Example Angular monorepo using Bazel
This is experimental, as part of Angular Labs! There may be breaking changes.
This is part of the ABC project. The overall goal is to make it possible to develop Angular applications the same way we do at Google.
Learn more about Bazel and Angular at https://bazel.angular.io
This example is deployed at https://bazel.angular.io/example
Guide to the example
This example is a monorepo, meant to show many different features and integrations that we expect are generally useful for enterprise use cases.
- Angular CLI: you can use the
ngcommand to run build, serve, test, and e2e
- Angular Libraries: to maximize build incrementality, each Angular module is compiled as a separate step. This lets us re-use Angular libraries without having to publish them as npm packages. See
src/todosfor a typical
NgModulecompiled as a library for use in the application, using the
ng_modulerule in the
- TypeScript Libraries: see
src/libfor a trivial example of a pure-TS library that's consumed in the application, using the
ts_libraryrule in the
- Sass: we use Sass for all styling. Angular components import Sass files, and these are built by Bazel as independent processes calling the modern Sass compiler (written in Dart).
- Material design: see
src/materialwhere we collect the material modules we use.
- Redux-style state management: see
src/reducerswhere we use the NgRx Store.
- Lazy loading: in production mode, the application is served in chunks. Run
ng serve --prod
- Differential loading: in production mode, we load a pair of
<script>tags. Modern browsers will load code in the ES2015 syntax, which is smaller and requires fewer polyfills. Older browsers will load ES5 syntax.
- Docker: see below where we package up the production app for deployment on Kubernetes.
You only need to install one build tool, and which one you choose typically depends on what kind of development you do most often.
If you're a frontend developer, you should install NodeJS and yarn.
package.json file has an
engines section which indicates the range of NodeJS and yarn versions that you could use.
You simply run
yarn commands shown below, and don't need to install Bazel or any other dependencies.
If you're a full-stack developer, you might be using Bazel for your backend already.
In this case, you should install Bazel following instructions at http://bazel.build.
ibazel, which is a watch mode for Bazel not included in the standard distribution. See https://github.com/bazelbuild/bazel-watcher#installation.
WORKSPACE file has a
check_bazel_version call which will print an error if your Bazel version is not in the supported range.
You simply run
bazel commands shown below, and don't need to install NodeJS, yarn, or any other dependencies.
First we'll run the development server:
$ ng serve # or $ ibazel run //src:devserver
This runs in "watch mode", which means it will watch any files that are inputs to the devserver, and when they change it will ask Bazel to re-build them. When the re-build is finished, it will trigger a LiveReload in the browser.
This command prints a URL on the terminal. Open that page to see the demo app
running. Now you can edit one of the source files (
src/lib/file.ts is an easy
one to understand and see the effect). As soon as you save a change, the app
should refresh in the browser with the new content. Our intent is that this time
is less than two seconds, even for a large application.
Control-C twice to kill the devserver.
We can also run all the unit tests:
$ ng test # or $ bazel test //src/...
Or run the end-to-end tests:
$ ng e2e # or $ bazel test //e2e/...
In this example, there is a unit test for the
hello-world component which uses
ts_web_test_suite rule. There are also protractor e2e tests for both the
devserver which use the
Note that Bazel will only re-run the tests whose inputs changed since the last run.
We can run the application in production mode, where the code has been bundled and optimized. This can be slower than the development mode, because any change requires re-optimizing the app. This example uses Rollup and Uglify, but other bundlers can be integrated with Bazel.
$ ng serve --prod # or $ bazel run //src:prodserver
The production bundle is code split and the
are lazy loaded. Code splitting is handled by the rollup_bundle rule
which now supports the new code splitting feature in rollup.
Note: code splitting is not supported in development mode yet so the
//src:devserver target does not serve a code split bundle. For this
reason, development and production use different main entry points
main.ts) and different root modules
app.module.ts). The difference in
the entry points and modules is how routes are loaded, with production
lazy loading routes and development using a custom
loader to disable lazy loading.
also called in the production entry point.
Having a local
node_modules folder setup by
npm is not
necessary when building this example with Bazel. This example makes use
of Bazel managed npm dependencies (https://github.com/bazelbuild/rules_nodejs#using-bazel-managed-dependencies)
which means Bazel will setup the npm dependencies in your
package.json for you
outside of your local workspace for use in the build.
However, you may still want to run
npm to manually
setup a local
node_modules folder for editor and tooling support.
We use the standard firebase deploy command.
yarn deploy to release changes to bazel.angular.io.
We use Bazel's docker support to package up our production server for deployment. Each time the app changes, we'll get a slim new docker layer with just the modified files, keeping the round-trip for deployment incremental and fast. This example is configured to run on Google Kubernetes Engine, so we can have an elastic pool of backend machines behind a load balancer. This setup is more expensive to operate than something like Firebase Functions where the backend code is spun up on-demand, but is also more adaptable to scenarios like backend servers that need to run other binaries on the machine.
The application is currently live at http://18.104.22.168/
To run it under docker:
$ bazel run src:nodejs_image -- --norun $ docker run --rm -p 8080:8080 bazel/src:nodejs_image
Deploy to production:
- Install gcloud and kubectl
- Authenticate to the Google Container Registry
gcloud auth configure-docker
- Authenticate to Kubernetes Engine
gcloud container clusters get-credentials angular-bazel-example --zone=us-west1-a
- For the first deployment:
bazel run :deploy.create
- To update:
bazel run :deploy.replace
# Run the binary without docker $ bazel run src:nodejs_image.binary # What's in the image? $ bazel build src:nodejs_image && file-roller dist/bin/src/nodejs_image-layer.tar # Tear down all running docker containers $ docker rm -f $(docker ps -aq) # Hop into the running image on kubernetes $ kubectl exec angular-bazel-example-prod-3285254973-ncv3g -it -- /bin/bash