Monorepo concepts, tips and tricks oriented around NextJs
Howtos for monorepo. New to monorepos ? check this FAQ. This example is managed by turborepo and yarn 4 with a / typescript path aliases approach. Not the only way to do.
Useful to
- Establish a structure and present a lifecycle perspective (dx, ci/cd, deployments...)
- How to create and consume shared packages, locales, assets, api types...
- Integrate tools & configs (eslint, jest, playwright, storybook, changelogs, versioning, codecov, codeclimate...).
- Clarify some advantages of monorepos (team cohesion, consistency, duplication, refactorings, atomic commits...).
- Create nextjs/vercel/prisma... bug reports with reproducible examples (initial goal of this repo).
If you are enjoying some of my OSS work in your company, I'd really appreciate a sponsorship, a coffee or a dropped star. That gives me some more time to improve it to the next level.
|
|
|
JetBrains | Embie.be | Vercel |
corepack enable
yarn install
.
├── apps
│ ├── nextjs-app (i18n, ssr, api, vitest)
│ └── vite-app
└── packages
├── common-i18n (locales...)
├── core-lib
├── db-main-prisma
├── eslint-config-bases (to shared eslint configs)
└── ui-lib (emotion, storybook)
- apps/nextjs-app: SSR, i18n, tailwind v3, emotion, graphQL, rest... README | DEMO/Vercel | CHANGELOG
- apps/vite-app: Basic vite-app. README | DEMO/Vercel | CHANGELOG
Apps should not depend on apps, they can depend on packages
- packages/core-lib: publishable. README | CHANGELOG
- packages/db-main-prisma: used by web-app. README | CHANGELOG
- packages/eslint-config-bases: README | CHANGELOG
- packages/ui-lib: publishable. README | CHANGELOG
- packages/common-i18n: README | CHANGELOG
Apps can depend on packages, packages can depend on each others...
Detailed folder structure
.
├── apps
│ ├── vite-app (Vite app as an example)
│ │ ├── src/
│ │ ├── package.json (define package workspace:package deps)
│ │ └── tsconfig.json (define path to packages)
│ │
│ └── nextjs-app (NextJS app with api-routes)
│ ├── e2e/ (E2E tests with playwright)
│ ├── public/
│ ├── src/
│ │ └── pages/api (api routes)
│ ├── CHANGELOG.md
│ ├── next.config.mjs
│ ├── package.json (define package workspace:package deps)
│ ├── tsconfig.json (define path to packages)
│ └── vitest.config.ts
│
├── packages
│ ├── core-lib (basic ts libs)
│ │ ├── src/
│ │ ├── CHANGELOG.md
│ │ ├── package.json
│ │ └── tsconfig.json
│ │
│ ├── db-main-prisma (basic db layer with prisma)
│ │ ├── e2e/ (E2E tests)
│ │ ├── prisma/
│ │ ├── src/
│ │ ├── CHANGELOG.md
│ │ ├── package.json
│ │ └── tsconfig.json
│ │
│ ├── eslint-config-bases
│ │ ├── src/
│ │ ├── CHANGELOG.md
│ │ ├── package.json
│ │ └── tsconfig.json
│ │
│ └── ui-lib (basic design-system in react)
│ ├── src/
│ ├── CHANGELOG.md
│ ├── package.json
│ └── tsconfig.json
│
├── static (no code: images, json, locales,...)
│ ├── assets
│ └── locales
├── docker (docker...)
│ ├── .dockerignore
│ ├── docker-compose.yml (compose specific for nextjs-app)
│ ├── docker-compose.db.yml (general services like postgresql...)
│ └── Dockerfile (multistage build for nextjs-app)
├── .yarnrc.yml
├── package.json (the workspace config)
└── tsconfig.base.json (base typescript config)
Root package.json with workspace directories
{
"name": "nextjs-monorepo-example",
// Set the directories where your apps, packages will be placed
"workspaces": ["apps/*", "packages/*"],
//...
}
The package manager will scan those directories and look for children package.json
. Their
content is used to define the workspace topology (apps, libs, dependencies...).
Create a folder in ./packages/ directory with the name of your package.
Create the package folder
mkdir packages/magnificent-poney
mkdir packages/magnificent-poney/src
cd packages/magnificent-poney
Initialize a package.json with the name of your package.
Rather than typing
yarn init
, prefer to take the ./packages/ui-lib/package.json as a working example and edit its values.
Example of package.json
{
"name": "@your-org/magnificent-poney",
"version": "0.0.0",
"private": true,
"scripts": {
"clean": "rimraf ./tsconfig.tsbuildinfo",
"lint": "eslint . --ext .ts,.tsx,.js,.jsx",
"typecheck": "tsc --project ./tsconfig.json --noEmit",
"test": "run-s 'test:*'",
"test:unit": "echo \"No tests yet\"",
"fix:staged-files": "lint-staged --allow-empty",
"fix:all-files": "eslint . --ext .ts,.tsx,.js,.jsx --fix",
},
"devDependencies": {
"@your-org/eslint-config-bases": "workspace:^",
},
}
First add the package to the app package.json. The recommended way is to use the workspace protocol supported by yarn and pnpm.
cd apps/my-app
yarn add @your-org/magnificent-poney@'workspace:^'
Inspiration can be found in apps/nextjs-app/package.json.
package.json
{
"name": "my-app",
"dependencies": {
"@your-org/magnificient-poney": "workspace:^",
},
}
Then add a typescript path alias in the app tsconfig.json. This will allow you to import it directly (no build needed)
Inspiration can be found in apps/nextjs-app/tsconfig.json.
Example of tsonfig.json
{
"compilerOptions": {
"baseUrl": "./src",
"paths": {
// regular app aliases
"@/components/*": ["./components/*"],
// packages aliases, relative to app_directory/baseUrl
"@your-org/magnificent-poney/*": [
"../../../packages/magnificent-poney/src/*",
],
"@your-org/magnificent-poney": [
"../../../packages/magnificent-poney/src/index",
],
},
},
}
PS:
- Don't try to set aliases in the global tsonfig.base.json to keep strict with graph dependencies.
- The star in
@your-org/magnificent-poney/*
allows you to import subfolders. If you use a barrel file (index.ts), the alias with star can be removed.
Edit your next.config.mjs
and enable the experimental.externalDir option.
Feedbacks here.
const nextConfig = {
experimental: {
externalDir: true,
},
};
Using a NextJs version prior to 10.2.0 ?
If you're using an older NextJs version and don't have the experimental flag, you can simply override your webpack config.
const nextConfig = {
webpack: (config, { defaultLoaders }) => {
// Will allow transpilation of shared packages through tsonfig paths
// @link https://github.com/vercel/next.js/pull/13542
const resolvedBaseUrl = path.resolve(config.context, "../../");
config.module.rules = [
...config.module.rules,
{
test: /\.(tsx|ts|js|jsx|json)$/,
include: [resolvedBaseUrl],
use: defaultLoaders.babel,
exclude: (excludePath) => {
return /node_modules/.test(excludePath);
},
},
];
return config;
},
};
PS: If your shared package make use of scss bundler... A custom webpack configuration will be necessary or use next-transpile-modules, see FAQ below.
The packages are now linked to your app, just import them like regular packages: import { poney } from '@your-org/magnificent-poney'
.
Optional
If you need to share some packages outside of the monorepo, you can publish them to npm or private repositories. An example based on microbundle is present in each package. Versioning and publishing can be done with atlassian/changeset, and it's simple as typing:
$ yarn g:changeset
Follow the instructions... and commit the changeset file. A "Version Packages" P/R will appear after CI checks. When merging it, a github action will publish the packages with resulting semver version and generate CHANGELOGS for you.
PS:
- Even if you don't need to publish, changeset can maintain an automated changelog for your apps. Nice !
- To disable automatic publishing of some packages, just set
"private": "true"
in their package.json.- Want to tune the behaviour, see .changeset/config.json.
Some convenience scripts can be run in any folder of this repo and will call their counterparts defined in packages and apps.
Name | Description |
---|---|
yarn g:changeset |
Add a changeset to declare a new version |
yarn g:codegen |
Run codegen in all workspaces |
yarn g:typecheck |
Run typechecks in all workspaces |
yarn g:lint |
Display linter issues in all workspaces |
yarn g:lint --fix |
Attempt to run linter auto-fix in all workspaces |
yarn g:lint-styles |
Display css stylelint issues in all workspaces |
yarn g:lint-styles --fix |
Attempt to run stylelint auto-fix issues in all workspaces |
yarn g:test |
Run unit and e2e tests in all workspaces |
yarn g:test-unit |
Run unit tests in all workspaces |
yarn g:test-e2e |
Run e2e tests in all workspaces |
yarn g:build |
Run build in all workspaces |
yarn g:clean |
Clean builds in all workspaces |
yarn g:check-dist |
Ensure build dist files passes es2017 (run g:build first). |
yarn g:check-size |
Ensure browser dist files are within size limit (run g:build first). |
yarn clean:global-cache |
Clean tooling caches (eslint, jest...) |
yarn deps:check --dep dev |
Will print what packages can be upgraded globally (see also .ncurc.yml) |
yarn deps:update --dep dev |
Apply possible updates (run yarn install && yarn dedupe after) |
yarn install:playwright |
Install playwright for e2e |
yarn dedupe |
Built-in yarn deduplication of the lock file |
Why using
:
to prefix scripts names ? It's convenient in yarn 3+, we can call those scripts from any folder in the monorepo.g:
is a shortcut forglobal:
. See the complete list in root package.json.
The global commands yarn deps:check
and yarn deps:update
will help to maintain the same versions across the entire monorepo.
They are based on the excellent npm-check-updates
(see options, i.e: yarn check:deps -t minor
).
After running
yarn deps:update
, ayarn install
is required. To prevent having duplicates in the yarn.lock, you can runyarn dedupe --check
andyarn dedupe
to apply deduplication. The duplicate check is enforced in the example github actions.
See an example in ./apps/nextjs-app/.eslintrc.js and our eslint-config-bases.
Check the .husky folder content to see what hooks are enabled. Lint-staged is used to guarantee that lint and prettier are applied automatically on commit and/or pushes.
Tests relies on ts-jest or vitest depending on the app. All setups supports typescript path aliases. React-testing-library is enabled whenever react is involved.
Configuration lives in the root folder of each apps/packages. As an example see
You'll find some example workflows for github action in .github/workflows. By default, they will ensure that
- You don't have package duplicates.
- You don't have typecheck errors.
- You don't have linter / code-style errors.
- Your test suite is successful.
- Your apps (nextjs) or packages can be successfully built.
- Basic check-size example in nextjs-app.
Each of those steps can be opted-out.
To ensure decent performance, those features are present in the example actions:
-
Caching of packages (node_modules...) - install around 25s
-
Caching of nextjs previous build - built around 20s
-
Triggered when changed using actions paths, ie:
paths: - "apps/nextjs-app/**" - "packages/**" - "package.json" - "tsconfig.base.json" - "yarn.lock" - ".yarnrc.yml" - ".github/workflows/**" - ".eslintrc.base.json" - ".eslintignore"
The ESLint plugin requires that the eslint.workingDirectories
setting is set:
"eslint.workingDirectories": [
{
"pattern": "./apps/*/"
},
{
"pattern": "./packages/*/"
}
],
More info here
Vercel support natively monorepos, see the vercel-monorepo-deploy document.
There's a basic example for building a docker image, read the docker doc.
Netlify, aws-amplify, k8s-docker, serverless-nextjs recipes might be added in the future. PR's welcome too.
- Ease of code reuse. You can easily extract shared libraries (like api, shared ui, locales, images...) and use them across apps without the need of handling them in separate git repos (removing the need to publish, version, test separately...). This limit the tendency to create code duplication amongst developers when time is short.
- Atomic commits. When projects that work together are contained in separate repositories, releases need to sync which versions of one project work with the other. In monorepo CI, sandboxes and releases are much easier to reason about (ie: dependency hell...). A pull-request contains all changes at once, no need to coordinate multiple packages versions to test it integrally (multiple published canary versions...).
- Code refactoring. Changes made on a library will immediately propagate to all consuming apps / packages. Typescript / typechecks, tests, ci, sandboxes... will improve the confidence to make a change (or the right one thanks to improved discoverability of possible side effects). It also limits the tendency to create tech debt as it invites the dev to refactor all the code that depends on a change.
- Collaboration across teams. Consistency, linters, discoverability, duplication... helps to maintain cohesion and collaboration across teams.
- Increased build time. Generally a concern but not relevant in this context thanks to the combination of nextjs/webpack5, typescript path aliases and yarn. Deps does not need to be build... modified files are included as needed and properly cached (nextjs webpack5, ci, deploy, docker/buildkit...).
- Versioning and publishing. Sometimes a concern when you want to use the shared libraries outside of the monorepo. See the notes about atlassian changeset. Not relevant here.
- Git repo size. All packages and apps and history will fit in the same git repository increasing its size and checkout time. Generally when you reach size problems, check for assets like images first and extract packages that don't churn anymore.
- Multi-languages. Setting up a monorepo containing code in multiple languages (php, ruby, java, node) is extremely difficult to handle due to nonexistence of mature tooling (bazel...).The general idea is to create a monorepo with the same stack (node, typescript...) and managed by the same package manager (yarn, pnpm,...)
Apps dependencies and devDependencies are pinned to exact versions. Packages deps will use semver compatible ones. For more info about this change see reasoning here and our renovabot.json5 configuration file.
To help keeping deps up-to-date, see the yarn deps:check && yarn deps:update
scripts and / or use the renovatebot.
When adding a dep through yarn cli (i.e.: yarn add something), it's possible to set the save-exact behaviour automatically by setting
defaultSemverRangePrefix: ""
in yarnrc.yml. But this would make the default for packages/* as well. Better to handleyarn add something --exact
on per-case basis.