Configuring BGP Peers

When kube-router is used to provide pod-to-pod networking, BGP is used to exchange routes across the nodes. Kube-router provides flexible networking models to support different deployments (public vs private cloud, routable vs non-routable pod IPs, service IPs, etc.).

Peering Within The Cluster

Full Node-To-Node Mesh

This is the default mode. All nodes in the clusters form iBGP peering relationships with rest of the nodes forming a full node-to-node mesh. Each node advertise the pod CIDR allocated to the nodes with its peers (the rest of the nodes in the cluster). There is no configuration required in this mode. All the nodes in the cluster are associated with the private ASN 64512 implicitly (which can be configured with --cluster-asn flag) and users are transparent to use of iBGP. This mode is suitable in public cloud environments or small cluster deployments.

Node-To-Node Peering Without Full Mesh

This model is used to support more than a single AS per cluster to allow for an AS per rack or an AS per node. Nodes in the cluster do not form full node-to-node meshes. Users have to explicitly select this mode by specifying --nodes-full-mesh=false when launching kube-router. In this mode kube-router expects each node will be configured with an ASN number from the node's API object annotations. Kube-router will use the node's kube-router.io/node.asn annotation value as the ASN number for the node.

Users can annotate node objects with the following command:

kubectl annotate node <kube-node> "kube-router.io/node.asn=64512"

Only nodes within same ASN form full mesh. Two nodes with different ASNs never get peered.

Route-Reflector setup Without Full Mesh

This model supports the common scheme of using a Route Reflector Server node to concentrate peering from client peers. This has the big advantage of not needing full mesh, and will scale better. In this mode kube-router expects each node is configured either in Route Reflector server mode or in Route Reflector client mode. This is done with node kube-router.io/rr.server=ClusterID, kube-router.io/rr.client=ClusterId respectively. In this mode each route reflector client will only peer with route reflector servers. Each route reflector server will only peer with other route reflector servers and with route reflector clients enabling reflection.

Users can annotate node objects with the following command for Route Reflector server mode:

kubectl annotate node <kube-node> "kube-router.io/rr.server=42"

and for Route Reflector client mode:

kubectl annotate node <kube-node> "kube-router.io/rr.client=42"

Only nodes with the same ClusterID in client and server mode will peer together.

When joining new nodes to the cluster, remember to annotate them with kube-router.io/rr.client=42, and then restart kube-router on the new nodes and the route reflector server nodes to let them successfully read the annotations and peer with each other.

Peering Outside The Cluster

Global External BGP Peers

An optional global BGP peer can be configured by specifying the parameters: --peer-router-asns and --peer-router-ips. When configured each node in the cluster forms a peer relationship with specified global peer. Pod CIDR and Cluster IPs get advertised to the global BGP peer. For redundancy, you can also configure more than one peer router by specifying a slice of BGP peers.

For example:

--peer-router-ips="192.168.1.99,192.168.1.100"
--peer-router-asns=65000,65000

Node Specific External BGP Peers

Alternatively, each node can be configured with one or more node specific BGP peers. Information regarding node specific BGP peer is read from node API object annotations:

kube-router.io/peer.ips
kube-router.io/peer.asns

For example, users can annotate node object with below commands:

kubectl annotate node <kube-node> "kube-router.io/peer.ips=192.168.1.99,192.168.1.100"
kubectl annotate node <kube-node> "kube-router.io/peer.asns=65000,65000"

AS Path Prepending

For traffic shaping purposes, you may want to prepend the AS path announced to peers. This can be accomplished on a per-node basis with annotations:

kube-router.io/path-prepend.as
kube-router.io/path-prepend.repeat-n

If you wanted to prepend all routes from a particular node with the AS 65000 five times, you would run the following commands:

kubectl annotate node <kube-node> "kube-router.io/path-prepend.as=65000"
kubectl annotate node <kube-node> "kube-router.io/path-prepend.repeat-n=5"

BGP Peer Local IP configuration

In some setups it might be desirable to set a local IP address used for connecting external BGP peers. This can be accomplished on nodes with annotations:

kube-router.io/peer.localips

If set, this must be a list with a local IP address for each peer, or left empty to use nodeIP.

Example:

kubectl annotate node <kube-node> "kube-router.io/peer.localips=10.1.1.1,10.1.1.2"

This will instruct kube-router to use IP 10.1.1.1 for first BGP peer as a local address, and use 10.1.1.2for the second.

BGP Peer Password Authentication

The examples above have assumed there is no password authentication with BGP peer routers. If you need to use a password for peering, you can use the --peer-router-passwords command-line option, the kube-router.io/peer.passwords node annotation, or the --peer-router-passwords-file command-line option.

Base64 Encoding Passwords

To ensure passwords are easily parsed, but not easily read by human eyes, kube-router requires that they are encoded as base64.

On a Linux or MacOS system you can encode your passwords on the command line:

$ printf "SecurePassword" | base64
U2VjdXJlUGFzc3dvcmQ=

Password Configuration Examples

In this CLI flag example the first router (192.168.1.99) uses a password, while the second (192.168.1.100) does not.

--peer-router-ips="192.168.1.99,192.168.1.100"
--peer-router-asns="65000,65000"
--peer-router-passwords="U2VjdXJlUGFzc3dvcmQK,"

Note the comma indicating the end of the first password.

Here's the same example but configured as node annotations:

kubectl annotate node <kube-node> "kube-router.io/peer.ips=192.168.1.99,192.168.1.100"
kubectl annotate node <kube-node> "kube-router.io/peer.asns=65000,65000"
kubectl annotate node <kube-node> "kube-router.io/peer.passwords=U2VjdXJlUGFzc3dvcmQK,"

Finally, to include peer passwords as a file you would run kube-router with the following option:

--peer-router-ips="192.168.1.99,192.168.1.100"
--peer-router-asns="65000,65000"
--peer-router-passwords-file="/etc/kube-router/bgp-passwords.conf"

The password file, closely follows the syntax of the command-line and node annotation options. Here, the first peer IP (192.168.1.99) would be configured with a password, while the second would not.

U2VjdXJlUGFzc3dvcmQK,

Note, complex parsing is not done on this file, please do not include any content other than the passwords on a single line in this file.

BGP Communities

Global peers support the addition of BGP communities via node annotations. Node annotations can be formulated either as:

a single 32-bit integer
two 16-bit integers separated by a colon (:)
common BGP community names (e.g. no-export, internet, no-peer, etc.) (see: WellKnownCommunityNameMap)

In the following example we add the NO_EXPORT BGP community to two of our nodes via annotation using all three forms of the annotation:

kubectl annotate node <kube-node> "kube-router.io/node.bgp.communities=4294967041"
kubectl annotate node <kube-node> "kube-router.io/node.bgp.communities=65535:65281"
kubectl annotate node <kube-node> "kube-router.io/node.bgp.communities=no-export"

Custom BGP Import Policy Reject

kube-router, by default, accepts all routes advertised by its neighbors.

If the bgp session with one neighbor dies, GoBGP deletes all routes received by it.

If one of the received routes is needed for this node to function properly (eg: custom static route), it could stop working.

In the following example we add custom prefixes that'll be set via a custom import policy reject rule annotation, protecting the node from losing required routes:

kubectl annotate node <kube-node> "kube-router.io/node.bgp.customimportreject=10.0.0.0/16, 192.168.1.0/24"

BGP listen address list

By default, the GoBGP server binds on the node IP address. However, in some cases nodes with multiple IP addresses desire to bind GoBGP to multiple local addresses. Local IP addresses on which GoGBP should listen on a node can be configured with annotation kube-router.io/bgp-local-addresses.

Here is sample example to make GoBGP server to listen on multiple IP address:

kubectl annotate node ip-172-20-46-87.us-west-2.compute.internal "kube-router.io/bgp-local-addresses=172.20.56.25,192.168.1.99"

Overriding the next hop

By default, kube-router populates the GoBGP RIB with node IP as next hop for the advertised pod CIDRs and service VIPs. While this works for most cases, overriding the next hop for the advertised routes is necessary when node has multiple interfaces over which external peers are reached. Next hops need to be defined as the interface over which external peer can be reached. Setting --override-nexthop to true leverages the BGP next-hop-self functionality implemented in GoBGP. The next hop will automatically be selected appropriately when advertising routes, irrespective of the next hop in the RIB.

Overriding the next hop and enable IPIP/tunnel

A common scenario exists where each node in the cluster is connected to two upstream routers that are in two different subnets. For example, one router is connected to a public network subnet and the other router is connected to a private network subnet. Additionally, nodes may be split across different subnets (e.g. different racks) each of which has their own routers.

In this scenario, --override-nexthop can be used to correctly peer with each upstream router, ensuring that the BGP next-hop attribute is correctly set to the node's IP address that faces the upstream router. The --enable-overlay option can be set to allow overlay/underlay tunneling across the different subnets to achieve an interconnected pod network.

This configuration would have the following effects:

Peering Outside the Cluster via one of themany means that kube-router makes available
Overriding Next Hop
Enabling overlays in either full mode or with nodes in different subnets

The warning here is that when using --override-nexthop in the above scenario, it may cause kube-router to advertise an IP address other than the node IP which is what kube-router connects the tunnel to when the --enable-overlay option is given. If this happens it may cause some network flows to become un-routable.

Specifically, people need to take care when combining --override-nexthop and --enable-overlay and make sure that they understand their network, the flows they desire, how the kube-router logic works, and the possible side effects that are created from their configuration. Please refer to this PR for the risk and impact discussion.