etcd

Etcd Helm Chart

Credit to https://github.com/ingvagabund. This is an implementation of that work

• kubernetes-retired/contrib#1295

Prerequisites Details

• Kubernetes 1.3 with alpha APIs enable
• PV support on the underlying infrastructure

PetSet Details

• http://kubernetes.io/docs/user-guide/petset/

PetSet Caveats

• http://kubernetes.io/docs/user-guide/petset/#alpha-limitations

Todo

• Implement SSL

Chart Details

This chart will do the following:

• Implemented a dynamically scalable etcd cluster using Kubernetes PetSets

Get this chart

Download the latest release of the chart from the releases page.

Alternatively, clone the repo if you wish to use the development snapshot:

$ git clone https://github.com/kubernetes/charts.git

Installing the Chart

To install the chart with the release name my-release:

$ helm install --name my-release etcd-x.x.x.tgz

Configuration

The following tables lists the configurable parameters of the etcd chart and their default values.

Parameter	Description	Default
Name	Spark master name	etcd
Image	Container image name	gcr.io/google_containers/etcd-amd64
ImageTag	Container image tag	2.2.5
ImagePullPolicy	Container pull policy	Always
Replicas	k8s petset replicas	3
Component	k8s selector key	etcd
Cpu	container requested cpu	100m
Memory	container requested memory	512Mi
ClientPort	k8s service port	2379
PeerPorts	Container listening port	2380
Storage	Persistent volume size	1Gi

Specify each parameter using the --set key=value[,key=value] argument to helm install.

Alternatively, a YAML file that specifies the values for the parameters can be provided while installing the chart. For example,

$ helm install --name my-release -f values.yaml etcd-x.x.x.tgz

Tip: You can use the default values.yaml

Deep dive

Cluster Health

$ for i in <0..n>; do kubectl exec <release-podname-$i> -- sh -c 'etcdctl cluster-health'; done

eg.

$ for i in {0..9}; do kubectl exec named-lynx-etcd-$i --namespace=etcd -- sh -c 'etcdctl cluster-health'; done
member 7878c44dabe58db is healthy: got healthy result from http://named-lynx-etcd-7.named-lynx-etcd:2379
member 19d2ab7b415341cc is healthy: got healthy result from http://named-lynx-etcd-4.named-lynx-etcd:2379
member 6b627d1b92282322 is healthy: got healthy result from http://named-lynx-etcd-3.named-lynx-etcd:2379
member 6bb377156d9e3fb3 is healthy: got healthy result from http://named-lynx-etcd-0.named-lynx-etcd:2379
member 8ebbb00c312213d6 is healthy: got healthy result from http://named-lynx-etcd-8.named-lynx-etcd:2379
member a32e3e8a520ff75f is healthy: got healthy result from http://named-lynx-etcd-5.named-lynx-etcd:2379
member dc83003f0a226816 is healthy: got healthy result from http://named-lynx-etcd-2.named-lynx-etcd:2379
member e3dc94686f60465d is healthy: got healthy result from http://named-lynx-etcd-6.named-lynx-etcd:2379
member f5ee1ca177a88a58 is healthy: got healthy result from http://named-lynx-etcd-1.named-lynx-etcd:2379
cluster is healthy

Failover

If any etcd member fails it gets re-joined eventually. You can test the scenario by killing process of one of the pets:

$ ps aux | grep etcd-1
$ kill -9 ETCD_1_PID

$ kubectl get pods -l "app=etcd"
NAME                 READY     STATUS        RESTARTS   AGE
etcd-0               1/1       Running       0          54s
etcd-2               1/1       Running       0          51s

After a while:

$ kubectl get pods -l "app=etcd"
NAME                 READY     STATUS    RESTARTS   AGE
etcd-0               1/1       Running   0          1m
etcd-1               1/1       Running   0          20s
etcd-2               1/1       Running   0          1m

You can check state of re-joining from etcd-1's logs:

$ kubectl logs etcd-1
Waiting for etcd-0.etcd to come up
Waiting for etcd-1.etcd to come up
ping: bad address 'etcd-1.etcd'
Waiting for etcd-1.etcd to come up
Waiting for etcd-2.etcd to come up
Re-joining etcd member
Updated member with ID 7fd61f3f79d97779 in cluster
2016-06-20 11:04:14.962169 I | etcdmain: etcd Version: 2.2.5
2016-06-20 11:04:14.962287 I | etcdmain: Git SHA: bc9ddf2
...

Scaling using kubectl

This is for reference. Scaling should be managed by helm upgrade

The etcd cluster can be scale up by running kubectl patch or kubectl edit. For instance,

$ kubectl get pods -l "app=etcd"
NAME      READY     STATUS    RESTARTS   AGE
etcd-0    1/1       Running   0          7m
etcd-1    1/1       Running   0          7m
etcd-2    1/1       Running   0          6m

$ kubectl patch petset/etcd -p '{"spec":{"replicas": 5}}'
"etcd" patched

$ kubectl get pods -l "app=etcd"
NAME      READY     STATUS    RESTARTS   AGE
etcd-0    1/1       Running   0          8m
etcd-1    1/1       Running   0          8m
etcd-2    1/1       Running   0          8m
etcd-3    1/1       Running   0          4s
etcd-4    1/1       Running   0          1s

Scaling-down is similar. For instance, changing the number of pets to 4:

$ kubectl edit petset/etcd
petset "etcd" edited

$ kubectl get pods -l "app=etcd"
NAME      READY     STATUS    RESTARTS   AGE
etcd-0    1/1       Running   0          8m
etcd-1    1/1       Running   0          8m
etcd-2    1/1       Running   0          8m
etcd-3    1/1       Running   0          4s

Once a pet is terminated (either by running kubectl delete pod etcd-ID or scaling down), content of /var/run/etcd/ directory is cleaned up. If any of the etcd pets restarts (e.g. caused by etcd failure or any other), the directory is kept untouched so the pet can recover from the failure.