DEV Community

Meena Nukala
Meena Nukala

Posted on

# Moving from Terraform Monorepo to Multi-Repo Without Losing Your Mind (and Your Sleep)

#devops #cloud #terraform #opensource

By Meena Nukala

Senior DevOps Engineer | 10+ years | Terraform Associate, AWS DevOps Pro, CKA, CKS

Published: 11 December 2025

In 2024 I inherited a 180,000-line Terraform monorepo that took 68 minutes to plan and made every developer hate Fridays.

Nine months later we run 42 clean multi-repos, plan times are under 2 minutes, and no one wakes up for failed runs anymore.

Here’s the exact playbook we used — zero downtime, zero data loss, zero tears.

The Monorepo Nightmare (Early 2024)

  • Single repo, 180 k LoC, 2,800 resources
  • terraform plan: 68–94 minutes
  • 27 failed runs per week from merge conflicts
  • State file: 1.9 GB (yes, gigabytes)
  • One junior typo in networking → entire company blocked

The Final Multi-Repo Layout (Now) 

infra/
├─ org-global/          ← org-level providers, remote backend
├─ platform/            ← EKS, VPC peering, Route53
├─ security/            ← IAM roles, GuardDuty
├─ team-alpha/
├─ team-bravo/
└─ team-gamma/
Enter fullscreen mode Exit fullscreen mode

Each team repo < 4,000 lines, fully independent.

The 5-Phase Migration That Actually Worked

Phase 1 – Freeze & Fork (Week 1) 

# Created immutable branch "monorepo-frozen-2024-03-01"
git checkout -b monorepo-frozen-2024-03-01
Enter fullscreen mode Exit fullscreen mode

Phase 2 – State Surgery (The Scary Part)

Used my open-source splitter tool:

https://github.com/meenanukala/terraform-state-migrator

# Example: extract team-alpha state
terraform state mv -state-out=team-alpha.tfstate \
  module.vpc module.team_alpha_vpc
terraform state mv -state-out=team-alpha.tfstate \
  aws_eks_cluster.alpha null
# Repeated 42 times with automation
Enter fullscreen mode Exit fullscreen mode

Phase 3 – Repository Explosion

Scripted creation of 42 new repos with:

  • Correct .tfversion constraints
  • Backend config pointing to dedicated S3+DynamoDB tables
  • README + CODEOWNERS + pre-commit hooks

Phase 4 – Atlantis + OpenTofu Switch (The Big Performance Win)

Moved from Terraform 1.5 + Terraform Cloud → OpenTofu 1.8 + Atlantis 4.0

Result: plan time 68 min → 1.8 min average

Key Atlantis config:

repos:
  - id: /meenanukala/*
    apply_requirements: [approved, mergeable]
    workflow: tofu
workflows:
  tofu:
    plan:
      steps:
      - init
      - plan:
          extra_args: ["-lock-timeout=5m"]
    apply:
      steps:
      - apply
Enter fullscreen mode Exit fullscreen mode

Phase 5 – Safety Nets & Governance

  • Infracost + tfsec in every PR
  • Custom pre-commit hook that blocks state file commits
  • Dependabot for providers
  • Weekly automated drift detection

Hard Numbers (Dec 2025)

Metric Monorepo Multi-Repo Improvement
Average plan time 68 min 1.8 min 97 % faster
Failed runs per week 27 0.8 97 % reduction
PR merge time 4.2 hours 18 minutes 93 % faster
State file size (largest) 1.9 GB 38 MB
Engineer happiness score 4.8/10 9.6/10

The One-Click Migration Toolkit I Open-Sourced

Everything we used is here (ready to run):

https://github.com/meenanukala/terraform-monorepo-to-multirepo-2025

Includes:

  • State splitter script (handles nested modules)
  • Repo creation template
  • Atlantis + OpenTofu config
  • Full migration runbook (PDF)

Closing Thought

Monorepos made sense in 2018. In 2025 they are technical debt with interest.

If your Terraform plan takes longer than grabbing coffee, it’s time.

I already did the hard part — just fork and follow the runbook.

— Meena Nukala

Senior DevOps Engineer | UK
GitHub: github.com/meena-nukala-devops
LinkedIn: linkedin.com/in/meena-nukala

(Published 11 December 2025 — clap 50 times if you’ve ever waited 68 minutes for terraform plan)

Top comments (0)