Cocoon is a Dart App Engine custom runtime (backend) with a frontend of Flutter apps (build and repository dashboard) and Angular 2 Dart (performance dashboard) apps. Cocoon coordinates and aggregates the results of flutter/flutter builds. It is not designed to help developers build Flutter apps. Cocoon is not a Google product.

Using Cocoon

Agents

The dashboard has a page showing the status of all our current agents.

Creating an agent

To create an agent in the dashboard, it needs an agentId and a list of capabilities (comma delimited). Clicking the "Create Agent" button will show a dialog for creating an agent.

An example of a valid agent would be agentId=bot-with-devices and capabilities=has-android-phone,has-iphone.

The dialog returns an authentication token, and prints it to the console. This token is not stored, so copy it immediately and add it to the agent's configuration file. If the token is lost or compromised, authorize the agent to generate a new token.

Authorizing an agent

Click on the dropdown for the agent, and click authorize agent. This will print a new generated token to the console.

This command invalidates any previously issued authentication tokens for the given agent. Only one authentication token is valid at any given moment in time. Therefore, if the agent is currently using a previously issued token its API requests will be rejected until it switches to using the newly created token.

Forcing a refresh from GitHub

The server is driven by commits made to https://github.com/flutter/flutter repo. It periodically syncs new commits. If you need to manually force a refresh, query https://flutter-dashboard.appspot.com/api/refresh-github-commits. You will need to be authenticated with Cocoon to do this.

Developing Cocoon

Cocoon has several components:

A server, which coordinates everything. This is a Dart App Engine application. If you have never used that before, you may want to peruse the samples for Dart App Engine. The server is found in app_dart.
An agent, a Dart program that runs on test hosts, each of which have a test device. Our "devicelab" consists of computers with devices that are running agents and talking to the server to get tasks to run (e.g. a benchmark). The agent is found in agent.
A Flutter app (generally used as a Web app) for the build and agent dashboards. The dashboard is found in app_flutter.
An Angular 2 for Dart Web app for the performance dashboard. We intend to reimplement this in the Flutter app eventually. The performance dashboard is found in app.

Cocoon creates a checklist for each Flutter commit. A checklist is made of multiple tasks. Tasks are performed by agents. An agent is a computer capable of running a subset of tasks in the checklist. To perform a task an agent reserves it in Cocoon. Cocoon issues tasks according to agents' capabilities. Each task has a list of required capabilities. For example, a task might require that a physical Android device is attached to an agent. It then lists "has-physical-android-phone" capability as required. Multiple agents may share the same capability. Cocoon will distribute tasks amongst agents. That's how Cocoon scales.

Getting started

First, set up a Flutter development environment. This will, as a side-effect, provide you with a Dart SDK. Your life will be easier if you add that (.../flutter/bin/cache/dart-sdk/bin/) to your path.

To update the production server, you will need the Google Cloud SDK. Since there is no Dart SDK, we just use the command line tools.

Developing the server

All the commands in this section assume that you are in the app_dart/ directory.

Running a local dev server

This is useful for developing backend functionality locally. This local dev server can be connected to the frontend applications by running dart dev/deploy.dart --project test --version test, and answering N when asked about deploying to App Engine. This will build the frontend files and copy them to the directory from which the server will serve them.

Set the environment variables GCLOUD_PROJECT and GCLOUD_KEY. Running dart bin/server.dart will give more explanation on what these values should be. You should also set COCOON_USE_IN_MEMORY_CACHE to true as you typically don't have access to the remote redis instance during local development.

If you see Serving requests at 0.0.0.0:8080 the dev server is working.

To develop and test some features, you need to have a local service account(key.json) with access to the project you will be connecting to.

If you work for Google you can use the key with flutter-dashboard project via internal doc.

Building the server for deployment

To run tests, build the app, and provide instructions for deploying to Google App Engine, run this command:

dart dev/deploy.dart --project {PROJECT} --version {VERSION}

You can test the new version by accessing {VERSION}-dot-flutter-dashboard.appspot.com in your browser. If the result is satisfactory, the new version can be activated by using the Cloud Console UI: https://pantheon.corp.google.com/appengine/versions?project=flutter-dashboard&serviceId=default

Optional flags

--profile: Deploy a profile mode of app_flutter application for debugging purposes.

Developing the dashboard

The dashboard application will use dummy data when it is not connected to the server, so it can be developed locally without a dev server.

To run the dashboard locally, go into the app_flutter directory and run flutter run -d web. The dashboard will be served from localhost (the exact address will be given on the console); copy the URL into your browser to view the application. (The dashboard should also be able to run on non-Web platforms, but since the Web is our main target that is the one that should generally be used for development.)

You may need to run flutter config --enable-web to enable Web support if you haven't done so in the past.

You can run flutter packages upgrade to update the dependencies. This may be necessary if you see a failure in the dependencies.

stuartmorgan/cocoon