acedergren/turborepo

Turborepo - Monorepo Architecture Expert

Assumption: You know turbo run build. This covers architectural decisions.

Arguments

• $ARGUMENTS: Monorepo decision, package boundary, or cache issue to analyze
  • Example: /turborepo why is turbo cache missing in CI
  • Example: /turborepo should packages/ui be split from packages/web-core
  • If empty: ask which Turborepo architecture problem is in scope

---

Before Adopting Turborepo: Strategic Assessment

| Signal | Recommendation | | ------ | -------------- | | 1-3 engineers | Polyrepo — monorepo overhead not worth it | | <20% shared code | Polyrepo | | >50% shared code + frequent coordination | Monorepo compelling | | Mixed languages (Go/Python/JS) | Nx or polyrepo — Turborepo is JS/TS focused | | All builds <5min total | Overhead not justified yet | | Breaking changes require 3+ repos | Monorepo wins | | Services deploy independently | Polyrepo |

Break-even: Monorepo worth it when 3+ apps share 30%+ code AND frequent coordination is required.

---

Critical Rule: Package Tasks, Not Root Tasks

The #1 Turborepo mistake: Putting task logic in root package.json.

// WRONG - defeats parallelization
// Root package.json
{ "scripts": { "build": "cd apps/web && next build && cd ../api && tsc" } }

// CORRECT - each package owns its task
// apps/web/package.json
{ "scripts": { "build": "next build" } }

// Root package.json - ONLY delegates
{ "scripts": { "build": "turbo run build" } }

Why: Turborepo can't parallelize sequential shell commands. Package tasks enable task graph parallelization.

---

Decision: When to Split a Package

Considering splitting code into a package?
│
├─ Used by 1 app only → DON'T split yet
│   └─ Keep in app; wait for second consumer
│      WHY: Premature abstraction, overhead > benefit
│
├─ Used by 2+ apps → MAYBE split
│   ├─ Stable API (rarely changes) → Split
│   ├─ Unstable (changes every sprint) → DON'T split yet
│   └─ Mixed team ownership → DON'T split (use import path)
│      WHY: Shared packages need stable APIs + clear owners
│
├─ Publishing to npm → MUST split
│
└─ CI builds > 10min → Split by stability, not domain
    └─ Stable packages cache; unstable packages always rebuild

Anti-pattern: Creating packages for "clean architecture" with no consumers. Every package adds build, test, and version overhead.

---

Anti-Patterns

❌ #1: Circular Dependencies

Symptom: turbo run build fails with "Could not resolve dependency graph"

packages/ui → packages/utils
packages/utils → packages/ui  // circular

Fix: Extract shared code to a third package (packages/shared).

For indirect cycles (A → B → C → A), use: npx madge --circular --extensions ts,tsx packages/

❌ #2: Overly Granular Packages

Symptom: Every feature touches 5+ packages; 10+ version bumps per sprint; pnpm workspace:* version hell.

Fix: Group by change frequency, not by domain:

packages/ui/            # All components (changes often)
packages/ui-primitives/ # Headless components (stable)
packages/icons/         # Generated SVGs (rarely changes)

Rule: Package boundary = different change frequency. Packages that always change together should be one package.

❌ #3: Missing Task Dependencies

Symptom: Tests pass locally, fail in CI with "Cannot find module './dist/index.js'"

Cause: Tests run before build completes — race condition.

// WRONG - no dependsOn for test
{ "tasks": { "build": { "outputs": ["dist/**"] }, "test": {} } }

// CORRECT
{
  "tasks": {
    "build": { "dependsOn": ["^build"], "outputs": ["dist/**"] },
    "test": { "dependsOn": ["build"] }
  }
}

^build = build this package's dependencies first. build = build this package first.

❌ #4: Cache Miss Hell

Symptom: Cache never hits; every run rebuilds everything.

Cause: inputs glob too broad — comment changes trigger rebuild.

// WRONG
{ "build": { "inputs": ["src/**"] } }

// CORRECT
{ "build": { "inputs": ["src/**/*.{ts,tsx}", "!src/**/*.test.ts"] } }

Debug:

turbo run build --dry --graph          # Visualize task graph
turbo run build --dry=json | jq '.tasks[] | select(.cache.status == "MISS")'

---

Decision: Monorepo vs Polyrepo

Starting new project?
│
├─ Single team, single product → Polyrepo (simpler)
│
├─ Shared UI library → Monorepo
│   └─ Develop library + test in consumers simultaneously
│
├─ Microservices in different languages → Polyrepo
│   └─ Turborepo is JS/TS focused
│
└─ Multiple teams, shared code, atomic changes needed → Monorepo

Practical advice: Start polyrepo, migrate to monorepo when the cross-repo coordination pain exceeds the tooling cost.

---

Package Boundary Patterns

By stability (recommended):

packages/core/      # Changes quarterly (semantic versioning)
packages/features/  # Changes weekly (workspace protocol)
packages/utils/     # Changes monthly

By consumer:

packages/public-api/  # External consumers — strict versioning
packages/internal/    # Internal apps — workspace protocol OK

By team: Only works if teams rarely share code. Otherwise creates silos.

---

Turborepo vs Alternatives

| Prefer Turborepo | Prefer Nx | Prefer Rush | | ---------------- | --------- | ----------- | | JS/TS monorepo | Project graph visualization needed | 100+ packages | | Vercel remote caching | Polyglot (JS + Python + Go) | Publishing to npm is primary goal | | pnpm/npm workspaces | Want opinionated project structure | Phantom dependency detection needed |

---

Error Recovery

Cache never hits

1. turbo run build --dry=json | jq '.tasks[0].hash' — see current hash
2. Narrow inputs glob to exclude non-code files
3. Fallback: "cache": false in turbo.json temporarily to debug without cache pressure

Circular dependency error

1. turbo run build --dry --graph=graph.html — visualize in browser
2. npx madge --circular --extensions ts,tsx packages/ — for indirect cycles
3. Extract common code to packages/shared

Tests fail in CI but pass locally

1. turbo run test --dry --graph — verify build runs before test
2. Add "dependsOn": ["build"] to test task
3. turbo run test --force — bypass cache to confirm ordering

Overly granular packages causing version hell

1. git log --oneline --since="1 month ago" -- packages/ — count version bumps per package
2. Packages that change together 5+ times → merge them
3. Fallback: use workspace:* to auto-link versions while planning merge

---

When to Load Full Reference

READ references/cli-options.md when: encountering 3+ unknown CLI flags, need advanced --filter patterns across 10+ packages, or setting up complex pipeline options.

READ references/remote-cache-setup.md when: setting up remote cache for teams, debugging cache auth errors, or configuring self-hosted cache with custom storage.

Do NOT load references for: basic architecture decisions, single cache miss debugging, or monorepo adoption decisions — all covered above.

---

Resources

• Official Docs: https://turbo.build/repo/docs