c12s-kubespray/docs/cilium.md
biqiang Wu c681435432
Add switch cilium_enable_bandwidth_manager (#9441)
Signed-off-by: dcwbq <biqiang.wu@daocloud.io>

Signed-off-by: dcwbq <biqiang.wu@daocloud.io>
2022-10-28 03:08:31 -07:00

6.8 KiB
Raw Blame History

Cilium

Kube-proxy replacement with Cilium

Cilium can run without kube-proxy by setting cilium_kube_proxy_replacement to strict.

Without kube-proxy, cilium needs to know the address of the kube-apiserver and this must be set globally for all cilium components (agents and operators). Hence, in this configuration in Kubespray, Cilium will always contact the external loadbalancer (even from a node in the control plane) and if there is no external load balancer It will ignore any local load balancer deployed by Kubespray and only contacts the first master.

Cilium Operator

Unlike some operators, Cilium Operator does not exist for installation purposes.

The Cilium Operator is responsible for managing duties in the cluster which should logically be handled once for the entire cluster, rather than once for each node in the cluster.

Adding custom flags to the Cilium Operator

You can set additional cilium-operator container arguments using cilium_operator_custom_args. This is an advanced option, and you should only use it if you know what you are doing.

Accepts an array or a string.

cilium_operator_custom_args: ["--foo=bar", "--baz=qux"]

or

cilium_operator_custom_args: "--foo=bar"

You do not need to add a custom flag to enable debugging. Instead, feel free to use the CILIUM_DEBUG variable.

Adding extra volumes and mounting them

You can use cilium_operator_extra_volumes to add extra volumes to the Cilium Operator, and use cilium_operator_extra_volume_mounts to mount those volumes. This is an advanced option, and you should only use it if you know what you are doing.

cilium_operator_extra_volumes:
  - configMap:
      name: foo
    name: foo-mount-path

cilium_operator_extra_volume_mounts:
  - mountPath: /tmp/foo/bar
    name: foo-mount-path
    readOnly: true

Choose Cilium version

cilium_version: v1.12.1

Add variable to config

Use following variables:

Example:

cilium_config_extra_vars:
  enable-endpoint-routes: true

Change Identity Allocation Mode

Cilium assigns an identity for each endpoint. This identity is used to enforce basic connectivity between endpoints.

Cilium currently supports two different identity allocation modes:

  • "crd" stores identities in kubernetes as CRDs (custom resource definition).
    • These can be queried with kubectl get ciliumid
  • "kvstore" stores identities in an etcd kvstore.

Enable Transparent Encryption

Cilium supports the transparent encryption of Cilium-managed host traffic and traffic between Cilium-managed endpoints either using IPsec or Wireguard.

Wireguard option is only available in Cilium 1.10.0 and newer.

IPsec Encryption

For further information, make sure to check the official Cilium documentation.

To enable IPsec encryption, you just need to set three variables.

cilium_encryption_enabled: true
cilium_encryption_type: "ipsec"

The third variable is cilium_ipsec_key. You need to create a secret key string for this variable. Kubespray does not automate this process. Cilium documentation currently recommends creating a key using the following command:

echo "3 rfc4106(gcm(aes)) $(echo $(dd if=/dev/urandom count=20 bs=1 2> /dev/null | xxd -p -c 64)) 128"

Note that Kubespray handles secret creation. So you only need to pass the key as the cilium_ipsec_key variable.

Wireguard Encryption

For further information, make sure to check the official Cilium documentation.

To enable Wireguard encryption, you just need to set two variables.

cilium_encryption_enabled: true
cilium_encryption_type: "wireguard"

Kubespray currently supports Linux distributions with Wireguard Kernel mode on Linux 5.6 and newer.

Bandwidth Manager

Ciliums bandwidth manager supports the kubernetes.io/egress-bandwidth Pod annotation.

Bandwidth enforcement currently does not work in combination with L7 Cilium Network Policies. In case they select the Pod at egress, then the bandwidth enforcement will be disabled for those Pods.

Bandwidth Manager requires a v5.1.x or more recent Linux kernel.

For further information, make sure to check the official Cilium documentation.

To use this function, set the following parameters

cilium_enable_bandwidth_manager: true

Install Cilium Hubble

k8s-net-cilium.yml:

cilium_enable_hubble: true ## enable support hubble in cilium
cilium_hubble_install: true ## install hubble-relay, hubble-ui
cilium_hubble_tls_generate: true ## install hubble-certgen and generate certificates

To validate that Hubble UI is properly configured, set up a port forwarding for hubble-ui service:

kubectl port-forward -n kube-system svc/hubble-ui 12000:80

and then open http://localhost:12000/.

Hubble metrics

cilium_enable_hubble_metrics: true
cilium_hubble_metrics:
  - dns
  - drop
  - tcp
  - flow
  - icmp
  - http

More

Upgrade considerations

Rolling-restart timeouts

Cilium relies on the kernel's BPF support, which is extremely fast at runtime but incurs a compilation penalty on initialization and update.

As a result, the Cilium DaemonSet pods can take a significant time to start, which scales with the number of nodes and endpoints in your cluster.

As part of cluster.yml, this DaemonSet is restarted, and Kubespray's default timeouts for this operation are not appropriate for large clusters.

This means that you will likely want to update these timeouts to a value more in-line with your cluster's number of nodes and their respective CPU performance. This is configured by the following values:

# Configure how long to wait for the Cilium DaemonSet to be ready again
cilium_rolling_restart_wait_retries_count: 30
cilium_rolling_restart_wait_retries_delay_seconds: 10

The total time allowed (count * delay) should be at least ($number_of_nodes_in_cluster * $cilium_pod_start_time) for successful rolling updates. There are no drawbacks to making it higher and giving yourself a time buffer to accommodate transient slowdowns.

Note: To find the $cilium_pod_start_time for your cluster, you can simply restart a Cilium pod on a node of your choice and look at how long it takes for it to become ready.

Note 2: The default CPU requests/limits for Cilium pods is set to a very conservative 100m:500m which will likely yield very slow startup for Cilium pods. You probably want to significantly increase the CPU limit specifically if short bursts of CPU from Cilium are acceptable to you.