Readme: origami-build-service

See the production service for API information.

Table Of Contents

• Requirements
• Running Locally
• Configuration
• Testing
• Deployment
• Monitoring
• Trouble-Shooting
• Project Structure
  • ES6/Promises patterns
• License

Requirements

Running Origami Build Service requires Node.js 8.x and npm.

Running Locally

Before we can run the application, we'll need to install dependencies:

npm install

Run the application in development mode with

make run-dev

Now you can access the app over HTTP on port 8080: http://localhost:8080/

Configuration

We configure Origami Build Service using environment variables. In development, configurations are set in a .env file. In production, these are set through Heroku config. Further documentation on the available options can be found in the [Origami Service documentation][service-options].

Required everywhere

• NODE_ENV: The environment to run the application in. One of production, development (default), or test (for use in automated tests).
• PORT: The port to run the application on.

Required in Heroku

• GITHUB_PASSWORD: The GitHub password corresponding to GITHUB_USERNAME.
• GITHUB_USERNAME: A GitHub username with permission to view required private repositories.
• GRAPHITE_API_KEY: The FT's internal Graphite API key
• GRAPHITE_HOST: The hostname of a Graphite server to gather metrics with.
• PREFERRED_HOSTNAME: The hostname to use in documentation and as a base URL in bundle requests. This defaults to www.ft.com/__origami/service/build.
• REGION: The region the application is running in. One of QA, EU, or US
• RELEASE_LOG_API_KEY: The change request API key to use when creating and closing release logs
• RELEASE_LOG_ENVIRONMENT: The Salesforce environment to include in release logs. One of Test or Production
• SENTRY_DSN: The Sentry URL to send error information to

Headers

The service can also be configured by sending HTTP headers, these would normally be set in your CDN config:

• FT-Origami-Service-Base-Path: The base path for the service, this gets prepended to all paths in the HTML and ensures that redirects work when the CDN rewrites URLs.

Testing

The tests are split into unit tests, integration tests, and an old suite of tests that we're migrating. To run tests on your machine you'll need to install Node.js and run make install. Then you can run the following commands:

make test              # run all the tests
make test-unit         # run the unit tests
make test-integration  # run the integration tests
make test-old          # run the old suite of tests

You can run the unit tests with coverage reporting, which expects 90% coverage or more:

make test-unit-coverage verify-coverage

The code will also need to pass linting on CI, you can run the linter locally with:

make verify

We run the tests and linter on CI, you can view results on CircleCI. make test and make verify must pass before we merge a pull request.

Deployment

The production (EU/US) and QA applications run on Heroku. We deploy continuously to QA via CircleCI, you should never need to deploy to QA manually. We use a Heroku pipeline to promote QA deployments to production.

You can promote either through the Heroku interface, or by running the following command locally:

make promote

Monitoring

• Grafana dashboard: graph memory, load, and number of requests
• Pingdom check (Production EU): checks that the EU production app is responding
• Sentry dashboard (Production): records application errors in the production app
• Sentry dashboard (QA): records application errors in the QA app
• Splunk (Production): query application logs (see below)

Logging

We use Splunk to store and query our application and CDN log files. Using Splunk we can answer many questions, such as: which product is using our services the most; which components are not being requested (good candidates to deprecate).

Here is an example query which was used to find out if our o-big-number component was being requested.

Here is an example query which shows the last hour of logs from our CDN.

Trouble-Shooting

We've outlined some common issues that can occur when running the Build Service:

What do I do if memory usage is high?

For now, restart the Heroku dynos:

heroku restart --app origami-build-service-eu

If this doesn't help, then a temporary measure could be to add more dynos to the production applications, or switch the existing ones to higher performance dynos.

What if I need to deploy manually?

If you really need to deploy manually, you should only do so to QA. Production deploys should always be a promotion from QA.

You'll need to provide an API key for change request logging. You can get this from the Origami LastPass folder in the note named Change Request API Keys. Now deploy to QA using the following:

CR_API_KEY=<API-KEY> make deploy

What do I do if my updated component is not appearing in bundles?

This is most likely due to the heavy caching we use.

First, change the hostname in your request to origami-build-service-eu.herokuapp.com. If your update appears now, then the CDN had cached the bundle. You'll need to wait for a while, or clear the CDN cache for your URL.

If your component still doesn't appear, then we've cached an older version on the file system. You can clear this by restarting the Heroku dynos:

heroku restart --app origami-build-service-eu

Project Structure

ES6/Promises patterns

You'll need to understand promises (Q implementation) and how they use the yield operator.

• Asynchronous programming changes use(func(arg)) into func(arg, use).
• Promises change func(arg, use) into func(arg).then(use).
• ES6 enables turning func(arg).then(use) into use(yield func(arg)).

The code uses promises and promise-wrapped versions of node functions. Q.denodeify(func) converts callback-based function into promise-returning function.

If you do doPromise = Q.denodeify(doAsync) then this code:

doAsync1(arg1, function(err, result1){
    if (err) throw err;
    doAsync2(arg2, function(err, result2) {
        if (err) throw err;
    });
});

is roughly equal to:

doPromise1(arg1)
.then(function(result1){
    return doPromise2(arg2);
})
.then(function(result2){
})
.done();

Instead of nesting callbacks, you chain .then() calls. Callbacks in then may return (a promise of) the result for the next then callback.

You can simplify further with Q.async(). This enables waiting for promises with the yield operator:

Q.async(function*(){
    var result1 = yield doPromise1(arg1);
    var result2 = yield doPromise2(arg2);
});

Q.async returns a promise as well, so we can chain it with other promises.

License

The Financial Times has published this software under the MIT license.