Use eviction policy Delete for Low priority VMSS workers

* Fix issue where Azure defaults to Deallocate eviction policy,
which required manually restarting deallocated workers
* Require terraform-provider-azurerm v1.19+ to support setting
the eviction_policy

CHANGES.md

@@ -13,6 +13,12 @@ Notable changes between versions.
   * Enable CoreDNS `loop` and `loadbalance` plugins ([#340](https://github.com/poseidon/typhoon/pull/340))
 * Use kubernetes-incubator/bootkube v0.14.0
 
+#### Azure
+
+* Use eviction policy `Delete` for `Low` priority virtual machine scale set workers ([#343](https://github.com/poseidon/typhoon/pull/343))
+  * Fix issue where Azure defaults to `Deallocate` eviction policy, which required manually restarting deallocated instances. `Delete` policy aligns Azure with AWS and GCP behavior.
+  * Require `terraform-provider-azurerm` v1.19+ (action required)
+
 #### Addons
 
 * Update Prometheus from v2.4.3 to v2.5.0

azure/container-linux/kubernetes/require.tf

@@ -5,7 +5,7 @@ terraform {
 }
 
 provider "azurerm" {
-  version = "~> 1.17"
+  version = "~> 1.19"
 }
 
 provider "local" {

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

@@ -67,8 +67,9 @@ resource "azurerm_virtual_machine_scale_set" "workers" {
   }
 
   # lifecycle
-  priority            = "${var.priority}"
   upgrade_policy_mode = "Manual"
+  priority            = "${var.priority}"
+  eviction_policy     = "Delete"
 }
 
 # Scale up or down to maintain desired number, tolerating deallocations.