Allow preemptible workers on AWS via spot instances

* Add `worker_price` to allow worker spot instances. Defaults
to empty string for the worker autoscaling group to use regular
on-demand instances.
* Add `spot_price` to internal `workers` module for spot worker
pools
* Note: Unlike GCP `preemptible` workers, spot instances require
you to pick a bid price.

CHANGES.md

@@ -2,7 +2,14 @@
 
 Notable changes between versions.
 
-## Latest
+#### AWS
+
+* Allow "preemptible" workers via spot instances ([#202](https://github.com/poseidon/typhoon/pull/202))
+  * Add `worker_price` to allow worker spot instances. Defaults to empty string for the worker autoscaling group to use regular on-demand instances.
+  * Add `spot_price` to internal `workers` module for spot [worker pools](https://typhoon.psdn.io/advanced/worker-pools/)
+  * Note: Unlike GCP `preemptible` workers, spot instances require you to pick a bid price.
+
+## v1.10.2
 
 * [Introduce](https://typhoon.psdn.io/announce/#april-26-2018) Typhoon for Fedora Atomic ([#199](https://github.com/poseidon/typhoon/pull/199))
 * Kubernetes [v1.10.2](https://github.com/kubernetes/kubernetes/blob/master/CHANGELOG-1.10.md#v1102)

aws/container-linux/kubernetes/variables.tf

@@ -59,6 +59,12 @@ variable "disk_type" {
   description = "Type of the EBS volume (e.g. standard, gp2, io1)"
 }
 
+variable "worker_price" {
+  type        = "string"
+  default     = ""
+  description = "Spot price in USD for autoscaling group spot instances. Leave as default empty string for autoscaling group to use on-demand instances. Note, switching in-place from spot to on-demand is not possible: https://github.com/terraform-providers/terraform-provider-aws/issues/4320"
+}
+
 variable "controller_clc_snippets" {
   type        = "list"
   description = "Controller Container Linux Config snippets"

aws/container-linux/kubernetes/workers.tf

@@ -10,6 +10,7 @@ module "workers" {
   instance_type   = "${var.worker_type}"
   os_channel      = "${var.os_channel}"
   disk_size       = "${var.disk_size}"
+  spot_price      = "${var.worker_price}"
 
   # configuration
   kubeconfig            = "${module.bootkube.kubeconfig}"

aws/container-linux/kubernetes/workers/variables.tf

@@ -52,6 +52,12 @@ variable "disk_type" {
   description = "Type of the EBS volume (e.g. standard, gp2, io1)"
 }
 
+variable "spot_price" {
+  type        = "string"
+  default     = ""
+  description = "Spot price in USD for autoscaling group spot instances. Leave as default empty string for autoscaling group to use on-demand instances. Note, switching in-place from spot to on-demand is not possible: https://github.com/terraform-providers/terraform-provider-aws/issues/4320"
+}
+
 variable "clc_snippets" {
   type        = "list"
   description = "Container Linux Config snippets"

aws/container-linux/kubernetes/workers/workers.tf

@@ -26,6 +26,12 @@ resource "aws_autoscaling_group" "workers" {
     create_before_destroy = true
   }
 
+  # Waiting for instance creation delays adding the ASG to state. If instances
+  # can't be created (e.g. spot price too low), the ASG will be orphaned.
+  # Orphaned ASGs escape cleanup, can't be updated, and keep bidding if spot is
+  # used. Disable wait to avoid issues and align with other clouds.
+  wait_for_capacity_timeout = "0"
+
   tags = [{
     key                 = "Name"
     value               = "${var.name}-worker"
@@ -37,6 +43,7 @@ resource "aws_autoscaling_group" "workers" {
 resource "aws_launch_configuration" "worker" {
   image_id      = "${data.aws_ami.coreos.image_id}"
   instance_type = "${var.instance_type}"
+  spot_price    = "${var.spot_price}"
 
   user_data = "${data.ct_config.worker_ign.rendered}"
 

aws/fedora-atomic/kubernetes/variables.tf

@@ -53,6 +53,12 @@ variable "disk_type" {
   description = "Type of the EBS volume (e.g. standard, gp2, io1)"
 }
 
+variable "worker_price" {
+  type        = "string"
+  default     = ""
+  description = "Spot price in USD for autoscaling group spot instances. Leave as default empty string for autoscaling group to use on-demand instances. Note, switching in-place from spot to on-demand is not possible: https://github.com/terraform-providers/terraform-provider-aws/issues/4320"
+}
+
 # configuration
 
 variable "ssh_authorized_key" {

aws/fedora-atomic/kubernetes/workers.tf

@@ -9,6 +9,7 @@ module "workers" {
   count           = "${var.worker_count}"
   instance_type   = "${var.worker_type}"
   disk_size       = "${var.disk_size}"
+  spot_price      = "${var.worker_price}"
 
   # configuration
   kubeconfig            = "${module.bootkube.kubeconfig}"

aws/fedora-atomic/kubernetes/workers/variables.tf

@@ -46,6 +46,12 @@ variable "disk_type" {
   description = "Type of the EBS volume (e.g. standard, gp2, io1)"
 }
 
+variable "spot_price" {
+  type        = "string"
+  default     = ""
+  description = "Spot price in USD for autoscaling group spot instances. Leave as default empty string for autoscaling group to use on-demand instances. Note, switching in-place from spot to on-demand is not possible: https://github.com/terraform-providers/terraform-provider-aws/issues/4320"
+}
+
 # configuration
 
 variable "kubeconfig" {

aws/fedora-atomic/kubernetes/workers/workers.tf

@@ -26,6 +26,12 @@ resource "aws_autoscaling_group" "workers" {
     create_before_destroy = true
   }
 
+  # Waiting for instance creation delays adding the ASG to state. If instances
+  # can't be created (e.g. spot price too low), the ASG will be orphaned.
+  # Orphaned ASGs escape cleanup, can't be updated, and keep bidding if spot is
+  # used. Disable wait to avoid issues and align with other clouds.
+  wait_for_capacity_timeout = "0"
+
   tags = [{
     key                 = "Name"
     value               = "${var.name}-worker"
@@ -37,6 +43,7 @@ resource "aws_autoscaling_group" "workers" {
 resource "aws_launch_configuration" "worker" {
   image_id      = "${data.aws_ami.fedora.image_id}"
   instance_type = "${var.instance_type}"
+  spot_price    = "${var.spot_price}"
 
   user_data = "${data.template_file.worker-cloudinit.rendered}"
 

docs/advanced/worker-pools.md

@@ -68,10 +68,11 @@ The AWS internal `workers` module supports a number of [variables](https://githu
 | instance_type | EC2 instance type | "t2.small" | "t2.medium" |
 | os_channel | Container Linux AMI channel | stable| "beta", "alpha" |
 | disk_size | Size of the disk in GB | 40 | 100 |
+| spot_price | Spot price in USD for workers. Leave as default empty string for regular on-demand instances | "" | "0.10" |
 | service_cidr | Must match `service_cidr` of cluster | "10.3.0.0/16" | "10.3.0.0/24" |
 | cluster_domain_suffix | Must match `cluster_domain_suffix` of cluster | "cluster.local" | "k8s.example.com" |
 
-Check the list of valid [instance types](https://aws.amazon.com/ec2/instance-types/).
+Check the list of valid [instance types](https://aws.amazon.com/ec2/instance-types/) or per-region and per-type [spot prices](https://aws.amazon.com/ec2/spot/pricing/).
 
 ## Google Cloud
 

docs/atomic/aws.md

@@ -227,6 +227,7 @@ Reference the DNS zone id with `"${aws_route53_zone.zone-for-clusters.zone_id}"`
 | worker_type | EC2 instance type for workers | "t2.small" | See below |
 | disk_size | Size of the EBS volume in GB | "40" | "100" |
 | disk_type | Type of the EBS volume | "gp2" | standard, gp2, io1 |
+| worker_price | Spot price in USD for workers. Leave as default empty string for regular on-demand instances | "" | "0.10" |
 | networking | Choice of networking provider | "calico" | "calico" or "flannel" |
 | network_mtu | CNI interface MTU (calico only) | 1480 | 8981 |
 | host_cidr | CIDR IPv4 range to assign to EC2 instances | "10.0.0.0/16" | "10.1.0.0/16" |

docs/cl/aws.md

@@ -244,6 +244,7 @@ Reference the DNS zone id with `"${aws_route53_zone.zone-for-clusters.zone_id}"`
 | os_channel | Container Linux AMI channel | stable | stable, beta, alpha |
 | disk_size | Size of the EBS volume in GB | "40" | "100" |
 | disk_type | Type of the EBS volume | "gp2" | standard, gp2, io1 |
+| worker_price | Spot price in USD for workers. Leave as default empty string for regular on-demand instances | "" | "0.10" |
 | controller_clc_snippets | Controller Container Linux Config snippets | [] | |
 | worker_clc_snippets | Worker Container Linux Config snippets | [] | |
 | networking | Choice of networking provider | "calico" | "calico" or "flannel" |