How To Use AWS IAM In Terraform

6 mins
Published on 15 February 2024

This blog post is a step-by-step guide on how to use AWS IAM in Terraform. After experiencing much pain and frustration, I finally figured out how to use AWS IAM in Terraform. This blog post will save you some time and frustration. The magical thing about this blog post is that you can do every step in this blog post locally on your machine without even owning an AWS account. This is because we will use the Localstack tool to simulate AWS locally on your machine.


To follow this blog post, you need to have the following tools installed on your machine:

  • Python 3.6 or later
  • Docker - For running Localstack
  • Localstack and development tools for interacting with Localstack
    python -m pip install terraform-local awscli-local localstack


To get started, complete the following steps; this is a one-time setup:

git clone
cd blog-how-to-use-aws-iam-in-terraform

Getting Started

IMPORTANT: From now on, the working directory should be the blog-how-to-use-aws-iam-in-terraform directory.

  1. Start Localstack - This will start a new Docker container, serving all AWS APIs locally on your machine on localhost:4566

    localstack start
  2. Create a new terminal
  3. Change to an example directory - For example examples/1-managed-arns-changes

    cd examples/1-managed-arns-changes
  4. IMPORTANT: We’ll be using tflocal, instead of terraform, communicating with Localstack.
  5. Initialize Terraform - This will download the required providers and modules.
    tflocal init
  6. Plan the changes - This will show you what Terraform will do when you apply the changes.
    tflocal plan -out plan.out
  7. Apply the changes - This will apply the changes to the Localstack container.
    tflocal apply plan.out
  8. Verify the changes - This will show the current state after changes
    tflocal show
  9. Plan and Apply the changes to check for weird behaviors
    tflocal plan -out plan.out && tflocal apply plan.out

That was it! You can now test all the examples in the examples directory. Each example is a separate Terraform configuration demonstrating a different aspect of using AWS IAM in Terraform.

Before you move on to the following example, make sure to destroy the resources created by the current example:

tflocal destroy -auto-approve

1 - Managed ARNs Changes

This example demonstrates the issue when setting the attribute managed_policy_arns in the aws_iam_role, and then creating a aws_iam_role_policy_attachment. This is a known conflict and it is also mentioned in the docs:

If you use this resource’s managed_policy_arns argument or inline_policy configuration blocks, this resource will take over exclusive management of the role’s respective policy types (e.g., both policy types if both arguments are used). These arguments are incompatible with other ways of managing a role’s policies, such as aws_iam_policy_attachment, aws_iam_role_policy_attachment, and aws_iam_role_policy. If you attempt to manage a role’s policies by multiple means, you will get resource cycling and/or errors. Source

After planning and applying this plan, you’ll experience a never-ending loop of changes, each time you plan and apply. The root cause of the loop is due aws_iam_role_policy_attachment, which attempts to change the existing attribute managed_policy_arns.

Here’s the output of the plan, for each “plan and apply”, the result, sadly, will be the same:

  # aws_iam_role.test_role will be updated in-place
  ~ resource "aws_iam_role" "test_role" {
        id                    = "test_role"
      ~ managed_policy_arns   = [
          - "arn:aws:iam::000000000000:policy/attached-policy",
            # (1 unchanged element hidden)
        name                  = "test_role"
        tags                  = {}
        # (8 unchanged attributes hidden)

Plan: 0 to add, 1 to change, 0 to destroy.

This is expected, and it is mentioned in the docs.

2 - EKS IRSA Module

The solution is rather simple, if all resources were managed in the same repository, one can adjust the resources and remove the aws_iam_role_policy_attachment resource, and instead use the inline_policy attribute in the aws_iam_role resource.

In reality, many Terraform modules will create and manage IAM Roles and Policies for you, and you might not have control over the resources. The solution is to pick the right policy attachment type and make sure that it doesn’t conflict with existing policies.

Resource NameAffected attribute in the aws_iam_role

Take a real-life example, assuming we use the Terraform module terraform-aws-modules/iam/aws//modules/iam-assumable-role-with-oidc to create an IAM Role for AWS Service Account (IRSA).

See examples/2-eks-irsa-module/, where we use the module terraform-aws-modules/iam/aws//modules/iam-assumable-role-with-oidc to create an IAM Role for AWS Service Account (IRSA), and then attach an additional IAM Policy as inline_policy.

After running tflocal init && tflocal plan -out plan.out && tflocal apply plan.out, you’ll see the following output:

# module.iam_assumable_role_with_oidc.aws_iam_role.this[0]:
resource "aws_iam_role" "this" {
    # ...
    arn                   = "arn:aws:iam::000000000000:role/role-with-oidc"
    managed_policy_arns   = [                   # <--- This is the interesting part
    # ...

To inspect the current state, run:

tflocal show

Assuming I want to add a policy to the role, I’ve realized the managed_policy_arns is already occupied by the module, and I can’t use aws_iam_role_policy_attachment or aws_iam_policy_attachment to attach the policy, as it will cause a conflict. So I must use aws_iam_role_policy to attach the policy as inline_policy, this will be done in the next example.

3 - EKS IRSA Module - Managed ARNs Changes - Solution

Finally, the solution is to use aws_iam_role_policy to attach the policy as inline_policy. This will not cause any conflicts, as the inline_policy attribute is not affected by the managed_policy_arns attribute.

# IAM Policy that will be attached to the role with "aws_iam_role_policy" to inline_policy
resource "aws_iam_policy" "attached_policy" {
  name        = "attached-policy"
  description = "A policy that will be attached to the role with aws_iam_role_policy to inline_policy"
  policy = jsonencode({
    Version = "2012-10-17"
    Statement = [
        Action = [
        Effect = "Allow"
        Resource = [

# Affects inline_policy, hence doesn't cause an issue this setup
resource "aws_iam_role_policy" "attached_policy" {
  name   = "attached_policy"
  role   = module.iam_assumable_role_with_oidc.iam_role_name
  policy = aws_iam_policy.attached_policy.policy

This example results in the following output:

No changes. Your infrastructure matches the configuration.

The module iam-assumable-role-with-oidc is not the perfect example, as it uses managed_policy_arns and inline_policy properly in aws_iam_role, and by properly, I mean not setting them at all, allowing aws_iam_role_policy_attachment and aws_iam_role_policyto be used safely, see check how iam-assuamble-role-with-oidc defines an aws_iam_role.

As you already know, each module that creates an aws_iam_role should be inspected thoroughly. If the module is configured with no inline_policy and managed_policy_arns, it is safe to use the IAM Policy attachment without fearing conflict.


In many cases, it’s possible to pass variables to a module and let the module create the IAM Role and Policy for you. However, in some cases, you may need to attach additional policies to the role from somewhere else in the infrastructure, from another stack or state, and you may not have control over the module. In such cases, you must know how to attach policies to the role and ensure that you don’t cause conflicts.

Ideally, all IAM Policies and Roles should be managed in the same state, and the “perfect solution” resides in examples/4-perfect, where all resources are managed in the same state. But this is far from reality, and that can only be done at your codebase level; you’ll always need to handle outputs from other stacks and attach policies to roles from different stacks.

Originally published at

Related Posts