chore: lambda otel (#16)

* chore: otel for lambdas

* chore: set otel sampling rates

* fix: fmt

* chore: mv aws_lambda_layers_account_id

* fix: pass in aws_lambda_layers_account_id var

* fix: use arn string for otel layer

* fix: layers not needed

Author: chrispsheehan

infra/live/global_vars.hcl

@@ -10,7 +10,8 @@ locals {
     "application-autoscaling:*",
     "cloudwatch:*",
     "sqs:*",
-    "cloudfront:*"
+    "cloudfront:*",
+    "xray:*"
   ]
 }
 

infra/live/prod/environment_vars.hcl

@@ -6,4 +6,5 @@ locals {
 inputs = {
   log_retention_days = local.log_retention_days
   deploy_branches    = local.deploy_branches
+  otel_sample_rate   = 0.1 # 10% of traces sampled
 }

infra/modules/aws/_shared/lambda/data.tf

@@ -1,3 +1,14 @@
+data "aws_iam_policy_document" "lambda_xray" {
+  statement {
+    effect = "Allow"
+    actions = [
+      "xray:PutTraceSegments",
+      "xray:PutTelemetryRecords",
+    ]
+    resources = ["*"]
+  }
+}
+
 data "aws_s3_bucket" "code_bucket" {
   bucket = var.code_bucket
 }

infra/modules/aws/_shared/lambda/main.tf

@@ -30,12 +30,21 @@ resource "aws_s3_object" "bootstrap_lambda_zip" {
   content_type = "application/zip"
 }
 
-resource "aws_lambda_function" "lambda" {
-  function_name = local.lambda_name
-  role          = aws_iam_role.iam_for_lambda.arn
-  handler       = local.lambda_handler
-  runtime       = local.lambda_runtime
+resource "aws_iam_policy" "lambda_xray" {
+  name   = "${local.lambda_name}-xray"
+  policy = data.aws_iam_policy_document.lambda_xray.json
+}
+
+resource "aws_iam_role_policy_attachment" "lambda_xray" {
+  role       = aws_iam_role.iam_for_lambda.name
+  policy_arn = aws_iam_policy.lambda_xray.arn
+}
 
+resource "aws_lambda_function" "lambda" {
+  function_name                  = local.lambda_name
+  role                           = aws_iam_role.iam_for_lambda.arn
+  handler                        = local.lambda_handler
+  runtime                        = local.lambda_runtime
   reserved_concurrent_executions = local.pc_reserved_count
 
   s3_bucket = data.aws_s3_bucket.code_bucket.bucket
@@ -44,8 +53,15 @@ resource "aws_lambda_function" "lambda" {
   # publish ONE immutable version so we can create an alias
   publish = true
 
+  tracing_config {
+    mode = "Active"
+  }
+
   environment {
-    variables = var.environment_variables
+    variables = merge(var.environment_variables, {
+      OTEL_TRACES_SAMPLER     = "parentbased_traceidratio"
+      OTEL_TRACES_SAMPLER_ARG = tostring(var.otel_sample_rate)
+    })
   }
 
   # tags for identifying the code deploy app and its deployment config. Used in CI/CD pipelines.

infra/modules/aws/_shared/lambda/variables.tf

@@ -1,4 +1,10 @@
 ### start of static vars set in root.hcl ###
+variable "otel_sample_rate" {
+  type        = number
+  description = "OpenTelemetry trace sampling rate — 0.0 = no traces, 1.0 = 100% of traces sampled"
+  default     = 1.0 # 100%
+}
+
 variable "project_name" {
   type        = string
   description = "Project name used in naming resources"

infra/modules/aws/api/main.tf

@@ -1,9 +1,10 @@
 module "lambda_api" {
   source = "../_shared/lambda"
 
-  project_name = var.project_name
-  environment  = var.environment
-  code_bucket  = var.code_bucket
+  project_name     = var.project_name
+  environment      = var.environment
+  code_bucket      = var.code_bucket
+  otel_sample_rate = var.otel_sample_rate
 
   lambda_name = local.lambda_name
 

infra/modules/aws/api/variables.tf

@@ -13,6 +13,12 @@ variable "code_bucket" {
   type        = string
   description = "Bucket where deployable code artifacts are uploaded"
 }
+
+variable "otel_sample_rate" {
+  type        = number
+  description = "OpenTelemetry trace sampling rate — 0.0 = no traces, 1.0 = 100% of traces sampled"
+  default     = 1.0 # 100%
+}
 ### end of static vars set in root.hcl ###
 
 variable "deployment_config" {

infra/modules/aws/lambda_worker/main.tf

@@ -1,9 +1,10 @@
 module "lambda_worker" {
   source = "../_shared/lambda"
 
-  project_name = var.project_name
-  environment  = var.environment
-  code_bucket  = var.code_bucket
+  project_name     = var.project_name
+  environment      = var.environment
+  code_bucket      = var.code_bucket
+  otel_sample_rate = var.otel_sample_rate
 
   lambda_name = local.lambda_name
 

infra/modules/aws/lambda_worker/variables.tf

@@ -13,6 +13,12 @@ variable "code_bucket" {
   type        = string
   description = "Bucket where deployable code artifacts are uploaded"
 }
+
+variable "otel_sample_rate" {
+  type        = number
+  description = "OpenTelemetry trace sampling rate — 0.0 = no traces, 1.0 = 100% of traces sampled"
+  default     = 1.0 # 100%
+}
 ### end of static vars set in root.hcl ###
 
 variable "sqs_queue_name" {