-system weave-net-1jkl6 -c weave -- /home/weave/weave --local status Version: 2.0.1 (up to date; next check at 2017/07/10 13:49:29) Service: router Protocol: weave 1..2 Name: 42:8e:e8:c4:52:1b(host-0) Encryption: disabled PeerDiscovery: enabled Targets: 3 Connections: 3 (2 established, 1 failed) Peers: 3 (with 6 established connections) TrustedSubnets: none Service: ipam Status: ready Range: 10.32.0.0/12 DefaultSubnet: 10.32.0.0/12
Troubleshooting Blocked Connections
If you suspect that legitimate traffic is being blocked by the Weave Network Policy Controller, the first thing to do is check the
weave-npc container’s logs.
To do this, first you have to find the name of the Weave Net pod running on the relevant host:
$ kubectl get pods -n kube-system -o wide | grep weave-net weave-net-08y45 2/2 Running 0 1m 10.128.0.2 host1 weave-net-2zuhy 2/2 Running 0 1m 10.128.0.4 host3 weave-net-oai50 2/2 Running 0 1m 10.128.0.3 host2
Select the relevant container, for example, if you want to look at host2 then pick
weave-net-oai50 and run:
$ kubectl logs <weave-pod-name-as-above> -n kube-system weave-npc
When the Weave Network Policy Controller blocks a connection, it logs the following details about it:
- protocol used,
- source IP and port,
- destination IP and port,
as per the below example:
TCP connection from 10.32.0.7:56648 to 10.32.0.11:80 blocked by Weave NPC. UDP connection from 10.32.0.7:56648 to 10.32.0.11:80 blocked by Weave NPC.
Things to watch out for
- Weave Net does not work on hosts running iptables 1.8 or above, only with 1.6. Track this via issue #3465
- Don’t turn on
--masquerade-allon kube-proxy: this will change the source address of every pod-to-pod conversation which will make it impossible to correctly enforce network policies that restrict which pods can talk.
- If you do set the
--cluster-cidroption on kube-proxy, make sure it matches the
IPALLOC_RANGEgiven to Weave Net (see below).
- IP forwarding must be enabled on each node, in order for pods to
access Kubernetes services or other IP addresses on another
network. Check this with
sysctl net.ipv4.ip_forward; the result should be
1. (Be aware that there can be security implications of enabling IP forwarding).
- Weave Net can be run on minikube v0.28 or later with the default CNI config shipped with minikube being disabled. See #3124 for more details.
- Weave Net has a problem with containerd versions 1.6.0 through 1.6.4. See Troubleshooting FailedCreatePodSandBox errors below.
Troubleshooting FailedCreatePodSandBox errors
If your Kubernetes cluster uses the
containerd runtime (versions 1.6.0 through 1.6.4), Weave Net will not be able to allocate IP addresses to pods. Your pods, except the ones that use HostNetworking, will be stuck at
You can examine any pod so affected by running
kubectl describe, for example:
$ kubectl describe pod -n kube-system coredns-78fcd69978-dbxs9
The events section will show repeated errors like the following:
Warning FailedCreatePodSandBox 3m6s kubelet Failed to create pod sandbox: rpc error: code = Unknown desc = failed to setup network for sandbox "09a23f79c96333b9f54e12df54e817837c8021cbaa32bdfeefbe2a1fb215d9ef": plugin type="weave-net" name="weave" failed (add): unable to allocate IP address: Post "http://127.0.0.1:6784/ip/09a23f79c96333b9f54e12df54e817837c8021cbaa32bdfeefbe2a1fb215d9ef": dial tcp 127.0.0.1:6784: connect: connection refused
You can verify that you are running an affected version of containerd by using the following:
$ kubectl get nodes -o wide NAME STATUS ROLES AGE VERSION INTERNAL-IP EXTERNAL-IP OS-IMAGE KERNEL-VERSION CONTAINER-RUNTIME host-1 Ready control-plane,master 13m v1.22.10 172.21.107.129 <none> Debian GNU/Linux 11 (bullseye) 5.10.0-14-amd64 containerd://1.6.4
The last column shows the container runtime and version.
Alternatively, you can run:
$ containerd -v containerd containerd.io 1.6.4 212e8b6fa2f44b9c21b2798135fc6fb7c53efc16
The problem can be solved by upgrading containerd to v1.6.5 or above. For example, on Debian Linux, using the docker official repositories, you can use:
sudo apt install containerd.io=1.6.6-1
The problem occurs because of a behaviour change in cni v1.1.0, which caused a regression issue in Weave. It was corrected in cni v1.1.1. Containerd 1.6.5 onwards uses cni 1.1.1 and above.
Changing Configuration Options
Manually editing the YAML file
You can manually edit the YAML file downloaded from our releases page,
- additional arguments may be supplied to the Weave router process by adding them to the
command:array in the YAML file,
- additional parameters can be set via the environment variables listed above; these can be inserted into the YAML file like this:
containers: - name: weave env: - name: IPALLOC_RANGE value: 10.0.0.0/16
The list of variables you can set is:
CHECKPOINT_DISABLE- if set to 1, disable checking for new Weave Net versions (default is blank, i.e. check is enabled)
CONN_LIMIT- soft limit on the number of connections between peers. Defaults to 200.
HAIRPIN_MODE- Weave Net defaults to enabling hairpin on the bridge side of the
vethpair for containers attached. If you need to disable hairpin, e.g. your kernel is one of those that can panic if hairpin is enabled, then you can disable it by setting
IPALLOC_RANGE- the range of IP addresses used by Weave Net and the subnet they are placed in (CIDR format; default
EXPECT_NPC- set to 0 to disable Network Policy Controller (default is on)
KUBE_PEERS- list of addresses of peers in the Kubernetes cluster (default is to fetch the list from the api-server)
IPALLOC_INIT- set the initialization mode of the IP Address Manager (defaults to consensus amongst the
WEAVE_EXPOSE_IP- set the IP address used as a gateway from the Weave network to the host network - this is useful if you are configuring the addon as a static pod.
WEAVE_METRICS_ADDR- address and port that the Weave Net daemon will serve Prometheus-style metrics on (defaults to 0.0.0.0:6782)
WEAVE_PASSWORD- shared key to use during session key generation to encrypt traffic between peers.
WEAVE_STATUS_ADDR- address and port that the Weave Net daemon will serve status requests on (defaults to disabled).
WEAVE_MTU- Weave Net defaults to 1376 bytes, but you can set a smaller size if your underlying network has a tighter limit, or set a larger size for better performance if your network supports jumbo frames - see here for more details.
NO_MASQ_LOCAL- set to 0 to disable preserving the client source IP address when accessing Service annotated with
service.spec.externalTrafficPolicy=Local. This feature works only with Weave IPAM (default).
IPTABLES_BACKEND- set to
Securing the Setup
You should set the environment variable
WEAVE_PASSWORD as stated in the previous section to enable the data plane encryption;
this is a recommended option in case you cannot be sure about the security of the fabric between your nodes.
A different option is to use
trusted-subnets and whitelist only the subnets that host your k8s nodes. Mind that depending on your circumstances that might allow a malicious container running in your cluster to access the weave dataplane, still.
Read on the Securing Connections Across Untrusted Networks document to see the alternatives.
To improve security drop
CAP_NET_RAW from pod capabilities: by default pods can forge packets from anywhere on the network, which enables attacks such as DNS spoofing.
securityContext: capabilities: drop: ["NET_RAW"]