Lecture 125: Cluster Upgrade Upgrading Nodes: Say I have the same setup in . . .

Kubernetes
#1

Rajasekhar:
Lecture 125: Cluster Upgrade
Upgrading Nodes:
Say I have the same setup in my production, and I upgraded the nodes .

When all the 3 nodes are updated. the 3rd node will have no pods running at the end of upgrade.
Node3 will be free until a new pod created. what if no such situation occurred and Node3 will be free and idle all time.

The load must be high on node 1 and 2 and they continue to serve the requests. Is there a way to utilise node3 immediate after upgrade?? @Mumshad Mannambeth

Thanks

#2

Priyanshu:
Yes simply by deploying containers on node 3 too after upgrade.
Not getting your exact problem!!

#3

Rajasekhar:
Thanks ,
What I did not get is : Say I’m upgrading my cluster .
Node1 : a) Drain the node. b) upgrade the node. c) uncordon the node. ( node1 does not have any pods now)

Node 2: a) Drain the node ( now pods on node2 will move to node1) . b) Upgrade the node . c) uncordon the node. ( node2 does not have any pods now)

Node3: a) Drain the node (now pods on node2 will move to node1) . b) Upgrade the node.
c) uncordon the node . ( node3 does not have pods now ) .

The Pods are running on Node1 and Node2 only . Until pods crashed on node1 and 2 OR new POD request comes Node3 will be free of PODs.

#4

Rajasekhar:
Say there is no request for new pods or no pods crashed on node1 or 2 . Node3 will be idle and doing nothing. how to assign the workload to Node3 in this case ? @Priyanshu

#5

An Đặng Quốc:
Here are some method: https://kubernetes.io/docs/tasks/configure-pod-container/assign-pods-nodes/#create-a-pod-that-gets-scheduled-to-specific-node

#6

An Đặng Quốc:
My favorite methods is assigning a nodeName to a pod

#7

Priyanshu:
@Rajasekhar if that node is a part of cluster, then api server automatically schedules pod to that node which are using less resources but for the stored data and real time data, you might need to use external services as new pod will schedule on node 3 so it will not have the old data

#8

Prasoon Mishra:
@Rajasekhar you can use HPA and assign minimum pod or you can increase replicas under deployment and then back to normal replica ,this might work.

#9

Rajasekhar:
@Prasoon Mishra That is a good idea… Thanks