Disable ECS AZ rebalancing so binpack placement applies

terraform/modules/container/README.md

@@ -163,7 +163,7 @@ No modules.
 | <a name="input_container_memory"></a> [container\_memory](#input\_container\_memory) | memory allocation in MB. 1024 is one full gig of memory | `number` | `1024` | no |
 | <a name="input_container_memory_reservation"></a> [container\_memory\_reservation](#input\_container\_memory\_reservation) | Soft memory limit in MiB. ECS subtracts this from a container instance when placing the task; the container may burst above it up to `container_memory`, which stays the hard cap. EC2 launch type only. | `number` | `256` | no |
 | <a name="input_container_port"></a> [container\_port](#input\_container\_port) | what port this container opens up to the outside | `number` | n/a | yes |
-| <a name="input_deployment_maximum_percent"></a> [deployment\_maximum\_percent](#input\_deployment\_maximum\_percent) | Ceiling on running tasks during a deploy. Defaults to 200. Values <= 100 are rejected by ECS when Availability Zone Rebalancing is enabled, which it is on this cluster. | `number` | `null` | no |
+| <a name="input_deployment_maximum_percent"></a> [deployment\_maximum\_percent](#input\_deployment\_maximum\_percent) | Ceiling on running tasks during a deploy. Defaults to 200, which lets a deploy start the new task before stopping the old one. | `number` | `null` | no |
 | <a name="input_deployment_minimum_healthy_percent"></a> [deployment\_minimum\_healthy\_percent](#input\_deployment\_minimum\_healthy\_percent) | Percent of the desired count that must stay running during a deploy. Leave null to derive from `environment`: 100 for prod, 0 everywhere else. | `number` | `null` | no |
 | <a name="input_environment"></a> [environment](#input\_environment) | n/a | `string` | n/a | yes |
 | <a name="input_health_check_path"></a> [health\_check\_path](#input\_health\_check\_path) | path for load balancer health checks. This path should return HTTP 200 if the app is up. This path does not need to follow the prefix of `path`, it can be any path | `string` | `"/"` | no |

terraform/modules/container/main.tf

@@ -142,9 +142,9 @@ locals {
   # second copy of the task's memory reservation. Non-prod may drop to zero, which lets a
   # deploy proceed even when the host has no memory to spare.
   deployment_min = var.deployment_minimum_healthy_percent != null ? var.deployment_minimum_healthy_percent : (var.environment == "prod" ? 100 : 0)
-  # max stays 200 everywhere: ECS Availability Zone Rebalancing rejects maximumPercent <= 100,
-  # so a non-prod 0/100 is refused at UpdateService. 0/200 still lets a deploy fall back to
-  # stop-then-start when the host has no spare memory, rather than deadlocking on placement.
+  # max stays 200 everywhere: prod needs it to start the new task before stopping the old,
+  # and non-prod's 0/200 still lets a deploy fall back to stop-then-start when the host has
+  # no spare memory, rather than deadlocking on placement.
   deployment_max = var.deployment_maximum_percent != null ? var.deployment_maximum_percent : 200
 
   # Memory is the scarce resource on the single host. Binpacking on it keeps tasks on as few
@@ -593,6 +593,11 @@ resource "aws_ecs_service" "fargate" {
   deployment_minimum_healthy_percent = local.deployment_min
   deployment_maximum_percent         = local.deployment_max
 
+  # Off because the cluster is one host in one Availability Zone, so there is nothing to
+  # rebalance across. While it was on, ECS accepted only placement strategies that spread
+  # by Availability Zone and rejected the binpack below at UpdateService (#255).
+  availability_zone_rebalancing = "DISABLED"
+
   dynamic "ordered_placement_strategy" {
     for_each = local.placement_strategies
     content {

terraform/modules/container/variables.tf

@@ -115,7 +115,7 @@ variable "deployment_minimum_healthy_percent" {
 }
 
 variable "deployment_maximum_percent" {
-  description = "Ceiling on running tasks during a deploy. Defaults to 200. Values <= 100 are rejected by ECS when Availability Zone Rebalancing is enabled, which it is on this cluster."
+  description = "Ceiling on running tasks during a deploy. Defaults to 200, which lets a deploy start the new task before stopping the old one."
   type        = number
   default     = null
 }

terraform/projects/home-unite-us/environment-prod.tf

@@ -46,7 +46,7 @@ module "prod_service" {
     // run on -- the RDS instance is not involved despite what the abandoned Terragrunt
     // state suggests. The task definition declares no volume and no mount, so this file
     // is created empty on every task start and destroyed with the task. Application
-    // data therefore does not survive a deploy, a host drain or an AZ rebalance; the
+    // data therefore does not survive a deploy or a host drain; the
     // log group shows 743 task generations. Reproduced here deliberately so the cutover
     // changes nothing, and recorded so the next reader does not mistake it for working
     // persistence. The absolute form is used because a relative URL would depend on the