feat: allow stacks to change cpu/mem requests (#2614)

chanzuckerberg · Oct 20, 2023 · a1affe8 · a1affe8
1 parent 964e735
commit a1affe8
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 1 deletion.
diff --git a/terraform/modules/happy-stack-eks/README.md b/terraform/modules/happy-stack-eks/README.md
@@ -59,7 +59,7 @@
 | <a name="input_image_tags"></a> [image\_tags](#input\_image\_tags) | Override image tag for each docker image | `map(string)` | `{}` | no |
 | <a name="input_k8s_namespace"></a> [k8s\_namespace](#input\_k8s\_namespace) | K8S namespace for this stack | `string` | n/a | yes |
 | <a name="input_routing_method"></a> [routing\_method](#input\_routing\_method) | Traffic routing method for this stack. Valid options are 'DOMAIN', when every service gets a unique domain name, or a 'CONTEXT' when all services share the same domain name, and routing is done by request path. | `string` | `"DOMAIN"` | no |
-| <a name="input_services"></a> [services](#input\_services) | The services you want to deploy as part of this stack. | <pre>map(object({<br>    name : string,<br>    service_type : optional(string, "INTERNAL"),<br>    allow_mesh_services : optional(list(object({<br>      service : optional(string, null),<br>      stack : optional(string, null),<br>      service_account_name : optional(string, null)<br>    })), null),<br>    ingress_security_groups : optional(list(string), []), // Only used for VPC service_type<br>    alb : optional(object({<br>      name : string,<br>      listener_port : number,<br>    }), null), // Only used for TARGET_GROUP_ONLY<br>    desired_count : optional(number, 2),<br>    max_count : optional(number, 2),<br>    scaling_cpu_threshold_percentage : optional(number, 80),<br>    port : optional(number, 80),<br>    scheme : optional(string, "HTTP"),<br>    cmd : optional(list(string), []),<br>    args : optional(list(string), []),<br>    image_pull_policy : optional(string, "IfNotPresent"), // Supported values: IfNotPresent, Always, Never<br>    tag_mutability : optional(bool, true),<br>    scan_on_push : optional(bool, false),<br>    service_port : optional(number, null),<br>    service_scheme : optional(string, "HTTP"),<br>    memory : optional(string, "100Mi"),<br>    cpu : optional(string, "100m"),<br>    gpu : optional(number, null), // Whole number of GPUs to request, 0 will schedule all available GPUs. Requires GPU-enabled nodes in the cluster, `k8s-device-plugin` installed, platform_architecture = "amd64", and additional_node_selectors = { "nvidia.com/gpu.present" = "true" } present.<br>    health_check_path : optional(string, "/"),<br>    aws_iam : optional(object({<br>      policy_json : optional(string, ""),<br>      service_account_name : optional(string, null),<br>    }), {}),<br>    path : optional(string, "/*"),  // Only used for CONTEXT and TARGET_GROUP_ONLY routing<br>    priority : optional(number, 0), // Only used for CONTEXT and TARGET_GROUP_ONLY routing<br>    success_codes : optional(string, "200-499"),<br>    synthetics : optional(bool, false),<br>    initial_delay_seconds : optional(number, 30),<br>    alb_idle_timeout : optional(number, 60) // in seconds<br>    period_seconds : optional(number, 3),<br>    platform_architecture : optional(string, "amd64"),     // Supported values: amd64, arm64; GPU nodes are amd64 only.<br>    additional_node_selectors : optional(map(string), {}), // For GPU use: { "nvidia.com/gpu.present" = "true" }<br>    bypasses : optional(map(object({                       // Only used for INTERNAL service_type<br>      paths   = optional(set(string), [])<br>      methods = optional(set(string), [])<br>    })), {})<br>    sidecars : optional(map(object({<br>      image : string<br>      tag : string<br>      port : optional(number, 80),<br>      scheme : optional(string, "HTTP"),<br>      memory : optional(string, "100Mi")<br>      cpu : optional(string, "100m")<br>      image_pull_policy : optional(string, "IfNotPresent") // Supported values: IfNotPresent, Always, Never<br>      health_check_path : optional(string, "/")<br>      initial_delay_seconds : optional(number, 30),<br>      period_seconds : optional(number, 3),<br>    })), {})<br>  }))</pre> | n/a | yes |
+| <a name="input_services"></a> [services](#input\_services) | The services you want to deploy as part of this stack. | <pre>map(object({<br>    name : string,<br>    service_type : optional(string, "INTERNAL"),<br>    allow_mesh_services : optional(list(object({<br>      service : optional(string, null),<br>      stack : optional(string, null),<br>      service_account_name : optional(string, null)<br>    })), null),<br>    ingress_security_groups : optional(list(string), []), // Only used for VPC service_type<br>    alb : optional(object({<br>      name : string,<br>      listener_port : number,<br>    }), null), // Only used for TARGET_GROUP_ONLY<br>    desired_count : optional(number, 2),<br>    max_count : optional(number, 2),<br>    scaling_cpu_threshold_percentage : optional(number, 80),<br>    port : optional(number, 80),<br>    scheme : optional(string, "HTTP"),<br>    cmd : optional(list(string), []),<br>    args : optional(list(string), []),<br>    image_pull_policy : optional(string, "IfNotPresent"), // Supported values: IfNotPresent, Always, Never<br>    tag_mutability : optional(bool, true),<br>    scan_on_push : optional(bool, false),<br>    service_port : optional(number, null),<br>    service_scheme : optional(string, "HTTP"),<br>    memory : optional(string, "100Mi"),<br>    memory_requests : optional(string, "100Mi"),<br>    cpu : optional(string, "100m"),<br>    cpu_requests : optional(string, "100m"),<br>    gpu : optional(number, null), // Whole number of GPUs to request, 0 will schedule all available GPUs. Requires GPU-enabled nodes in the cluster, `k8s-device-plugin` installed, platform_architecture = "amd64", and additional_node_selectors = { "nvidia.com/gpu.present" = "true" } present.<br>    health_check_path : optional(string, "/"),<br>    aws_iam : optional(object({<br>      policy_json : optional(string, ""),<br>      service_account_name : optional(string, null),<br>    }), {}),<br>    path : optional(string, "/*"),  // Only used for CONTEXT and TARGET_GROUP_ONLY routing<br>    priority : optional(number, 0), // Only used for CONTEXT and TARGET_GROUP_ONLY routing<br>    success_codes : optional(string, "200-499"),<br>    synthetics : optional(bool, false),<br>    initial_delay_seconds : optional(number, 30),<br>    alb_idle_timeout : optional(number, 60) // in seconds<br>    period_seconds : optional(number, 3),<br>    platform_architecture : optional(string, "amd64"),     // Supported values: amd64, arm64; GPU nodes are amd64 only.<br>    additional_node_selectors : optional(map(string), {}), // For GPU use: { "nvidia.com/gpu.present" = "true" }<br>    bypasses : optional(map(object({                       // Only used for INTERNAL service_type<br>      paths   = optional(set(string), [])<br>      methods = optional(set(string), [])<br>    })), {})<br>    sidecars : optional(map(object({<br>      image : string<br>      tag : string<br>      port : optional(number, 80),<br>      scheme : optional(string, "HTTP"),<br>      memory : optional(string, "100Mi")<br>      cpu : optional(string, "100m")<br>      image_pull_policy : optional(string, "IfNotPresent") // Supported values: IfNotPresent, Always, Never<br>      health_check_path : optional(string, "/")<br>      initial_delay_seconds : optional(number, 30),<br>      period_seconds : optional(number, 3),<br>    })), {})<br>  }))</pre> | n/a | yes |
 | <a name="input_skip_config_injection"></a> [skip\_config\_injection](#input\_skip\_config\_injection) | Skip injecting app configs into the services / tasks | `bool` | `false` | no |
 | <a name="input_stack_name"></a> [stack\_name](#input\_stack\_name) | Happy Path stack name | `string` | n/a | yes |
 | <a name="input_stack_prefix"></a> [stack\_prefix](#input\_stack\_prefix) | Do bucket storage paths and db schemas need to be prefixed with the stack name? (Usually '/{stack\_name}' for dev stacks, and '' for staging/prod stacks) | `string` | `""` | no |

diff --git a/terraform/modules/happy-stack-eks/main.tf b/terraform/modules/happy-stack-eks/main.tf
@@ -169,7 +169,9 @@ module "services" {
   max_count                        = try(each.value.max_count, each.value.desired_count)
   scaling_cpu_threshold_percentage = each.value.scaling_cpu_threshold_percentage
   memory                           = each.value.memory
+  memory_requests                  = each.value.memory_requests
   cpu                              = each.value.cpu
+  cpu_requests                     = each.value.cpu_requests
   gpu                              = each.value.gpu
   health_check_path                = each.value.health_check_path
   k8s_namespace                    = var.k8s_namespace

diff --git a/terraform/modules/happy-stack-eks/variables.tf b/terraform/modules/happy-stack-eks/variables.tf
@@ -69,7 +69,9 @@ variable "services" {
     service_port : optional(number, null),
     service_scheme : optional(string, "HTTP"),
     memory : optional(string, "100Mi"),
+    memory_requests : optional(string, "100Mi"),
     cpu : optional(string, "100m"),
+    cpu_requests : optional(string, "100m"),
     gpu : optional(number, null), // Whole number of GPUs to request, 0 will schedule all available GPUs. Requires GPU-enabled nodes in the cluster, `k8s-device-plugin` installed, platform_architecture = "amd64", and additional_node_selectors = { "nvidia.com/gpu.present" = "true" } present.
     health_check_path : optional(string, "/"),
     aws_iam : optional(object({