Configuring the Calico Enterprise CNI plugins
- Operator
- Manifest
The Calico Enterprise CNI plugins do not need to be configured directly when installed by the operator. For a complete operator configuration reference, see the installation API reference documentation.
The Calico Enterprise CNI plugin is configured through the standard CNI configuration mechanism
A minimal configuration file that uses Calico Enterprise for networking and IPAM looks like this
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam"
}
}
If the cnx-node
container on a node registered with a NODENAME
other than the node hostname, the CNI plugin on this node must be configured with the same nodename
:
{
"name": "any_name",
"nodename": "<NODENAME>",
"type": "calico",
"ipam": {
"type": "calico-ipam"
}
}
Additional configuration can be added as detailed below.
Generic
Datastore type
The Calico Enterprise CNI plugin supports the following datastore:
datastore_type
(kubernetes)
Logging
Logging is always to stderr
. Logs are also written to /var/log/calico/cni/cni.log
on each host by default.
Logging can be configured using the following options in the netconf.
Option name | Default | Description |
---|---|---|
log_level | INFO | Logging level. Allowed levels are ERROR , WARNING , INFO , and DEBUG . |
log_file_path | /var/log/calico/cni/cni.log | Location on each host to write CNI log files to. Logging to file can be disabled by removing this option. |
log_file_max_size | 100 | Max file size in MB log files can reach before they are rotated. |
log_file_max_age | 30 | Max age in days that old log files will be kept on the host before they are removed. |
log_file_max_count | 10 | Max number of rotated log files allowed on the host before they are cleaned up. |
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"log_level": "DEBUG",
"log_file_path": "/var/log/calico/cni/cni.log",
"ipam": {
"type": "calico-ipam"
}
}
IPAM
When using Calico Enterprise IPAM, the following flags determine what IP addresses should be assigned. NOTE: These flags are strings and not boolean values.
assign_ipv4
(default:"true"
)assign_ipv6
(default:"false"
)
A specific IP address can be chosen by using CNI_ARGS
and setting IP
to the desired value.
By default, Calico Enterprise IPAM will assign IP addresses from all the available IP pools.
Optionally, the list of possible IPv4 and IPv6 pools can also be specified via the following properties:
ipv4_pools
: An array of CIDR strings or pool names. (e.g.,"ipv4_pools": ["10.0.0.0/24", "20.0.0.0/16", "default-ipv4-ippool"]
)ipv6_pools
: An array of CIDR strings or pool names. (e.g.,"ipv6_pools": ["2001:db8::1/120", "namedpool"]
)
Example CNI config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam",
"assign_ipv4": "true",
"assign_ipv6": "true",
"ipv4_pools": ["10.0.0.0/24", "20.0.0.0/16", "default-ipv4-ippool"],
"ipv6_pools": ["2001:db8::1/120", "default-ipv6-ippool"]
}
}
ipv6_pools
will be respected only when assign_ipv6
is set to "true"
.
Any IP pools specified in the CNI config must have already been created. It is an error to specify IP pools in the config that do not exist.
Container settings
The following options allow configuration of settings within the container namespace.
- allow_ip_forwarding (default is
false
)
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam"
},
"container_settings": {
"allow_ip_forwarding": true
}
}
Readiness Gates
The following option makes CNI plugin wait for specified endpoint(s) to be ready before configuring pod networking.
readiness_gates
This is an optional property that takes an array of URLs. Each URL specified will be polled for readiness and pod networking will continue startup once all readiness_gates are ready.
Example CNI config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam"
},
"readiness_gates": ["http://localhost:9099/readiness", "http://localhost:8888/status"]
}
Kubernetes specific
When using the Calico Enterprise CNI plugin with Kubernetes, the plugin must be able to access the Kubernetes API server to find the labels assigned to the Kubernetes pods. The recommended way to configure access is through a kubeconfig
file specified in the kubernetes
section of the network config. e.g.
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"kubernetes": {
"kubeconfig": "/path/to/kubeconfig"
},
"ipam": {
"type": "calico-ipam"
}
}
As a convenience, the API location can also be configured directly, e.g.
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"kubernetes": {
"k8s_api_root": "http://127.0.0.1:8080"
},
"ipam": {
"type": "calico-ipam"
}
}
Enabling Kubernetes policy
If you wish to use the Kubernetes NetworkPolicy
resource then you must set a policy type in the network config.
There is a single supported policy type, k8s
. When set,
you must also run tigera/kube-controllers
with the policy, profile, and workloadendpoint controllers enabled.
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"policy": {
"type": "k8s"
},
"kubernetes": {
"kubeconfig": "/path/to/kubeconfig"
},
"ipam": {
"type": "calico-ipam"
}
}
When using type: k8s
, the Calico Enterprise CNI plugin requires read-only Kubernetes API access to the Pods
resource in all namespaces.
Enabling policy setup timeout
The Calico Enterprise CNI plugin can be configured to prevent new pods from starting their containers until one of the following conditions occurs:
- The pod's policy has finished being programmed.
- A specified amount of time has elapsed.
By enabling this feature, you can avoid errors that can occur when a pod tries to start before the pod's policy is programmed by its host.
- Operator
- Manifest
The policy setup timeout can be configured by setting the linuxPolicySetupTimeoutSeconds
field in the calicoNetwork spec of the default operator.tigera.io/v1/installation
resource.
The following example configures the CNI to delay a pod from starting its containers for up to 10 seconds, or until the pod's dataplane has been programmed:
kind: Installation
apiVersion: operator.tigera.io/v1
metadata:
name: default
spec:
calicoNetwork:
linuxPolicySetupTimeoutSeconds: 10
The policy setup timeout can be configured by setting the policy_setup_timeout_seconds
option in the CNI config.
Example CNI config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"policy_setup_timeout_seconds": 10,
"ipam": {
"type": "calico-ipam"
}
}
The Calico Enterprise CNI plugin reads Felix's endpoint-status
directory to determine when the dataplane has been programmed for a pod.
If left unset, the Calico Enterprise CNI plugin will look for the directory at /var/run/calico/endpoint-status
. The path /var/run/calico
is commonly mounted to the Calico Enterprise DaemonSet, meaning it can be written to by the Felix container, and read by the (host-namespace) Calico Enterprise CNI plugin.
To enable the endpoint-status
directory, and adjust which directory of the Felix container it is written to, the endpointStatusPathPrefix
option must be configured for Felix.
To adjust where the Calico Enterprise CNI plugin looks for the endpoint-status
directory in the host filesystem, you must set the endpoint_status_dir
option.
Example CNI config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"policy_setup_timeout_seconds": 10,
"endpoint_status_dir": "/path/to/endpoint-status",
"ipam": {
"type": "calico-ipam"
}
}
IPAM
Using host-local IPAM
Calico can be configured to use host-local IPAM instead of the default calico-ipam
. Host
local IPAM uses a pre-determined CIDR per-host, and stores allocations locally on each node. This is in contrast to Calico IPAM, which dynamically
allocates blocks of addresses and single addresses alike in response to cluster needs.
Host local IPAM is generally only used on clusters where integration with the Kubernetes route controller is necessary. Note that some Calico features - such as the ability to request a specific address or pool for a pod - require Calico IPAM to function, and will not work with host-local IPAM enabled.
- Operator
- Manifest
The host-local
IPAM plugin can be configured by setting the Spec.CNI.IPAM.Plugin
field to HostLocal
on the operator.tigera.io/Installation API.
Calico will use the host-local
IPAM plugin to allocate IPv4 addresses from the node's IPv4 pod CIDR if there is an IPv4 pool configured in Spec.IPPools
, and an IPv6 address from the node's IPv6 pod CIDR if
there is an IPv6 pool configured in Spec.IPPools
.
The following example configures Calico to assign dual-stack IPs to pods using the host-local IPAM plugin.
kind: Installation
apiVersion: operator.tigera.io/v1
metadata:
name: default
spec:
calicoNetwork:
ipPools:
- cidr: 192.168.0.0/16
- cidr: 2001:db8::/64
cni:
type: Calico
ipam:
type: HostLocal
When using the CNI host-local
IPAM plugin, two special values - usePodCidr
and usePodCidrIPv6
- are allowed for the subnet field (either at the top-level, or in a "range"). This tells the plugin to determine the subnet to use from the Kubernetes API based on the Node.podCIDR field. Calico Enterprise does not use the gateway
field of a range so that field is not required and it will be ignored if present.
usePodCidr
and usePodCidrIPv6
can only be used as the value of the subnet
field, it cannot be used in
rangeStart
or rangeEnd
so those values are not useful if subnet
is set to usePodCidr
.
Calico Enterprise supports the host-local IPAM plugin's routes
field as follows:
-
If there is no
routes
field, Calico Enterprise will install a default0.0.0.0/0
, and/or::/0
route into the pod (depending on whether the pod has an IPv4 and/or IPv6 address). -
If there is a
routes
field then Calico Enterprise will program only the routes in the routes field into the pod. Since Calico Enterprise implements a point-to-point link into the pod, thegw
field is not required and it will be ignored if present. All routes that Calico Enterprise installs will have Calico Enterprise's link-local IP as the next hop.
Calico Enterprise CNI plugin configuration:
node_name
- The node name to use when looking up the CIDR value (defaults to current hostname)
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"kubernetes": {
"kubeconfig": "/path/to/kubeconfig",
"node_name": "node-name-in-k8s"
},
"ipam": {
"type": "host-local",
"ranges": [[{ "subnet": "usePodCidr" }], [{ "subnet": "usePodCidrIPv6" }]],
"routes": [{ "dst": "0.0.0.0/0" }, { "dst": "2001:db8::/96" }]
}
}
When making use of the usePodCidr
or usePodCidrIPv6
options, the Calico Enterprise CNI plugin requires read-only Kubernetes API access to the Nodes
resource.
Configuring node and typha
When using host-local
IPAM with the Kubernetes API datastore, you must configure both cnx-node and the Typha deployment to use the Node.podCIDR
field by setting the environment variable USE_POD_CIDR=true
in each.
Using Kubernetes annotations
Specifying IP pools on a per-namespace or per-pod basis
In addition to specifying IP pools in the CNI config as discussed above, Calico Enterprise IPAM supports specifying IP pools per-namespace or per-pod using the following Kubernetes annotations.
-
cni.projectcalico.org/ipv4pools
: A list of configured IPv4 Pools from which to choose an address for the pod.Example:
annotations:
'cni.projectcalico.org/ipv4pools': '["default-ipv4-ippool"]' -
cni.projectcalico.org/ipv6pools
: A list of configured IPv6 Pools from which to choose an address for the pod.Example:
annotations:
'cni.projectcalico.org/ipv6pools': '["2001:db8::1/120"]'
If provided, these IP pools will override any IP pools specified in the CNI config.
This requires the IP pools to exist before ipv4pools
or
ipv6pools
annotations are used. Requesting a subset of an IP pool
is not supported. IP pools requested in the annotations must exactly
match a configured IPPool resource.
The Calico Enterprise CNI plugin supports specifying an annotation per namespace. If both the namespace and the pod have this annotation, the pod information will be used. Otherwise, if only the namespace has the annotation the annotation of the namespace will be used for each pod in it.
Requesting a specific IP address
You can also request a specific IP address through Kubernetes annotations with Calico Enterprise IPAM. There are two annotations to request a specific IP address:
-
cni.projectcalico.org/ipAddrs
: A list of IPv4 and/or IPv6 addresses to assign to the Pod. The requested IP addresses will be assigned from Calico Enterprise IPAM and must exist within a configured IP pool.Example:
annotations:
'cni.projectcalico.org/ipAddrs': '["192.168.0.1"]' -
cni.projectcalico.org/ipAddrsNoIpam
: A list of IPv4 and/or IPv6 addresses to assign to the Pod, bypassing IPAM. Any IP conflicts and routing have to be taken care of manually or by some other system. Calico Enterprise will only distribute routes to a Pod if its IP address falls within a Calico Enterprise IP pool using BGP mode. Calico will not distribute ipAddrsNoIpam routes when operating in VXLAN mode. If you assign an IP address that is not in a Calico Enterprise IP pool or if its IP address falls within a Calico Enterprise IP pool that uses VXLAN encapsulation, you must ensure that routing to that IP address is taken care of through another mechanism.Example:
annotations:
'cni.projectcalico.org/ipAddrsNoIpam': '["10.0.0.1"]'The ipAddrsNoIpam feature is disabled by default. It can be enabled in the feature_control section of the CNI network config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam"
},
"feature_control": {
"ip_addrs_no_ipam": true
}
}cautionThis feature allows for the bypassing of network policy via IP spoofing. Users should make sure the proper admission control is in place to prevent users from selecting arbitrary IP addresses.
- The
ipAddrs
andipAddrsNoIpam
annotations can't be used together. - You can only specify one IPv4/IPv6 or one IPv4 and one IPv6 address with these annotations.
- When
ipAddrs
oripAddrsNoIpam
is used withipv4pools
oripv6pools
,ipAddrs
/ipAddrsNoIpam
take priority.
Requesting a floating IP
You can request a floating IP address for a pod through Kubernetes annotations with Calico Enterprise.
The specified address must belong to an IP Pool for advertisement to work properly.
-
cni.projectcalico.org/floatingIPs
: A list of floating IPs which will be assigned to the pod's workload endpoint.Example:
annotations:
'cni.projectcalico.org/floatingIPs': '["10.0.0.1"]'The floatingIPs feature is disabled by default. It can be enabled in the feature_control section of the CNI network config:
{
"name": "any_name",
"cniVersion": "0.1.0",
"type": "calico",
"ipam": {
"type": "calico-ipam"
},
"feature_control": {
"floating_ips": true
}
}cautionThis feature can allow pods to receive traffic which may not have been intended for that pod. Users should make sure the proper admission control is in place to prevent users from selecting arbitrary floating IP addresses.
Using IP pools node selectors
Nodes will only assign workload addresses from IP pools which select them. By
default, IP pools select all nodes, but this can be configured using the
nodeSelector
field. Check out the IP pool resource document
for more details.
Example:
-
Create (or update) an IP pool that only allocates IPs for nodes where it contains a label
rack=0
.kubectl create -f -<<EOF
apiVersion: projectcalico.org/v3
kind: IPPool
metadata:
name: rack-0-ippool
spec:
cidr: 192.168.0.0/24
ipipMode: Always
natOutgoing: true
nodeSelector: rack == "0"
EOF -
Label a node with
rack=0
.kubectl label nodes kube-node-0 rack=0
Check out the usage guide on assign IP addresses based on topology
for a full example.
CNI network configuration lists
The CNI 0.3.0 spec supports "chaining" multiple CNI plugins together. Calico Enterprise supports the following Kubernetes CNI plugins, which are enabled by default. Although chaining other CNI plugins may work, we support only the following tested CNI plugins.
Port mapping plugin
Calico Enterprise is required to implement Kubernetes host port functionality and is enabled by default.
Be aware of the following portmap plugin CNI issue where draining nodes may take a long time with a cluster of 100+ nodes and 4000+ services.
To disable it, remove the portmap section from the CNI network configuration in the Calico Enterprise manifests.
{
"type": "portmap",
"snat": true,
"capabilities": { "portMappings": true }
}
Traffic shaping plugin
The traffic shaping Kubernetes CNI plugin supports pod ingress and egress traffic shaping. This bandwidth management technique delays the flow of certain types of network packets to ensure network performance for higher priority applications. It is enabled by default.
You can add the kubernetes.io/ingress-bandwidth
and kubernetes.io/egress-bandwidth
annotations to your pod. For example, the following sets a 1 megabit-per-second connection for ingress and egress traffic.
apiVersion: v1
kind: Pod
metadata:
annotations:
kubernetes.io/ingress-bandwidth: 1M
kubernetes.io/egress-bandwidth: 1M
...
To disable it, remove the bandwidth section from the CNI network configuration in the Calico Enterprise manifests.
{
"type": "bandwidth",
"capabilities": { "bandwidth": true }
}
Order of precedence
If more than one of these methods are used for IP address assignment, they will take on the following precedence, 1 being the highest:
- Kubernetes annotations
- CNI configuration
- IP pool node selectors
Calico Enterprise IPAM will not reassign IP addresses to workloads that are already running. To update running workloads with IP addresses from a newly configured IP pool, they must be recreated. We recommend doing this before going into production or during a maintenance window.
Specify num_queues for veth interfaces
num_rx_queues
and num_tx_queues
can be set using the num_queues
option in the CNI configuration. Default: 1
For example:
{
"num_queues": 3
}