Terraform replacement: prove old and new can coexist

Operational decision

A receipt worker pool needs a launch-template change. In an isolated environment, plan the replacement and list every new instance, address, disk, and dependent object that must exist while the old pool serves traffic. Reserve quota for overlap, use distinct names where required, and verify readiness before redirecting work. The fragment is only an operation-order rule; the deployment still needs a data and traffic migration. If the object is a stateful database, decide whether replication, snapshot restore, or a maintenance window is needed before selecting the rule. Simulate failure after creation but before old-resource deletion and record how to identify and retire the stray replacement. Compare the final plan with actual provider inventory after apply. A prevent-destroy guard can block an accidental replacement while present, but removing that resource block removes the guard's protection.

resource "aws_autoscaling_group" "receipt_workers" {
  name_prefix         = "receipt-workers-"
  max_size            = 6
  min_size            = 3
  desired_capacity    = 3
  vpc_zone_identifier = var.worker_subnet_ids

  lifecycle {
    create_before_destroy = true
  }
}

Cost and verification

Overlap can temporarily double compute, addresses, volume attachments, and load on a shared database. Quota requests and data synchronization may take longer than the apply itself. Destroy-first avoids overlap but can create downtime; create-first reduces that window only when both generations can coexist and cutover is verified. Track peak resource count, ready capacity, old-generation traffic, and orphan count. A low-cost test should deliberately exhaust one quota to prove the rollback path retains the old generation.

Common Mistakes

• Do not assume create-before-destroy solves a globally unique name conflict.
• Do not rely on prevent-destroy after removing the resource block.
• Do not mark the replacement complete before traffic and data have moved.

Connected lessons

• DevOps: delivery, infrastructure, and reliable operations
• Terraform state: shared ownership and safe plans
• Provider quota preflight: reserve capacity for rollback and recovery
• Graceful Pod shutdown: stop accepting work before exit
• Rollback image retention: keep every approved fallback pullable

Practice and check

• Project: hand off Terraform-managed infrastructure without a replacement surprise
• DevOps Terraform ownership decisions

Continue with: Infrastructure plans: classify every action before apply.