Research and develop method to scale up an existing cluster
One of the desired features of Jetlag would be the ability to scale up a cluster after an initial cluster is installed. We need to research scale up methods for assisted-installer based cluster, and determine if these are feasible via a playbook to make this a simple task.
A new feature GA'd in 4.17 that could be an implementation here https://docs.openshift.com/container-platform/4.17/nodes/nodes/nodes-nodes-adding-node-iso.html
Going through the docs initially, the generate-discovery-iso and boot-iso jetlag tasks already accomplish the steps required to add nodes through this method and already includes a httpd server to host the iso.
I've been working on this using this doc as a template for workflow: https://access.redhat.com/solutions/6968529
Oh! Ok.
That makes sense for versions less than 4.17 where this feature was added.
Maybe we look to use this new approach for 4.17 and onward?
Can someone work on this feature with priority as still JetSki is the only option to deploy a large cluster? We are using JetSki to deploy a smaller cluster and then scale the nodes which helps in debugging hardware issues. However JetSki is not actively mainted whereas Jetlag has more users and contributors. So enhancing Jetlag to include scale up features will be helpful for people in transition to Jetlag.
https://github.com/radez/jetlag/tree/scale_out I used this branch this morning to scale out on 4.17
- Deploy your cluster with only the initial workers in the [worker] inventory group.
- Add the new workers records to the [worker] inventory group
- in vars/scale_out.yml Set current_worker_count to the initial number of workers and scale_out_count to the number of workers you've added to the [worker] group and intend to scale the cluster to.
- Run ansible-playbook -i ansible/inventory/cloudX.local ansible/mno-scale-out.yml
I'm working on adding these instructions to a doc file and I'll put in a patch
@radez would you consider this issue complete now? Is there any other scale up work needed?
Closing as #613 was merged