c12s-kubespray/docs/calico.md
Cristian Calin ec0c0d4a28
Calico enable support for eBPF (#7618)
* Calico: align manifests with upstream

* allow enabling typha prometheus metrics

* Calico: enable eBPF support

* manage the kubernetes-services-endpoint configmap

* Calico: document the use of eBPF dataplane

* Calico: improve checks before deployment

* enforce disabling kube-proxy when using eBPF dataplane
* ensure calico_version is supported
2021-06-07 04:58:39 -07:00

332 lines
11 KiB
Markdown

# Calico
Check if the calico-node container is running
```ShellSession
docker ps | grep calico
```
The **calicoctl.sh** is wrap script with configured access credentials for command calicoctl allows to check the status of the network workloads.
* Check the status of Calico nodes
```ShellSession
calicoctl.sh node status
```
* Show the configured network subnet for containers
```ShellSession
calicoctl.sh get ippool -o wide
```
* Show the workloads (ip addresses of containers and their location)
```ShellSession
calicoctl.sh get workloadEndpoint -o wide
```
and
```ShellSession
calicoctl.sh get hostEndpoint -o wide
```
## Configuration
### Optional : Define datastore type
The default datastore, Kubernetes API datastore is recommended for on-premises deployments, and supports only Kubernetes workloads; etcd is the best datastore for hybrid deployments.
Allowed values are `kdd` (default) and `etcd`.
Note: using kdd and more than 50 nodes, consider using the `typha` daemon to provide scaling.
To re-define you need to edit the inventory and add a group variable `calico_datastore`
```yml
calico_datastore: kdd
```
### Optional : Define network backend
In some cases you may want to define Calico network backend. Allowed values are `bird`, `vxlan` or `none`. Bird is a default value.
To re-define you need to edit the inventory and add a group variable `calico_network_backend`
```yml
calico_network_backend: none
```
### Optional : Define the default pool CIDRs
By default, `kube_pods_subnet` is used as the IP range CIDR for the default IP Pool, and `kube_pods_subnet_ipv6` for IPv6.
In some cases you may want to add several pools and not have them considered by Kubernetes as external (which means that they must be within or equal to the range defined in `kube_pods_subnet` and `kube_pods_subnet_ipv6` ), it starts with the default IP Pools of which IP range CIDRs can by defined in group_vars (k8s_cluster/k8s-net-calico.yml):
```ShellSession
calico_pool_cidr: 10.233.64.0/20
calico_pool_cidr_ipv6: fd85:ee78:d8a6:8607::1:0000/112
```
### Optional : BGP Peering with border routers
In some cases you may want to route the pods subnet and so NAT is not needed on the nodes.
For instance if you have a cluster spread on different locations and you want your pods to talk each other no matter where they are located.
The following variables need to be set:
`peer_with_router` to enable the peering with the datacenter's border router (default value: false).
you'll need to edit the inventory and add a hostvar `local_as` by node.
```ShellSession
node1 ansible_ssh_host=95.54.0.12 local_as=xxxxxx
```
### Optional : Defining BGP peers
Peers can be defined using the `peers` variable (see docs/calico_peer_example examples).
In order to define global peers, the `peers` variable can be defined in group_vars with the "scope" attribute of each global peer set to "global".
In order to define peers on a per node basis, the `peers` variable must be defined in hostvars.
NB: Ansible's `hash_behaviour` is by default set to "replace", thus defining both global and per node peers would end up with having only per node peers. If having both global and per node peers defined was meant to happen, global peers would have to be defined in hostvars for each host (as well as per node peers)
Since calico 3.4, Calico supports advertising Kubernetes service cluster IPs over BGP, just as it advertises pod IPs.
This can be enabled by setting the following variable as follow in group_vars (k8s_cluster/k8s-net-calico.yml)
```yml
calico_advertise_cluster_ips: true
```
Since calico 3.10, Calico supports advertising Kubernetes service ExternalIPs over BGP in addition to cluster IPs advertising.
This can be enabled by setting the following variable in group_vars (k8s_cluster/k8s-net-calico.yml)
```yml
calico_advertise_service_external_ips:
- x.x.x.x/24
- y.y.y.y/32
```
### Optional : Define global AS number
Optional parameter `global_as_num` defines Calico global AS number (`/calico/bgp/v1/global/as_num` etcd key).
It defaults to "64512".
### Optional : BGP Peering with route reflectors
At large scale you may want to disable full node-to-node mesh in order to
optimize your BGP topology and improve `calico-node` containers' start times.
To do so you can deploy BGP route reflectors and peer `calico-node` with them as
recommended here:
* <https://hub.docker.com/r/calico/routereflector/>
* <https://docs.projectcalico.org/v3.1/reference/private-cloud/l3-interconnect-fabric>
You need to edit your inventory and add:
* `calico_rr` group with nodes in it. `calico_rr` can be combined with
`kube_node` and/or `kube_control_plane`. `calico_rr` group also must be a child
group of `k8s_cluster` group.
* `cluster_id` by route reflector node/group (see details
[here](https://hub.docker.com/r/calico/routereflector/))
Here's an example of Kubespray inventory with standalone route reflectors:
```ini
[all]
rr0 ansible_ssh_host=10.210.1.10 ip=10.210.1.10
rr1 ansible_ssh_host=10.210.1.11 ip=10.210.1.11
node2 ansible_ssh_host=10.210.1.12 ip=10.210.1.12
node3 ansible_ssh_host=10.210.1.13 ip=10.210.1.13
node4 ansible_ssh_host=10.210.1.14 ip=10.210.1.14
node5 ansible_ssh_host=10.210.1.15 ip=10.210.1.15
[kube_control_plane]
node2
node3
[etcd]
node2
node3
node4
[kube_node]
node2
node3
node4
node5
[k8s_cluster:children]
kube_node
kube_control_plane
calico_rr
[calico_rr]
rr0
rr1
[rack0]
rr0
rr1
node2
node3
node4
node5
[rack0:vars]
cluster_id="1.0.0.1"
```
The inventory above will deploy the following topology assuming that calico's
`global_as_num` is set to `65400`:
![Image](figures/kubespray-calico-rr.png?raw=true)
### Optional : Define default endpoint to host action
By default Calico blocks traffic from endpoints to the host itself by using an iptables DROP action. When using it in kubernetes the action has to be changed to RETURN (default in kubespray) or ACCEPT (see <https://github.com/projectcalico/felix/issues/660> and <https://github.com/projectcalico/calicoctl/issues/1389).> Otherwise all network packets from pods (with hostNetwork=False) to services endpoints (with hostNetwork=True) within the same node are dropped.
To re-define default action please set the following variable in your inventory:
```yml
calico_endpoint_to_host_action: "ACCEPT"
```
## Optional : Define address on which Felix will respond to health requests
Since Calico 3.2.0, HealthCheck default behavior changed from listening on all interfaces to just listening on localhost.
To re-define health host please set the following variable in your inventory:
```yml
calico_healthhost: "0.0.0.0"
```
## Config encapsulation for cross server traffic
Calico supports two types of encapsulation: [VXLAN and IP in IP](https://docs.projectcalico.org/v3.11/networking/vxlan-ipip). VXLAN is supported in some environments where IP in IP is not (for example, Azure).
*IP in IP* and *VXLAN* is mutualy exclusive modes.
Configure Ip in Ip mode. Possible values is `Always`, `CrossSubnet`, `Never`.
```yml
calico_ipip_mode: 'Always'
```
Configure VXLAN mode. Possible values is `Always`, `CrossSubnet`, `Never`.
```yml
calico_vxlan_mode: 'Never'
```
If you use VXLAN mode, BGP networking is not required. You can disable BGP to reduce the moving parts in your cluster by `calico_network_backend: vxlan`
## Cloud providers configuration
Please refer to the official documentation, for example [GCE configuration](http://docs.projectcalico.org/v1.5/getting-started/docker/installation/gce) requires a security rule for calico ip-ip tunnels. Note, calico is always configured with ``calico_ipip_mode: Always`` if the cloud provider was defined.
### Optional : Ignore kernel's RPF check setting
By default the felix agent(calico-node) will abort if the Kernel RPF setting is not 'strict'. If you want Calico to ignore the Kernel setting:
```yml
calico_node_ignorelooserpf: true
```
Note that in OpenStack you must allow `ipip` traffic in your security groups,
otherwise you will experience timeouts.
To do this you must add a rule which allows it, for example:
### Optional : Felix configuration via extraenvs of calico node
Possible environment variable parameters for [configuring Felix](https://docs.projectcalico.org/reference/felix/configuration)
```yml
calico_node_extra_envs:
FELIX_DEVICEROUTESOURCEADDRESS: 172.17.0.1
```
```ShellSession
neutron security-group-rule-create --protocol 4 --direction egress k8s-a0tp4t
neutron security-group-rule-create --protocol 4 --direction igress k8s-a0tp4t
```
### Optional : Use Calico CNI host-local IPAM plugin
Calico currently supports two types of CNI IPAM plugins, `host-local` and `calico-ipam` (default).
To allow Calico to determine the subnet to use from the Kubernetes API based on the `Node.podCIDR` field, enable the following setting.
```yml
calico_ipam_host_local: true
```
Refer to Project Calico section [Using host-local IPAM](https://docs.projectcalico.org/reference/cni-plugin/configuration#using-host-local-ipam) for further information.
## eBPF Support
Calico supports eBPF for its data plane see [an introduction to the Calico eBPF Dataplane](https://www.projectcalico.org/introducing-the-calico-ebpf-dataplane/) for further information.
Note that it is advisable to always use the latest version of Calico when using the eBPF dataplane.
### Enabling eBPF support
To enable the eBPF dataplane support ensure you add the following to your inventory. Note that the `kube-proxy` is incompatible with running Calico in eBPF mode and the kube-proxy should be removed from the system.
```yaml
calico_bpf_enabled: true
kube_proxy_remove: true
```
### Cleaning up after kube-proxy
Calico node cannot clean up after kube-proxy has run in ipvs mode. If you are converting an existing cluster to eBPF you will need to ensure the `kube-proxy` DaemonSet is deleted and that ipvs rules are cleaned.
To check that kube-proxy was running in ipvs mode:
```ShellSession
# ipvsadm -l
```
To clean up any ipvs leftovers:
```ShellSession
# ipvsadm -C
```
### Calico access to the kube-api
Calico node, typha and kube-controllers need to be able to talk to the kubernetes API. Please reference the [Enabling eBPF Calico Docs](https://docs.projectcalico.org/maintenance/ebpf/enabling-bpf) for guidelines on how to do this.
Kubespray sets up the `kubernetes-services-endpoint` configmap based on the contents of the `loadbalancer_apiserver` inventory variable documented in [HA Mode](./ha-mode.md).
If no external loadbalancer is used, Calico eBPF can also use the localhost loadbalancer option. In this case Calico Automatic Host Endpoints need to be enabled to allow services like `coredns` and `metrics-server` to communicate with the kubernetes host endpoint. See [this blog post](https://www.projectcalico.org/securing-kubernetes-nodes-with-calico-automatic-host-endpoints/) on enabling automatic host endpoints.
```yaml
loadbalancer_apiserver_localhost: true
use_localhost_as_kubeapi_loadbalancer: true
```
### Tunneled versus Direct Server Return
By default Calico usese Tunneled service mode but it can use direct server return (DSR) in order to optimize the return path for a service.
To configure DSR:
```yaml
calico_bpf_service_mode: "DSR"
```
### eBPF Logging and Troubleshooting
In order to enable Calico eBPF mode logging:
```yaml
calico_bpf_log_level: "Debug"
```
To view the logs you need to use the `tc` command to read the kernel trace buffer:
```ShellSession
tc exec bpf debug
```
Please see [Calico eBPF troubleshooting guide](https://docs.projectcalico.org/maintenance/troubleshoot/troubleshoot-ebpf#ebpf-program-debug-logs).