Use an Ansible group for TDP cluster instead of "all"

[rpignolet]

In Kubespray inventory sample (https://github.com/kubernetes-sigs/kubespray/blob/master/inventory/sample/inventory.ini) they use a group k8s_cluster used by Kubespray playbooks to know what Ansible hosts are part of Kubernetes cluster.

Inside our actual getting started, we use all Ansible groups to configure /etc/hosts and yum packages for example. I think we should create an Ansible group tdp_cluster or hadoop_cluster and never use the all Ansible groups to not configure hosts which are not part of TDP/Hadoop cluster.

[gboutry]

I'm all for having a separate entity designating the cluster, moreover, I would propose we move the cluster definitions vars that were inside the group hadoop towards a group with the new name.

The plan, for now, was to move them to the all group.

[leopaul36]

Agreed, it would make sense for this file https://github.com/TOSIT-IO/tdp-getting-started/blob/master/inventory/group_vars/all.yml to become tdp_cluster.yml

[jusstol]

I completely agree. On our side at RTE, we created the groupnodes group to our inventory just to be able to put every nodes as children of a specific environment (all nodes are children of C03, which is child of OPF). We had to create this group because children of groups must be groups, and using the special group all is not allowed because all groups are already children of all :

ERROR! Adding group 'OPF' as child to 'all' creates a recursive dependency loop.

Our inventory :

....

[groupnodes]
worker-01
worker-02
worker-03
master-01
master-02
master-03
edge-01

[C03:children]
groupnodes

[OPF:children]
C03

[rpignolet]

Poll result -> use tdp_cluster instead of all.