Terraform Provider GSI

Overview

This provider is a workaround to a limitation of the AWS provider (and TF language) which prevents the management of global secondary indexes associated with an autoscaler.

The approach is to decouple the table resource (still owned by the AWS provider) from the GSI resource while ignoring the GSIs on the table.

provider "gsi" {}

resource "aws_dynamodb_table" "test_table" {
  name           = "test_table"
  read_capacity  = 5
  write_capacity = 5
  hash_key       = "UserId"

  attribute {
    name = "UserId"
    type = "S"
  }

  lifecycle {
    ignore_changes = [global_secondary_index]
  }
}

resource "gsi_global_secondary_index" "test_index" {
  name       = "test_index"
  table_name = aws_dynamodb_table.test_table.name

  hash_key       = "UserId"
  hash_key_type  = "S"
  range_key      = "OrderId"
  range_key_type = "S"

  read_capacity   = 5
  write_capacity  = 5
  projection_type = "KEYS_ONLY"
}

If you have an autoscaler (the whole point of using this resource), consider adding a depends_on the GSIs since the autoscaler cannot reference a GSI that does not exits yet.

Since you might have a lot of existing GSIs already, you can use auto_import = true in the provider configuration and then remove it once the migration is done. When set, the first create will automatically import the GSI if one with the same name exists. Note that it will not attempt to correct drift so it might be a two step process to get to a clean plan.

Build

Run the following command to build the provider

go build -o terraform-provider-gsi

Run acceptance tests

(cd docker-compose; docker-compose up -d)
make testacc

Run unit tests

make test