Opscraft
Terraform · Lesson 5 of 12 · 4 min read

Meta-arguments and lifecycle

count vs for_each

# count — indexed list
resource "aws_iam_user" "dev" {
  count = length(var.developers)          # ["asha", "bala", "chitra"]
  name  = var.developers[count.index]
}
# addresses: aws_iam_user.dev[0], [1], [2]

# for_each — keyed by map keys or set values
resource "aws_iam_user" "dev" {
  for_each = toset(var.developers)
  name     = each.value
}
# addresses: aws_iam_user.dev["asha"], ["bala"], ["chitra"]

The count index problem: remove "asha" from the list and with count every later element shifts index — Terraform sees [0] changed from asha to bala, [1] from bala to chitra, and [2] deleted: it modifies or recreates resources you didn't touch. With for_each, removing "asha" deletes only dev["asha"].

Rules of thumb:

  • for_each for collections of distinct things (subnets, users, buckets, services).
  • count for "zero or one" conditional creation, or truly identical copies.
resource "aws_cloudwatch_log_group" "flow_logs" {
  count = var.enable_flow_logs ? 1 : 0
  name  = "/vpc/${local.name_prefix}/flow-logs"
}
# reference: aws_cloudwatch_log_group.flow_logs[0].arn  (or one(aws_cloudwatch_log_group.flow_logs[*].arn))

Subnets from a map of objects:

variable "private_subnets" {
  default = {
    a = { cidr = "10.20.0.0/19",  az = "ap-south-1a" }
    b = { cidr = "10.20.32.0/19", az = "ap-south-1b" }
    c = { cidr = "10.20.64.0/19", az = "ap-south-1c" }
  }
}
resource "aws_subnet" "private" {
  for_each          = var.private_subnets
  vpc_id            = aws_vpc.main.id
  cidr_block        = each.value.cidr
  availability_zone = each.value.az
  tags = { Name = "${local.name_prefix}-private-${each.key}", "kubernetes.io/role/internal-elb" = "1" }
}

for_each keys must be known at plan time — you can't key on an attribute only known after apply (like a generated ID); you'll get "Invalid for_each argument". Use static keys (names) instead.

Enlarge diagramCOUNT: REMOVE "BALA"FOR_EACH: REMOVE "BALA"dev[0] ashano changedev[1] bala → chitrarenamed!dev[2] chitradestroyeddev["asha"]no changedev["bala"]destroyed (intended)dev["chitra"]no change
Figure 1. Why for_each is safer: remove one item from a count list and every later index shifts, so Terraform rewrites resources that did not change.
Swipe sideways to see all of it, or tap to open it full screen.

lifecycle

resource "aws_db_instance" "orders" {
  # ...
  deletion_protection = true                     # AWS-side protection too
  lifecycle {
    prevent_destroy       = true                 # plan errors if anything would destroy it
    create_before_destroy = false
    ignore_changes        = [password]           # managed elsewhere / rotated
  }
}

resource "aws_launch_template" "nodes" {
  lifecycle { create_before_destroy = true }     # new one exists before old is removed
}

resource "aws_autoscaling_group" "web" {
  lifecycle {
    ignore_changes       = [desired_capacity]    # the autoscaler changes this
    replace_triggered_by = [aws_launch_template.nodes.latest_version]
  }
}
  • prevent_destroy — guardrail for critical resources (databases, state buckets, hosted zones). Combine with provider-side deletion protection.
  • create_before_destroy — for replacements that must not cause downtime (certificates, launch templates, security groups attached elsewhere). Names must be unique-able (use name_prefix).
  • ignore_changes — attributes another system manages (tags from a tagging tool, ASG desired capacity, Kubernetes-managed annotations).
  • replace_triggered_by — force replacement when something else changes.
  • precondition / postcondition blocks validate assumptions with custom error messages.

Forcing a single replacement:

terraform apply -replace="aws_instance.bastion"     # modern replacement for `terraform taint`
terraform plan -target=module.vpc                   # limit scope — emergencies only; leaves config and state out of sync
Enlarge diagramdestroy oldgap: nothing existscreate newcreate newboth exist brieflydestroy olddefaultcreate_before_destroy
Figure 2. create_before_destroy changes the order of a replacement so the new resource exists before the old one goes.
Swipe sideways to see all of it, or tap to open it full screen.

depends_on

Use only for dependencies Terraform can't see through references:

resource "aws_lambda_function" "processor" {
  # ...
  depends_on = [aws_iam_role_policy_attachment.processor_logs]
}