Starting out with RKE? Check out this [blog post](http://rancher.com/an-introduction-to-rke/) or the [Quick Start Guide](https://github.com/rancher/rke/wiki/Quick-Start-Guide)
In case both are defined, the system images configuration will take precedence over `kubernetes_version`. Since the `kubernetes_version` options was added mainly to be used by Rancher v2.0, it has a limited number of supported tags that can be found [here](https://github.com/rancher/types/blob/master/apis/management.cattle.io/v3/k8s_defaults.go#L14).
RKE will deploy the addons yaml after the cluster starts, RKE first uploads this yaml file as a configmap in kubernetes cluster and then run a kubernetes job that mounts this config map and deploy the addons.
Note that we are using `|-` because the addons option is a multi line string option, where you can specify multiple yaml files and separate them with `---`
RKE uses kubernetes Jobs to deploy addons. In some cases, addons deployment takes longer than expected. Starting with version `0.1.7-rc1`, RKE provides an option to controle the job check timeout in seconds:
Critical addons will cause RKE to error out if they fail to deploy for any reason. While noncritical addons will just log a warning and continue with the deployment. Currently only the network plugin is considered critical.
RKE is HA ready, you can specify more than one controlplane host in the `cluster.yml` file, and rke will deploy master components on all of them, the kubelets are configured to connect to `127.0.0.1:6443` by default which is the address of `nginx-proxy` service that proxy requests to all master nodes.
to start an HA cluster, just specify more than one host with role `controlplane`, and start the cluster normally.
RKE supports adding/removing nodes for worker and controlplane hosts, in order to add additional nodes you will only need to update the `cluster.yml` file with additional nodes and run `rke up` with the same file.
RKE supports kubernetes cluster upgrade through changing the image version of services, in order to do that change the image option for each services, for example:
Service can also be upgraded by changing any of the services arguments or extra args and run `rke up` again with the updated configuration file.
> Please note that changing the following arguments: `service_cluster_ip_range` or `cluster_cidr` will result in a broken cluster, because currently the network pods will not be automatically upgraded.
RKE will ask some questions around the cluster file like number of the hosts, ips, ssh users, etc, `--empty` option will generate an empty cluster.yml file, also if you just want to print on the screen and not save it in a file you can use `--print`.
RKE will deploy Nginx controller by default, user can disable this by specifying `none` to ingress `provider` option in the cluster configuration, user also can specify list of options for nginx config map listed in this [doc](https://github.com/kubernetes/ingress-nginx/blob/master/docs/user-guide/configmap.md), and command line extra_args listed in this [doc](https://github.com/kubernetes/ingress-nginx/blob/master/docs/user-guide/cli-arguments.md), for example:
By default, RKE will deploy ingress controller on all schedulable nodes (controlplane and workers), to specify only certain nodes for ingress controller to be deployed, user has to specify `node_selector` for the ingress and the right label on the node, for example:
RKE will deploy Nginx Ingress controller as a DaemonSet with `hostnetwork: true`, so ports `80`, and `443` will be opened on each node where the controller is deployed.
As of `v0.1.3-rc2` using `extra_args` will add new arguments and **override** existing defaults. For example, if you need to modify the default admission controllers list, you need to change the default list and add apply it using `extra_args`.
RKE Supports x509 authentication strategy. You can additionally define a list of SANs (Subject Alternative Names) to add to the Kubernetes API Server PKI certificates. This allows you to connect to your Kubernetes cluster API Server through a load balancer, for example, rather than a single node.
RKE supports using external etcd instead of deploying etcd servers, to enable external etcd the following parameters should be populated:
```
services:
etcd:
path: /etcdcluster
external_urls:
- https://etcd-example.com:2379
ca_cert: |-
-----BEGIN CERTIFICATE-----
xxxxxxxxxx
-----END CERTIFICATE-----
cert: |-
-----BEGIN CERTIFICATE-----
xxxxxxxxxx
-----END CERTIFICATE-----
key: |-
-----BEGIN PRIVATE KEY-----
xxxxxxxxxx
-----END PRIVATE KEY-----
```
Note that RKE only supports connecting to TLS enabled etcd setup, user can enable multiple endpoints in the `external_urls` field. RKE will not accept having external urls and nodes with `etcd` role at the same time, user should only specify either etcd role for servers or external etcd but not both.
Starting from v0.1.3 rke supports cloud providers.
### AWS Cloud Provider
To enable AWS cloud provider, you can set the following in the cluster configuration file:
```
cloud_provider:
name: aws
```
AWS cloud provider has to be enabled on ec2 instances with the right IAM role.
### Azure Cloud provider
Azure cloud provider can be enabled by passing `azure` as the cloud provider name and set of options to the configuration file:
```
cloud_provider:
name: azure
cloud_config:
aadClientId: xxxxxxxxxxxx
aadClientSecret: xxxxxxxxxxx
location: westus
resourceGroup: rke-rg
subnetName: rke-subnet
subscriptionId: xxxxxxxxxxx
vnetName: rke-vnet
tenantId: xxxxxxxxxx
securityGroupName: rke-nsg
```
You also have to make sure that the Azure node name must match the kubernetes node name, you can do that by changing the value of hostname_override in the config file:
Using RKE's pluggable user addons, it's possible to deploy Rancher 2.x server in HA with a single command. Detailed instructions can be found [here](https://rancher.com/docs/rancher/v2.x/en/installation/ha-server-install/).
- Container volumes may have some issues in Atomic OS due to SELinux, most of volumes are mounted in rke with option `z`, however user still need to run the following commands before running rke:
```
# mkdir /opt/cni /etc/cni
# chcon -Rt svirt_sandbox_file_t /etc/cni
# chcon -Rt svirt_sandbox_file_t /opt/cni
```
- OpenSSH 6.4 shipped by default on Atomic CentOS which doesn't support SSH tunneling and therefore breaks rke, upgrading OpenSSH to the latest version supported by Atomic host will solve this problem:
```
# atomic host upgrade
```
- Atomic host doesn't come with docker group by default, you can change ownership of docker.sock to enable specific user to run rke:
You can configure a Rancher Kubernetes Engine (RKE) cluster to automatically take snapshots of etcd. In a disaster scenario, you can restore these snapshots, which are stored on other cluster nodes.
### One-Time Snapshots
RKE introduce a new command that can take a snapshot of a running etcd node in rke cluster, the snapshot will be automatically saved in `/opt/rke/etcd-snapshots`, the commands works as following:
```
./rke etcd snapshot-save --config cluster.yml
WARN[0000] Name of the snapshot is not specified using [rke_etcd_snapshot_2018-05-17T23:32:08+02:00]
INFO[0000] Starting saving snapshot on etcd hosts
INFO[0000] [dialer] Setup tunnel for host [x.x.x.x]
INFO[0001] [dialer] Setup tunnel for host [y.y.y.y]
INFO[0002] [dialer] Setup tunnel for host [z.z.z.z]
INFO[0003] [etcd] Saving snapshot [rke_etcd_snapshot_2018-05-17T23:32:08+02:00] on host [x.x.x.x]
INFO[0004] [etcd] Successfully started [etcd-snapshot-once] container on host [x.x.x.x]
INFO[0004] [etcd] Saving snapshot [rke_etcd_snapshot_2018-05-17T23:32:08+02:00] on host [y.y.y.y]
INFO[0005] [etcd] Successfully started [etcd-snapshot-once] container on host [y.y.y.y]
INFO[0005] [etcd] Saving snapshot [rke_etcd_snapshot_2018-05-17T23:32:08+02:00] on host [z.z.z.z]
INFO[0006] [etcd] Successfully started [etcd-snapshot-once] container on host [z.z.z.z]
INFO[0006] Finished saving snapshot [rke_etcd_snapshot_2018-05-17T23:32:08+02:00] on all etcd hosts
```
The command will save a snapshot of etcd from each etcd node in the cluster config file and will save it in `/opt/rke/etcd-snapshots`. This command also creates a container for taking the snapshot. When the process completes, the container is automatically removed.
### Etcd Recurring Snapshots
To schedule a recurring automatic etcd snapshot save, enable the `etcd-snapshot` service. `etcd-snapshot` runs in a service container alongside the `etcd` container. `etcd-snapshot` automatically takes a snapshot of etcd and stores them to its local disk in `/opt/rke/etcd-snapshots`.
To enable `etcd-snapshot` in RKE CLI, configure the following three variables:
```
services:
etcd:
snapshot: true
creation: 5m0s
retention: 24h
```
-`snapshot`: Enables/disables etcd snapshot recurring service in the RKE cluster.
Default value: `false`.
-`creation`: Time period in which `etcd-sanpshot` take snapshots.
Default value: `5m0s`
-`retention`: Time period before before an etcd snapshot expires. Expired snapshots are purged.
Default value: `24h`
After RKE runs, view the `etcd-snapshot` logs to confirm backups are being created automatically:
```
# docker logs etcd-snapshot
time="2018-05-04T18:39:16Z" level=info msg="Initializing Rolling Backups" creation=1m0s retention=24h0m0s
Backups are saved to the following directory: `/opt/rke/etcd-snapshots/`. Backups are created on each node that runs etcd.
### Etcd Disaster recovery
`etcd snapshot-restore` is used for etcd Disaster recovery, it reverts to any snapshot stored in `/opt/rke/etcd-snapshots` that you explicitly define. When you run `etcd snapshot-restore`, RKE removes the old etcd container if it still exists. To restore operations, RKE creates a new etcd cluster using the snapshot you choose.
>**Warning:** Restoring an etcd snapshot deletes your current etcd cluster and replaces it with a new one. Before you run the `etcd snapshot-restore` command, backup any important data in your current cluster.
After taking the etcd backup on node2 we should be able to save this backup in a persistence place, one of the options to do that is to save the backup taken on a s3 bucket or tape backup, for example:
To simulate the failure lets powerdown node2 completely:
```
root@node2:~# poweroff
```
Now its time to pull the backup saved on s3 on a new node:
| Name | IP | Role |
|:-----:|:--------:|:----------------------:|
| node1 | 10.0.0.1 | [controlplane, worker] |
| ~~node2~~ | ~~10.0.0.2~~ | ~~[etcd]~~ |
| node3 | 10.0.0.3 | [etcd] |
| | | |
```
root@node3:~# mkdir -p /opt/rke/etcdbackup
root@node3:~# s3cmd get s3://rke-etcd-backup/snapshot.db /opt/rke/etcdbackup/snapshot.db
```
### 5. Restore etcd on the new node
Now lets do a restore to restore and run etcd on the third node, in order to do that you have first to add the third node to the cluster configuration file:
The previous command will restore the etcd data dir from the snapshot and run etcd container on this node, the final step is to restore the operations on the cluster by making the k8s api to point to the new etcd, to do that we run `rke up` again on the new cluster.yml file:
```
rke up --config cluster.yml
```
You can make sure that operations have been restored by checking the nginx deployment we created earlier:
