ci(infra): Optimize ECS (#627)

TracecatHQ · Dec 19, 2024 · c14229f · c14229f
1 parent 4dba340
commit c14229f
Show file tree

Hide file tree

Showing 7 changed files with 62 additions and 2 deletions.
diff --git a/deployments/aws/ecs/ecs-api.tf b/deployments/aws/ecs/ecs-api.tf
@@ -8,6 +8,11 @@ resource "aws_ecs_task_definition" "api_task_definition" {
   execution_role_arn       = aws_iam_role.api_execution.arn
   task_role_arn            = aws_iam_role.api_worker_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TracecatApiContainer"
@@ -34,6 +39,13 @@ resource "aws_ecs_task_definition" "api_task_definition" {
         maxAttempts = 3
         backoffTime = 10
       }
+      healthCheck = {
+        command     = ["CMD", "python", "-c", "import httpx; httpx.get('http://localhost:8000/').raise_for_status()"]
+        interval    = 30
+        timeout     = 5
+        retries     = 3
+        startPeriod = 60
+      }
     }
   ])
 }

diff --git a/deployments/aws/ecs/ecs-caddy.tf b/deployments/aws/ecs/ecs-caddy.tf
@@ -8,6 +8,11 @@ resource "aws_ecs_task_definition" "caddy_task_definition" {
   execution_role_arn       = aws_iam_role.caddy_execution.arn
   task_role_arn            = aws_iam_role.caddy_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TracecatCaddyContainer"

diff --git a/deployments/aws/ecs/ecs-executor.tf b/deployments/aws/ecs/ecs-executor.tf
@@ -8,6 +8,11 @@ resource "aws_ecs_task_definition" "executor_task_definition" {
   execution_role_arn       = aws_iam_role.worker_execution.arn
   task_role_arn            = aws_iam_role.api_worker_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TracecatExecutorContainer"
@@ -44,6 +49,13 @@ resource "aws_ecs_task_definition" "executor_task_definition" {
         maxAttempts = 3
         backoffTime = 10
       }
+      healthCheck = {
+        command     = ["CMD", "python", "-c", "import httpx; httpx.get('http://localhost:8002/').raise_for_status()"]
+        interval    = 30
+        timeout     = 5
+        retries     = 3
+        startPeriod = 60
+      }
     }
   ])
 }

diff --git a/deployments/aws/ecs/ecs-temporal-ui.tf b/deployments/aws/ecs/ecs-temporal-ui.tf
@@ -10,6 +10,11 @@ resource "aws_ecs_task_definition" "temporal_ui_task_definition" {
   execution_role_arn       = aws_iam_role.temporal_ui_execution.arn
   task_role_arn            = aws_iam_role.temporal_ui_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TemporalUiContainer"

diff --git a/deployments/aws/ecs/ecs-temporal.tf b/deployments/aws/ecs/ecs-temporal.tf
@@ -8,6 +8,11 @@ resource "aws_ecs_task_definition" "temporal_task_definition" {
   execution_role_arn       = aws_iam_role.temporal_execution.arn
   task_role_arn            = aws_iam_role.temporal_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TemporalContainer"

diff --git a/deployments/aws/ecs/ecs-ui.tf b/deployments/aws/ecs/ecs-ui.tf
@@ -7,6 +7,11 @@ resource "aws_ecs_task_definition" "ui_task_definition" {
   memory                   = var.ui_memory
   execution_role_arn       = aws_iam_role.ui_execution.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name  = "TracecatUiContainer"

diff --git a/deployments/aws/ecs/ecs-worker.tf b/deployments/aws/ecs/ecs-worker.tf
@@ -8,6 +8,11 @@ resource "aws_ecs_task_definition" "worker_task_definition" {
   execution_role_arn       = aws_iam_role.worker_execution.arn
   task_role_arn            = aws_iam_role.api_worker_task.arn
 
+  runtime_platform {
+    operating_system_family = "LINUX"
+    cpu_architecture        = "ARM64"
+  }
+
   container_definitions = jsonencode([
     {
       name    = "TracecatWorkerContainer"
@@ -43,8 +48,7 @@ resource "aws_ecs_service" "tracecat_worker" {
   name                 = "tracecat-worker"
   cluster              = aws_ecs_cluster.tracecat_cluster.id
   task_definition      = aws_ecs_task_definition.worker_task_definition.arn
-  launch_type          = "FARGATE"
-  desired_count        = 1
+  desired_count        = 2
   force_new_deployment = var.force_new_deployment
 
   network_configuration {
@@ -80,6 +84,18 @@ resource "aws_ecs_service" "tracecat_worker" {
     }
   }
 
+  capacity_provider_strategy {
+    capacity_provider = "FARGATE"
+    weight            = 1
+    base              = 1
+  }
+
+  capacity_provider_strategy {
+    capacity_provider = "FARGATE_SPOT"
+    weight            = 1
+    base              = 0
+  }
+
   depends_on = [
     aws_ecs_service.temporal_service,
     aws_ecs_service.tracecat_executor