This guide will teach you:
- How to compile your own copy of Metabase
- How to set up a development environment
- How to run the Metabase Server
- How to contribute back to the Metabase project
In general, we like to have an open issue for every pull request as a place to discuss the nature of any bug or proposed improvement. Each pull request should address a single issue, and contain both the fix as well as a description of how the pull request and tests that validate that the PR fixes the issue in question.
For significant feature additions, it is expected that discussion will have taken place in the attached issue. Any feature that requires a major decision to be reached will need to have an explicit design document written. The goals of this document are to make explicit the assumptions, constraints and tradeoffs any given feature implementation will contain. The point is not to generate documentation but to allow discussion to reference a specific proposed design and to allow others to consider the implications of a given design.
We don't like getting sued, so before merging any pull request, we'll need each person contributing code to sign a Contributor License Agreement here
The development scripts are designed for Linux/Mac environment, so we recommend using the latest Windows 10 version with WSL (Windows Subsystem for Linux) and Ubuntu on Windows. The Ubuntu Bash shell works well for both backend and frontend development.
If you have problems with your development environment, make sure that you are not using any development commands outside the Bash shell. As an example, Node dependencies installed in normal Windows environment will not work inside Ubuntu Bash environment.
These are the set of tools which are required in order to complete any build of the Metabase code. Follow the links to download and install them on your own before continuing.
- Oracle JDK 8 (http://www.oracle.com/technetwork/java/javase/downloads/index.html)
- Node.js (http://nodejs.org/)
- Yarn package manager for Node.js
- Leiningen (http://leiningen.org/)
If you are developing on Windows, make sure to use Ubuntu on Windows and follow instructions for Ubuntu/Linux instead of installing ordinary Windows versions.
The entire Metabase application is compiled and assembled into a single .jar file which can run on any modern JVM. There is a script which will execute all steps in the process and output the final artifact for you.
./bin/build
After running the build script simply look in target/uberjar for the output .jar file and you are ready to go.
See this guide.
If you plan to work on the Metabase code and make changes then you'll need to understand a few more things.
The Metabase application has two basic compnents:
- a backend written in Clojure which contains a REST API as well as all the relevant code for talking to databases and processing queries.
- a frontend written as a Javascript single-page application which provides the web UI.
Both components are built and assembled together into a single jar file which runs the entire application.
Metabase depends on lots of other 3rd party libraries to run, so as you are developing you'll need to keep those up to date. Leiningen will automatically fetch Clojure dependencies when needed, but for JavaScript dependencies you'll need to kick off the installation process manually when needed.
# javascript dependencies
$ yarnRun your backend development server with
lein ring server
Start the frontend build process with
yarn run build-hot
We use these technologies for our FE build process to allow us to use modules, es6 syntax, and css variables.
- webpack
- babel
- cssnext
Frontend tasks are executed using yarn run. All available tasks can be found in package.json under scripts.
To build the frontend client without watching for changes, you can use:
$ yarn run buildIf you're working on the frontend directly, you'll most likely want to reload changes on save, and in the case of React components, do so while maintaining state. To start a build with hot reloading, use:
$ yarn run build-hotNote that at this time if you change CSS variables, those changes will only be picked up when a build is restarted.
There is also an option to reload changes on save without hot reloading if you prefer that.
$ yarn run build-watchAll frontend tests are located in frontend/test directory. Run all frontend tests with
./bin/build version uberjar && yarn run test
which will first build the backend JAR and then run integration, unit and Karma browser tests in sequence.
Integration tests simulate realistic sequences of user interactions. They render a complete DOM tree using Enzyme and use temporary backend instances for executing API calls.
Integration tests use an enforced file naming convention <test-suite-name>.integ.js to separate them from unit tests.
Useful commands:
./bin/build version uberjar # Builds the JAR without frontend assets; run this every time you need to update the backend
yarn run test-integrated-watch # Watches for file changes and runs the tests that have changed
yarn run test-integrated-watch -- TestFileName # Watches the files in paths that match the given (regex) stringThe way integration tests are written is a little unconventional so here is an example that hopefully helps in getting up to speed:
import {
useSharedAdminLogin,
createTestStore,
} from "__support__/integrated_tests";
import {
click
} from "__support__/enzyme_utils"
import { mount } from "enzyme"
import { FETCH_DATABASES } from "metabase/redux/metadata";
import { INITIALIZE_QB } from "metabase/query_builder/actions";
import RunButton from "metabase/query_builder/components/RunButton";
describe("Query builder", () => {
beforeAll(async () => {
// Usually you want to test stuff where user is already logged in
// so it is convenient to login before any test case.
useSharedAdminLogin()
})
it("should let you run a new query", async () => {
// Create a superpowered Redux store.
// Remember `await` here!
const store = await createTestStore()
// Go to a desired path in the app. This is safest to do before mounting the app.
store.pushPath('/question')
// Get React container for the whole app and mount it using Enzyme
const app = mount(store.getAppContainer())
// Usually you want to wait until the page has completely loaded, and our way to do that is to
// wait until the completion of specified Redux actions. `waitForActions` is also useful for verifying that
// specific operations are properly executed after user interactions.
// Remember `await` here!
await store.waitForActions([FETCH_DATABASES, INITIALIZE_QB])
// You can use `enzymeWrapper.debug()` to see what is the state of DOM tree at the moment
console.log(app.debug())
// You can use `testStore.debug()` method to see which Redux actions have been dispatched so far.
// Note that as opposed to Enzyme's debugging method, you don't need to wrap the call to `console.log()`.
store.debug();
// For simulating user interactions like clicks and input events you should use methods defined
// in `enzyme_utils.js` as they abstract away some React/Redux complexities.
click(app.find(RunButton))
// Note: In pretty rare cases where rendering the whole app is problematic or slow, you can just render a single
// React container instead with `testStore.connectContainer(container)`. In that case you are not able
// to click links that lead to other router paths.
});
})
You can also skim through __support__/integrated_tests.js and __support__/enzyme_utils.js to see all available methods.
Unit tests are focused around isolated parts of business logic.
Unit tests use an enforced file naming convention <test-suite-name>.unit.js to separate them from integration tests.
yarn run test-unit # Run all tests at once
yarn run test-unit-watch # Watch for file changes
If you need to test code which uses browser APIs that are only available in real browsers, you can add a Karma test to frontend/test/legacy-karma directory.
yarn run test-karma # Run all tests once
yarn run test-karma-watch # Watch for file changes
Leiningen and your REPL are the main development tools for the backend. There are some directions below on how to setup your REPL for easier development.
And of course your Jetty development server is available via
lein ring server
Run unit tests with
lein test
or a specific test with
lein test metabase.api.session-test
By default, the tests only run against the h2 driver. You can specify which drivers to run tests against with the env var ENGINES:
ENGINES=h2,postgres,mysql,mongo lein test
At the time of this writing, the valid engines are h2, postgres, mysql, mongo, sqlserver, sqlite, druid, bigquery, oracle, vertica, and redshift. Some of these engines require additional parameters
when testing since they are impossible to run locally (such as Redshift and Bigquery). The tests will fail on launch and let you know what parameters to supply if needed.
Due to some issues with the way we've structured our test setup code, you currently always need to include h2 in the ENGINES list. Thus to test something like bigquery you should specify ENGINES=h2,bigquery. Fortunately the H2 tests are fast so this should not make a noticeable difference.
lein eastwood && lein bikeshed && lein docstring-checker && ./bin/reflection-linter
.dir-locals.el contains some Emacs Lisp that tells clojure-mode how to indent Metabase macros and which arguments are docstrings. Whenever this file is updated,
Emacs will ask you if the code is safe to load. You can answer ! to save it as safe.
By default, Emacs will insert this code as a customization at the bottom of your init.el.
You'll probably want to tell Emacs to store customizations in a different file. Add the following to your init.el:
(setq custom-file (concat user-emacs-directory ".custom.el")) ; tell Customize to save customizations to ~/.emacs.d/.custom.el
(ignore-errors ; load customizations from ~/.emacs.d/.custom.el
(load-file custom-file))Start up an instant cheatsheet for the project + dependencies by running
lein instant-cheatsheet
Copyright © 2017 Metabase, Inc
Distributed under the terms of the GNU Affero General Public License (AGPL) except as otherwise noted. See individual files for details.