sfn.tf

# ──────────────────────────────────────────────
# Step Functions State Machine
#
# Uses .sync:2 integration so Step Functions
# waits for each ECS task to finish before
# moving to the next state — no polling needed.
#
# Each step overrides the container command to
# run a different Python file. All share the
# same task definition and Docker image.
# ──────────────────────────────────────────────

locals {
  # Reusable network config block for all ECS steps
  ecs_network_config = {
    AwsvpcConfiguration = {
      Subnets        = var.subnet_ids
      SecurityGroups = var.security_group_ids
      AssignPublicIp = "DISABLED"
    }
  }
}

resource "aws_sfn_state_machine" "main" {
  name     = "${local.prefix}-state-machine"
  role_arn = aws_iam_role.sfn.arn
  type     = "STANDARD"

  logging_configuration {
    log_destination        = "${aws_cloudwatch_log_group.sfn.arn}:*"
    include_execution_data = true
    level                  = "ERROR"
  }

  definition = jsonencode({
    Comment = "Runs script_1, script_2, script_3 sequentially on Fargate"
    StartAt = "RunScript1"

    States = {

      RunScript1 = {
        Type     = "Task"
        Resource = "arn:aws:states:::ecs:runTask.sync:2"
        Parameters = {
          LaunchType     = "FARGATE"
          Cluster        = aws_ecs_cluster.main.arn
          TaskDefinition = aws_ecs_task_definition.app.arn
          NetworkConfiguration = local.ecs_network_config
          Overrides = {
            ContainerOverrides = [{
              Name    = "app"
              Command = ["script_1.py"]
            }]
          }
        }
        ResultPath = "$.script1Result"
        Retry = [{
          ErrorEquals     = ["States.TaskFailed"]
          IntervalSeconds = 30
          MaxAttempts     = 2
          BackoffRate     = 2.0
        }]
        Catch = [{
          ErrorEquals = ["States.ALL"]
          Next        = "JobFailed"
          ResultPath  = "$.error"
        }]
        Next = "RunScript2"
      }

      RunScript2 = {
        Type     = "Task"
        Resource = "arn:aws:states:::ecs:runTask.sync:2"
        Parameters = {
          LaunchType     = "FARGATE"
          Cluster        = aws_ecs_cluster.main.arn
          TaskDefinition = aws_ecs_task_definition.app.arn
          NetworkConfiguration = local.ecs_network_config
          Overrides = {
            ContainerOverrides = [{
              Name    = "app"
              Command = ["script_2.py"]
            }]
          }
        }
        ResultPath = "$.script2Result"
        Retry = [{
          ErrorEquals     = ["States.TaskFailed"]
          IntervalSeconds = 30
          MaxAttempts     = 2
          BackoffRate     = 2.0
        }]
        Catch = [{
          ErrorEquals = ["States.ALL"]
          Next        = "JobFailed"
          ResultPath  = "$.error"
        }]
        Next = "RunScript3"
      }

      RunScript3 = {
        Type     = "Task"
        Resource = "arn:aws:states:::ecs:runTask.sync:2"
        Parameters = {
          LaunchType     = "FARGATE"
          Cluster        = aws_ecs_cluster.main.arn
          TaskDefinition = aws_ecs_task_definition.app.arn
          NetworkConfiguration = local.ecs_network_config
          Overrides = {
            ContainerOverrides = [{
              Name    = "app"
              Command = ["script_3.py"]
            }]
          }
        }
        ResultPath = "$.script3Result"
        Retry = [{
          ErrorEquals     = ["States.TaskFailed"]
          IntervalSeconds = 30
          MaxAttempts     = 2
          BackoffRate     = 2.0
        }]
        Catch = [{
          ErrorEquals = ["States.ALL"]
          Next        = "JobFailed"
          ResultPath  = "$.error"
        }]
        Next = "JobSucceeded"
      }

      JobSucceeded = {
        Type = "Succeed"
      }

      JobFailed = {
        Type  = "Fail"
        Error = "JobFailed"
        Cause = "A script step failed after retries. Check CloudWatch logs."
      }
    }
  })
}

# ──────────────────────────────────────────────
# Outputs
# ──────────────────────────────────────────────
output "state_machine_arn" {
  description = "ARN of the Step Functions state machine"
  value       = aws_sfn_state_machine.main.arn
}

output "state_machine_name" {
  description = "Name of the Step Functions state machine"
  value       = aws_sfn_state_machine.main.name
}