Terraform at team scale without DynamoDB state locks.

DynamoDB-backed Terraform state locks are the default for a reason. They work. But at a certain team size, they stop being free. This is what broke for us at ~40 engineers, and what we replaced them with.

What broke

• CI runs queueing on lock contention during peak deploy windows
• Stale locks from cancelled CI jobs requiring manual unlock
• Cross-team blast radius when a single workspace held a lock too long
• No visibility into who held a lock without grepping CloudTrail

What we replaced it with

A queue, not a lock. Our CI runner now serializes plans-and-applies per workspace with explicit ownership signals: owner, expected duration, escalation channel, surfaced in the PR. The state lock still exists at the Terraform level, but it's almost never the bottleneck anymore.

terraform {
  backend "s3" {
    bucket = "company-tfstate"
    key    = "platform/${terraform.workspace}/terraform.tfstate"
    region = "us-east-1"
    # No DynamoDB lock table, serialization handled by CI queue
  }
}