Assume I have a cluster with 2 nodes and a POD with 2 replicas. Can I have the guarantee that my 2 replicas are deployed in 2 differents nodes. So that when a node is down, the application keeps running. By default does the scheduler work on best effort mode to assign the 2 replicas in distinct nodes?
Daemonset is not a good option. It will schedule one pod on every node. In future if you scale your cluster and then pods get scaled as many as nodes. Instead Use pod affinity to schedule no more than one pod on any node
You can use kind: DeamonSet . Here is a link to Kubernetes DeamonSet documentation.
A DaemonSet ensures that all (or some) Nodes run a copy of a Pod. As nodes are added to the cluster, Pods are added to them. As nodes are removed from the cluster, those Pods are garbage collected. Deleting a DaemonSet will clean up the Pods it created.
Here is a link to documentation about DeamonSets in OpenShift Example might look like the following:
This is available on Openshift >= 3.2 version of openshift This use case is to run a specific docker container (veermuchandi/welcome) on all nodes (or a set nodes with specific label
Enable HostPorts expose on Openshift
$ oc edit scc restricted #as system:admin user
change allowHostPorts: true and save
apiVersion: extensions/v1beta1
kind: DaemonSet
metadata:
name: welcome
spec:
template:
metadata:
name: welcome
labels:
daemon: welcome
spec:
containers:
- name: c
image: veermuchandi/welcome
ports:
- containerPort: 8080
hostPort: 8080
name: serverport$ oc create -f myDaemonset.yaml #with system:admin user
Source available here
Pod anti-affinity can also to repel the pod from each other. so no two pods can be scheduled on same node.
Use following configurations.
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx
spec:
replicas: 2
selector:
matchLabels:
app: nginx
template:
metadata:
labels:
app: nginx
spec:
affinity:
podAntiAffinity:
requiredDuringSchedulingIgnoredDuringExecution:
- labelSelector:
matchExpressions:
- key: app
operator: In
values:
- nginx
topologyKey: "kubernetes.io/hostname"
containers:
- name: nginx
image: nginxThis will use the anti-affinity feature so if you are having more than 2 nodes the there will be guarantee that no two pod will be scheduled on same node.