The args with using the command `-d` is no longer necessary due to the
fact that the images already ship with a default value of using `-f`
which now runs in foreground without having to enable logging, which
should significantly decrease log noise.
The executor logs service needs to be a headless service so that
the scheduler can connect to the specific executor pod running a
job. Set "ClusterIP: None" for this.
To attach a service to a GCE loadbalancer, the service must be a
NodePort not a ClusterIP. Allow users to specify that as an
option. ClusterIP is the default.
Add options so that a user can create an externally managed secret
with the tenant config file and point the scheduler to that instead
of the one written by the helm values.
So that a user can manage the nodepool config file (which is a secret)
outside of helm, add an option where the user can specify the name
of that secret as a value, and if it's set, helm will not write the
secret and will tell k8s to mount that secret as the nodepool config