Dedup pod killer svc and separate nodeSelectors

[smirco-zoox]

Node Selectors

Node selector terms are an object not a string.
I've also added separate nodeSelector terms for each component.
In our setup podKiller and csiController can run on any node, however the nodeDriver daemonset should only run on nodes with a specific label.

Pod Killer Cache svc

In our environment we have multiple quobyte clusters and therefore need to deploy this chart multiple times. As a result all manifests must be uniquely named. pkc was creating a service with the same name for each deplyment

[venkatsc]

@smirco-zoox
I think, cacheNodeSelector would be more opt

[smirco-zoox]

why? it's json path is quobyte.podKiller.nodeSelector

I'd argue that all the other components should also have paths like this to be more consistent with how podKiller is setup.

e.g. quobyte.csiController.nodeSelector instead of quobyte.csiControllerNodeSelector

However this restructures your values quite a bit more, so is a more major release of the helm chart.

[venkatsc]

why? it's json path is quobyte.podKiller.nodeSelector

Because, pod killer has two components:

• pod mount monitor that runs along with other Quobyte CSI container in a single pod on each node with quobyte.csiControllerNodeSelector
• pod killer cache runs as a separate pod on node with quobyte.podKiller.[cache]NodeSelector

This could create some ambiguity, so please modify the naming to avoid this ambiguity.

I will make a new release. Since it is a simple configuration change, it can be easily adapted.