Skip to content

Latest commit

 

History

History
84 lines (58 loc) · 7.63 KB

Architecture.md

File metadata and controls

84 lines (58 loc) · 7.63 KB

Architecture

✨ Today ✨

Fetching data from the server and rendering the UI to display that data is a classic pattern. The approach we try to follow (see The Need for Speed) is to "render as soon as possible" and "save optimistically."

In short, say a component <Items /> needs to display a data object passed through the props by the parent, we will also give the component a fetchingData prop, so it can render accordingly. There are 4 possible situations (the component may choose to render more than one situation in the same way):

  • data is falsy and fetchingData is truthy: first data load, or reset, we can render for example an empty "skeleton" while we wait for data
  • data and fetchingData are both falsy: data load returned an empty set, we can display a "no data available" message for example
  • data is truthy and fetchingData is falsy: display the data "normally"
  • data and fetchingData are both truthy: a data refresh, either don't do anything and wait for data to come back, or display some kind of loading indicator

For forms, we try as much as possible to "save optimistically," meaning when the user "saves" the form, we immediately update the app state (and thus the UI), and then send the new data to the server to be saved. If the server returns an error, we should be able to rollback the app state and display some kind of error message.

Details on blip's data fetching strategy

As of our refactoring in 2015–2016 to use React Router and Redux in blip, we have established a single pattern for data fetching for all page-level "smart" components (found in app/pages/) that are both defined as the component for a route (see app/routes.js) and connected to the redux store.

Each of these pages defines a mapDispatchToProps function for use as the second argument to react-redux's connect utility. The mapDispatchToProps function uses the bindActionCreators utility from redux to wrap every action creator needed on the page in a redux dispatch call so that it can be used directly in the page. However, with our setup in blip, this dispatch-wrapped action creator is still not quite ready for direct use, so each page also defines a getFetchers function which also binds our api singleton:

let getFetchers = (dispatchProps, ownProps, api) => {
  return [
    dispatchProps.fetchPendingReceivedInvites.bind(null, api),
    dispatchProps.fetchAssociatedAccounts.bind(null, api)
  ];
};

and sometimes other function arguments:

let getFetchers = (dispatchProps, ownProps, api, options) => {
  return [
    dispatchProps.fetchPatient.bind(null, api, ownProps.match.params.id),
    dispatchProps.fetchPatientData.bind(null, api, options, ownProps.match.params.id)
  ];
};

The getFetchers function is called from the mergeProps function provided as the third argument to the react-redux connect utility; getFetchers always return an array of functions and is always assigned to a prop in the inner page component (as opposed to the outer component wrapped with connect) called fetchers. Each inner page component defines an instance method called doFetching which simply loops through the fetchers included (if any) in the props provided as doFetching's sole function argument and calls each function in turn:

  doFetching: function(props) {
    if (!props.fetchers) {
      return
    }

    props.fetchers.forEach(fetcher => {
      fetcher();
    });
  },

Every page also integrates the fetching into the React component lifecycle in a consistent manner by calling the doFetching instance method from the componentDidMount lifecycle method:

  componentDidMount: function() {
    this.doFetching(this.props);
  },

🚀 The Future

Aside from continuing to improve on the blip app experience by iterating on current features and adding new features, we at Tidepool have planned changes or are aware that scaling or opening the platform to third-party development may push us to make changes in a few areas that have implications for blip's architecture. Currently (as of mid-November, 2016) we have no concrete plans or solutions for the issues discussed below; they are all open questions.

Data caching

The data fetching pattern outlined above is not the most efficient pattern, since data that is already available in state may be re-fetched on mount of a different component (that shares some of the same data needs as the previously mounted component). On the one hand, the current strategy ensures that data never gets stale in the app (as long as the user is still doing things). On the other hand, some kind of caching strategy, which could be client- or server-side, could reduce load on Tidepool's servers. Whether or how we implement caching to improve data fetching efficiency in blip is as of yet an open question.

Switch to OAuth for authentication

Tidepool plans to use OAuth 2 as our eventual authentication standard both internally and to support third-party clients. This will likely have an impact on blip's architecture as the platform account-related features in blip (e.g., sign-up, account settings, etc.) will need to be provided via a traditional server-rendered web application to ensure the application's security.

Combining functionality with the uploader in an Electron app

Google has announced the end of Chrome apps, and so we are already planning to replace our Chrome uploader application with an Electron application. This provides an opportunity to simplify the Tidepool user experience (especially for PwD and caregiver home users, as opposed to clinical users) by combining the uploading and data visualization functionalities into a single Electron application, or, in other words, combining blip's functionality with the uploader's.

Although blip and the uploader share some architectural features in common such as very similar redux implementations for state management, there are also a number of areas where combining the two applications may pose an architectural challenge, such as:

  • client-side routing, since the uploader does not (currently) use a router
  • styles, since the uploader uses CSS modules and blip's uses class-prefixing to define local styles
  • different approaches to wrapping platform-client in an api.js utility