Enable more NodeGroup ASG options

[gitfool]

The NodeGroup ASG uses a CF template internally and does not expose a few things we use, like LoadBalancerNames (or TargetGroupARNs), NotificationConfiguration and detailed UpdatePolicy. Another example of where an escape hatch would tide us over, if available.

[gitfool]

Not much success trying to workaround this so far. Given the CF template is so thin, I'm wondering why it's used at all, rather than having first class Pulumi objects that can be manipulated, which would work better with an escape hatch too.

[lukehoban]

Given the CF template is so thin, I'm wondering why it's used at all, rather than having first class Pulumi objects that can be manipulated, which would work better with an escape hatch too.

The lifecycle behavior around cfn-signal that allows waiting on a template deployment successfully completing the initialization info the EC2 instances currently requires coordination with CloudFormation. This is a standard pattern for this in both Terraform and Pulumi.

This should not prevent allowing all necessary parameterization here. It should be possible to pass through any of the inputs used here through node group options.

And of course, as you note, the transformations escape hatch will be a general purpose tool for addressing this sort of thing in the near future.

[metral]

@gitfool Were you able to leverage transformations get passed this?

[kahkeng]

+1 I would like to be able to use ASG's mixedInstancesPolicy when configuring unmanaged EKS node groups. It looks like Pulumi's aws.autoscaling.Group construct lets you pass this in (see https://github.com/pulumi/pulumi-aws/blob/master/sdk/nodejs/autoscaling/group.ts#L103). I wonder if eks.NodeGroup could be streamlined with this given it is also dealing with unmanaged node groups.

Alternatively, is there a way to use autoscaling groups created with aws.autoscaling.Group with Pulumi's eks wrapper?