Document how clients not on K8s can join a DC in K8s. (#9438)

This commit is contained in:
Nitya Dhanushkodi 2021-01-07 16:14:07 -08:00 committed by GitHub
parent a9285ad76a
commit dd81e4a1cc
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 129 additions and 14 deletions

View File

@ -11,8 +11,21 @@ description: >-
Consul clients running on non-Kubernetes nodes can join a Consul cluster running within Kubernetes. Consul clients running on non-Kubernetes nodes can join a Consul cluster running within Kubernetes.
## Auto-join ## Networking
Within one datacenter, Consul typically requires a fully connected
[network](/docs/architecture). This means the IPs of every client and server
agent should be routable by every other client and server agent in the
datacenter. Clients need to be able to [gossip](/docs/architecture/gossip) with
every other agent and make RPC calls to servers. Servers need to be able to
gossip with every other agent. See [Architecture](/docs/architecture) for more details.
-> **Consul Enterprise customers** may use [network
segments](/docs/enterprise/network-segments) to enable non-fully-connected
topologies. However, out-of-cluster nodes must still be able to communicate
with the server pod or host IP addresses.
## Auto-join
The recommended way to join a cluster running within Kubernetes is to The recommended way to join a cluster running within Kubernetes is to
use the ["k8s" cloud auto-join provider](/docs/agent/cloud-auto-join#kubernetes-k8s). use the ["k8s" cloud auto-join provider](/docs/agent/cloud-auto-join#kubernetes-k8s).
@ -29,20 +42,122 @@ started using the [official Helm chart](/docs/k8s/helm):
```shell-session ```shell-session
$ consul agent -retry-join 'provider=k8s label_selector="app=consul,component=server"' $ consul agent -retry-join 'provider=k8s label_selector="app=consul,component=server"'
``` ```
-> **Note:** This auto-join command only connects on the default gossip port
8301, whether you are joining on the pod network or via host ports. Either a
consul server or client that is already a member of the datacenter should be
listening on this port for the external client agent to be able to use
auto-join.
By default, Consul will join the default gossip port. Pods may set an ### Auto-join on the Pod network
annotation `consul.hashicorp.com/auto-join-port` to an integer value or In the default Consul Helm chart installation, Consul clients and servers are
a named port to specify the port for the auto-join to return. This enables routable only via their pod IPs for server RPCs and gossip (HTTP
different pods to have different exposed ports. API calls to Consul clients can also be made through host IPs). This means any
external client agents joining the Consul cluster running on Kubernetes would
need to be able to have connectivity to those pod IPs.
## Networking In many hosted Kubernetes environments, you will need to explicitly configure
your hosting provider to ensure that pod IPs are routable from external VMs.
See [Azure AKS
CNI](https://docs.microsoft.com/en-us/azure/aks/concepts-network#azure-cni-advanced-networking),
[AWS EKS
CNI](https://docs.aws.amazon.com/eks/latest/userguide/pod-networking.html) and
[GKE VPC-native
clusters](https://cloud.google.com/kubernetes-engine/docs/concepts/alias-ips).
Given you have the [official Helm chart](/docs/k8s/helm) installed with the default values, do the following to join an external client agent.
1. Make sure the pod IPs of the clients and servers in Kubernetes are
routable from the VM and that the VM can access port 8301 (for gossip) and
port 8300 (for server RPC) on those pod IPs.
1. Make sure that the client and server pods running in Kubernetes can route
to the VM's advertise IP on its gossip port (default 8301).
2. Make sure you have the `kubeconfig` file for the Kubernetes cluster in `$HOME/.kube/config` on the external VM.
2. On the external VM, run:
```bash
consul agent \
-advertise="$ADVERTISE_IP" \
-retry-join='provider=k8s label_selector="app=consul,component=server"'
-bind=0.0.0.0 \
-hcl='leave_on_terminate = true' \
-hcl='ports { grpc = 8502 }' \
-config-dir=$CONFIG_DIR \
-datacenter=$DATACENTER \
-data-dir=$DATA_DIR \
```
3. Check if the join was successful by running `consul members`. Sample output:
```shell-session
/ $ consul members
Node Address Status Type Build Protocol DC Segment
consul-consul-server-0 10.138.0.43:9301 alive server 1.9.1 2 dc1 <all>
external-agent 10.138.0.38:8301 alive client 1.9.0 2 dc1 <default>
gke-external-agent-default-pool-32d15192-grs4 10.138.0.43:8301 alive client 1.9.1 2 dc1 <default>
gke-external-agent-default-pool-32d15192-otge 10.138.0.44:8301 alive client 1.9.1 2 dc1 <default>
gke-external-agent-default-pool-32d15192-vo7k 10.138.0.42:8301 alive client 1.9.1 2 dc1 <default>
```
### Auto-join via host ports
If your external VMs can't connect to Kubernetes pod IPs, but they can connect
to the internal host IPs of the nodes in the Kubernetes cluster, you have the
option to expose the clients and server ports on the host IP instead.
1. Install the [official Helm chart](/docs/k8s/helm) with the following values:
```yaml
client:
exposeGossipPorts: true # exposes client gossip ports as hostPorts
server:
exposeGossipAndRPCPorts: true # exposes the server gossip and RPC ports as hostPorts
ports:
# Configures the server gossip port
serflan:
# Note that this needs to be different than 8301, to avoid conflicting with the client gossip hostPort
port: 9301
```
This will expose the client gossip ports, the server gossip ports and the server RPC port at `hostIP:hostPort`. Note that `hostIP` is the **internal** IP of the VM that the client/server pods are deployed on.
1. Make sure the IPs of the Kubernetes nodes are routable from the VM and
that the VM can access ports 8301 and 9301 (for gossip) and port 8300 (for
server RPC) on those node IPs.
1. Make sure the client and server pods running in Kubernetes can route to
the VM's advertise IP on its gossip port (default 8301).
3. Make sure you have the `kubeconfig` file for the Kubernetes cluster in `$HOME/.kube/config` on the external VM.
4. On the external VM, run (note the addition of `host_network=true` in the retry-join argument):
```bash
consul agent \
-advertise="$ADVERTISE_IP" \
-retry-join='provider=k8s host_network=true label_selector="app=consul,component=server"'
-bind=0.0.0.0 \
-hcl='leave_on_terminate = true' \
-hcl='ports { grpc = 8502 }' \
-config-dir=$CONFIG_DIR \
-datacenter=$DATACENTER \
-data-dir=$DATA_DIR \
```
3. Check if the join was successful by running `consul members`. Sample output:
```shell-session
/ $ consul members
Node Address Status Type Build Protocol DC Segment
consul-consul-server-0 10.138.0.43:9301 alive server 1.9.1 2 dc1 <all>
external-agent 10.138.0.38:8301 alive client 1.9.0 2 dc1 <default>
gke-external-agent-default-pool-32d15192-grs4 10.138.0.43:8301 alive client 1.9.1 2 dc1 <default>
gke-external-agent-default-pool-32d15192-otge 10.138.0.44:8301 alive client 1.9.1 2 dc1 <default>
gke-external-agent-default-pool-32d15192-vo7k 10.138.0.42:8301 alive client 1.9.1 2 dc1 <default>
```
## Manual join
If you are unable to use auto-join, you can also follow the instructions in
either of the auto-join sections but instead of using a `provider` key in the
`-retry-join` flag, you would need to pass the address of at least one
consul server, e.g: `-retry-join=$CONSUL_SERVER_IP:$SERVER_SERFLAN_PORT`.
However, rather than hardcoding the IP, it's recommended to set up a DNS entry
that would resolve to the consul servers' pod IPs (if using the pod network) or
host IPs that the server pods are running on (if using host ports).
Consul typically requires a fully connected network.
Because the Consul Helm chart currently doesn't allow exposing servers' gossip ports via a `hostPort`,
nodes outside of Kubernetes joining a cluster running within Kubernetes must be able to communicate
to pod IPs via the network. Note that the auto-join provider discussed above will use pod IPs by default.
-> **Consul Enterprise customers** may use
[network segments](/docs/enterprise/network-segments) to
enable non-fully-connected topologies. However, out-of-cluster nodes must still
be able to communicate with the server pod or host IP addresses.