Logical redux of Node+Pod affinity+anti-affinity

[timothysc]

As of today affinity and anti-affinity specification are very explicit, and lengthy to specify.

Node Affinity Examples:
https://github.com/kubernetes/kubernetes/blob/master/plugin/pkg/scheduler/algorithm/priorities/node_affinity_test.go

Inter-Pod Affinity Examples:
https://github.com/kubernetes/kubernetes/blob/master/plugin/pkg/scheduler/algorithm/priorities/interpod_affinity_test.go

imo ~90% users can typically understand affinity in terms of "groupby" and "spreadby".

groupby (type(pod,node), key=value)

• pod.groupby(node, rack=rack1)
• pod.spreadby(pod, type=highiops)

There is more to this, but hopefully the gist is there.