Skip to content

Manage terraform state backends for multiple stacks. Generates backend file for each stack managed. Stacks can consist of multiple tfstate files, and state lock is at stack level for optimum protection of stack state.

License

schollii/terraform-aws-multi-stack-backends

Repository files navigation

This terraform module makes it easy to boostrap the storage of terraform state remote backends in AWS S3. It is basically an alternative to using Terraform Cloud, insofar as tfstate storage goes.

I created this module because I got tired of having to do a song and dance every time I was standing up a new stack: create terraform code in a separate folder from the stack code to create several AWS resources specific to the stack (namely its own bucket for storing its tfstate, its own lock table etc), run terraform init and apply, then create a backend.tf file for that folder and run terraform init -migrate-state, THEN create a backends.tf file in the stack's folder with the right values based on the tfstate folder. That got tiring really quickly, and I didn't want to have to store tfstate in Terraform Cloud mostly because I didn't like the idea of depending on a third party (relative to AWS) to host the all-important tfstate which contains sensitive information.

And then when your stacks get large enough, you'll want to subdivide them into smaller "substacks", eg one for the VPC/networking, one for some servers, one for databases, etc.

So this module automates much of this toil. In its simplest form, you use this module inside a " tfstate backends manager" root module that you create, you just have to give it a simple mapping of (arbitrary) stack names to their paths on the filesystem. Then run terraform apply, and finally run terraform init -migrate-state -force-copy for each stack and the manager module itself (so even the manager stores its state in S3!) and you're DONE! You can go about your business of writing your infra code! AND the bucket storing your tfstates will have auto-replication to a different region, both buckets will be server-side encrypted and versioned, each stack will have its own lock, etc.

This "tfstate backends manager" module (maybe that's what I should have called it) supports more advanced uses as well:

  • any stack can have substacks; each will have its own key in the S3 bucket
  • you could have multiple "managers" in one AWS account, each can act as the host for a separate set of stacks
  • the module can generate IAM policies for each stack that limit access to that stacks' tfstate only in either read-only mode or read-write; you can then attach those policies to your IAM users/roles/groups (you do that part)

The list of stacks to manage is a tree-like map:

stack ID -> sub-stack ID -> information about the stack or sub-stack

Currently the info consists only of the filesystem path, but more info could be added in the future. When a stack does not have sub-stacks, you define just one sub-stack ID, and give it a name like "all" or "main".

Examples

Example:

# your main.tf for the tfstate manager
module "tfstate_manager" {
  source = "schollii/multi-stack-backends/aws"

  stacks_map = {
    stack-1 = {
      all = {
        path = "../stack1"
      }
    },
    stack-2 = {
      network = {
        path = "../stack2-network"
      }
      cluster = {
        path = "../stack2-cluster"
      }
    },
  }
}

The keys like "network" and "cluster" are entirely up to you. See the examples/simple/README.md for details including diagrams that illustrate the different pieces managed by this module.

There isn't much more to using this module than the above and the examples. The remaining sections provide useful info for various tasks.

Renaming a tfstates manager bucket

It may happen that the backends manager bucket needs to be renamed. Eg, following some naming policy changes in your organization. Unfortunately AWS does not make this easy: buckets cannot be renamed. The following procedure is one that I use so that all AWS resources get renamed according to the new bucket name:

  1. Ensure that no one else will be running terraform apply on either the manager or any of its stacks / sub-stacks.
  2. Change the value of backends_bucket_name used by the backends manager terraform code (this new value is referred to here as NEW_BACKENDS_BUCKET_NAME).
  3. Examine scripts/rename-backends-manager-bucket.sh to understand what it does. Test it out first on your system, eg on the examples/simple/tfstate-s3-manager folder. (REMINDER: per the license terms described in this git repo, this script is provided as-is without any warranty implied - it is your responsibility to ensure that it will not cause loss of data!).
  4. Run that script scripts/rename-backends-manager-bucket.sh NEW_BUCKET_NAME.
  5. Any of your terraform_remote_state data sources that point to the old bucket name of this manager need to be have their bucket argument adjusted.
  6. Run terraform plan in all of the stacks managed by this backends manager, and confirm that no init is needed and no changes are planned.
  7. Once you are satisfied that the new setup is working, you can manually delete the 2 old buckets.

That should be it!

Upgrades

This section documents upgrades necessary between adjacent major releases.

0.6.x to 1.0

  1. (Optional) The module no longer assumes that IAM policies for the tfstate access are required. These policies are not needed by the module, they are merely a convenience (making it easy for you to control access to the tfstates stored in the backend bucket). To generate the policies as in 0.6, set one or both create_tfstate_access_polic* variables to true, depending on your needs.
  2. (Optional) The module no longer assumes a default set of tags. Only var.extra_tags is used, and by default it is empty. The terraform plan will clearly say what the old values were, if you need some of them.
  3. Rename variable this_tfstate_in_s3 to manager_tfstate_in_s3
  4. The module now requires a value for backends_bucket_name (it no longer provides a default). This won't likely affect you because you almost certainly had to specify a bucket name anyway, due to AWS uniqueness constraints on S3 bucket names. But if you were able to not specify it (eg if you were the first to use that default bucket name in yours AWS partition), you must now add backends_bucket_name = "tfstate-s3-backends". However, I highly recommend that you rename your backends bucket to something unique to you or your organization / employer. Eg ORG_NAME-GROUP_NAME-tfstate-backends. The procedure to do this is in a separate section of this readme.
  5. The module itself no longer defines providers, as recommended in the terraform documentation. Rather, it requires they be passed in to the module. Therefore, before applying,
    • ensure you have these two blocks (as done in examples/simple/tfstate-s3-manager/terraform.tf):
      provider "aws" {
        alias  = "tfstate_backends"
        region = "us-east-1"
      }
      
      provider "aws" {
        alias  = "tfstate_backends_replica"
        region = "us-west-1"
      }
    • add the following to the module block that references this module (as done in examples/simple/tfstate-s3-manager/terraform.tf):
      providers = {
        aws = aws.tfstate_backends
        aws.replica = aws.tfstate_backends_replica
      }
  6. (Optional) The module now uses the aws_s3_bucket_* resources instead of the inline blocks that the AWS provider has deprecated, like acl, server-side encryption, etc. This will cause terraform to plan creation of those resources. This is ok, but if you prefer (like me) not to, you can import them into the tfstate. I use the bash script scripts/upgrade-to-1.0.sh. Again before using it, examine it and check that it will work with your version of bash and your setup. Testing is easy: add an "echo" in front of the terraform commands.

Acknowledgements

My code used some of https://github.com/nozaq/terraform-aws-remote-state-s3-backend as a starting point.

About

Manage terraform state backends for multiple stacks. Generates backend file for each stack managed. Stacks can consist of multiple tfstate files, and state lock is at stack level for optimum protection of stack state.

Topics

Resources

License

Stars

Watchers

Forks

Packages

No packages published