When launching into a shared VPC, the VPC & the Internet Gateway will be reused. By default we create a new subnet per zone, and a new route table, but you can also use a shared subnet (see below).
- Use
kops create cluster
with the--vpc
argument for your existing VPC:
export KOPS_STATE_STORE=s3://<somes3bucket>
export CLUSTER_NAME=<sharedvpc.mydomain.com>
export VPC_ID=vpc-12345678 # replace with your VPC id
export NETWORK_CIDR=10.100.0.0/16 # replace with the cidr for the VPC ${VPC_ID}
kops create cluster --zones=us-east-1b --name=${CLUSTER_NAME} --vpc=${VPC_ID}
- Then
kops edit cluster ${CLUSTER_NAME}
will show you something like:
metadata:
creationTimestamp: "2016-06-27T14:23:34Z"
name: ${CLUSTER_NAME}
spec:
cloudProvider: aws
networkCIDR: ${NETWORK_CIDR}
networkID: ${VPC_ID}
nonMasqueradeCIDR: 100.64.0.0/10
subnets:
- cidr: 172.20.32.0/19
name: us-east-1b
type: Public
zone: us-east-1b
Verify that networkCIDR
& networkID
match your VPC CIDR & ID.
You likely need to set the CIDR on each of the Zones, because subnets in a VPC cannot overlap.
- You can then run
kops update cluster
in preview mode (without--yes
). You don't need any arguments, because they're all in the cluster spec:
kops update cluster ${CLUSTER_NAME}
Review the changes to make sure they are OK - the Kubernetes settings might not be ones you want on a shared VPC (in which case, open an issue!)
Note also the Kubernetes VPCs (currently) require EnableDNSHostnames=true
. kops will detect the required change,
but refuse to make it automatically because it is a shared VPC. Please review the implications and make the change
to the VPC manually.
- Once you're happy, you can create the cluster using:
kops update cluster ${CLUSTER_NAME} --yes
This will add an additional Tag to your aws vpc resource. This tag will be removed automatically if you delete your kops cluster.
"kubernetes.io/cluster/<cluster-name>" = "shared"
Prior to kops 1.8 this Tag Key was KubernetesCluster
which is obsolete and should
not be used anymore as it only supports one cluster.
AWS now allows you to add more CIDRs to a VPC, the param AdditionalNetworkCIDRs
allows you to specify any additional CIDRs added to the VPC.
metadata:
creationTimestamp: "2016-06-27T14:23:34Z"
name: ${CLUSTER_NAME}
spec:
cloudProvider: aws
networkCIDR: 10.1.0.0/16
additionalNetworkCIDRs:
- 10.2.0.0/16
networkID: vpc-00aa5577
subnets:
- cidr: 10.1.0.0/19
name: us-east-1b
type: Public
zone: us-east-1b
id: subnet-1234567
- cidr: 10.2.0.0/19
name: us-east-1b
type: Public
zone: us-east-1b
id: subnet-1234568
kops
can create a cluster in shared subnets in both public and private network topologies. Doing so is not recommended unless you are using external networking
- Use kops create cluster with the
--subnets
argument for your existing subnets:
export KOPS_STATE_STORE=s3://<somes3bucket>
export CLUSTER_NAME=<sharedvpc.mydomain.com>
export VPC_ID=vpc-12345678 # replace with your VPC id
export NETWORK_CIDR=10.100.0.0/16 # replace with the cidr for the VPC ${VPC_ID}
export SUBNET_ID=subnet-12345678 # replace with your subnet id
export SUBNET_CIDR=10.100.0.0/24 # replace with your subnet CIDR
export SUBNET_IDS=$SUBNET_IDS # replace with your comma separated subnet ids
kops create cluster --zones=us-east-1b --name=${CLUSTER_NAME} --subnets=${SUBNET_IDS}
--vpc
is optional when specifying --subnets
. When creating a cluster with a
private topology and shared subnets, the utility subnets should be specified similarly with --utility-subnets
.
- Then
kops edit cluster ${CLUSTER_NAME}
will show you something like:
metadata:
creationTimestamp: "2016-06-27T14:23:34Z"
name: ${CLUSTER_NAME}
spec:
cloudProvider: aws
networkCIDR: ${NETWORK_CIDR}
networkID: ${VPC_ID}
nonMasqueradeCIDR: 100.64.0.0/10
subnets:
- cidr: ${SUBNET_CIDR}
id: ${SUBNET_ID}
name: us-east-1b
type: Public
zone: us-east-1b
- Once you're happy, you can create the cluster using:
kops update cluster ${CLUSTER_NAME} --yes
If you run in AWS private topology with shared subnets, and you would like Kubernetes to provision resources in these shared subnets, you must create tags on them.
This is important, for example, if your utility
subnets are shared, you will not be able to launch any services that create Elastic Load Balancers (ELBs).
Prior to kops 1.8 KubernetesCluster
tag was used for this. This lead to several problems if there were more than one Kubernetes Cluster in a subnet.
After you upgraded to kops 1.8 remove KubernetesCluster
Tag from subnets otherwise kubernetes.io/cluster/<clustername>
won't have any effect!
These are currently needed Tags on shared resources:
Public Subnets:
"kubernetes.io/cluster/<cluster-name>" = "shared"
"kubernetes.io/role/elb" = "1"
"SubnetType" = "Utility"
Private Subnets:
"kubernetes.io/cluster/<cluster-name>" = "shared"
"kubernetes.io/role/internal-elb" = "1"
"SubnetType" = "Private"
On AWS in private topology, kops
creates one NAT Gateway (NGW) per AZ. If your shared VPC is already set up with an NGW in the subnet that kops
deploys private resources to, it is possible to specify the ID and have kops
/kubernetes
use it.
If you don't want to use NAT Gateways but have setup EC2 NAT Instances in your VPC that you can share, it's possible to specify the IDs of said instances and have kops
/kubernetes
use them.
After creating a basic cluster spec, edit your cluster to specify NGW:
kops edit cluster ${CLUSTER_NAME}
spec:
subnets:
- cidr: 10.20.64.0/21
name: us-east-1a
egress: nat-987654321
type: Private
zone: us-east-1a
- cidr: 10.20.96.0/21
name: us-east-1b
egress: i-987654321
type: Private
zone: us-east-1a
- cidr: 10.20.32.0/21
name: utility-us-east-1a
type: Utility
zone: us-east-1a
Please note:
- You must specify pre-create subnets for all the subnets, or for none of them.
- kops won't alter your existing subnets. Therefore they must be correctly set up with route tables etc. The Public or Utility subnets should have public IPs and an internet gateway configured as their default route in their route table. Private subnets should not have public IPs, and will typically have a NAT gateway configured as their default route.
- kops won't create a route-table at all if we're not creating subnets.
- In the example above the first subnet is using a shared NAT Gateway while the second one is using a shared NAT Instance