Data flows

Orbit enables the coordination of many independent sources of data, each of which may internally handle queries and updates differently. It's easy to imagine tying oneself in knots with such potential complexity. However, by following Orbit's conventions and a few guidelines, you can compose data to flow predictably and reliably through your application.

"Request up, sync down"

Orbit divides the movement of data into two different "flows":

• Request flow - Requests to query or update data originate from an application and flow upstream to a source that can fulfill the request.

• Sync flow - Mutations then flow back downstream and are synchronized with all the sources that are concerned.

Every source interface has events and methods that correspond with one of these flows:

• The Updatable and Queryable interfaces all participate in the request flow.

• The Syncable interface participates in the sync flow.

"Request up, sync down" is a variant of the "data down, actions up" mnemonic popularized in the Ember.js community. In fact, most frontend frameworks adopt a similar pattern for handling actions and returning data. Orbit fits well with these patterns: an action triggered by a user can spawn an Orbit "request", which can lead to responses that "sync" data back down, typically ending in an update to a view. In this way, "request up, sync down" can be seen as a continuation of the "data down, actions up" pattern.

Coordinating sources

Request and sync flows can be coordinated across sources by configuring an event listener for one source that triggers actions on another.

Let's take a look at what events can trigger other actions:

• Update events (beforeUpdate, update) can trigger update.

• Query events (beforeQuery, query, beforePull, pull) can trigger pull.

• Change events (transform, beforeSync, sync) can trigger sync.

Blocking vs. non-blocking

We can coordinate sources through simple event listeners, such as:

memory.on('beforeUpdate', transform => {
  remote.update(transform);
});

The above listener is "non-blocking" because it doesn't return anything to the emitter. The call to remote.update() is async and may take a while to complete, so it will proceed in parallel with the memory source being updated.

As an alternative, we can use a "blocking" strategy in our event listener by simply returning a promise:

memory.on('beforeUpdate', transform => remote.update(transform));

This will prevent the memory source from updating before the transform has been pushed up to the remote source. An error in remote.update will cause memory.update to error as well.

Coordination guidelines

Here are some guidelines for working with data flows:

• Consider the full arc of each request—how it will flow up to be fulfilled, and how results and/or errors will be synchronized on the way back down.

• Avoid spawning requests from the synchronization flow. A change event (beforeSync, sync, or transform) should only ever trigger a sync action.

• For pessimistic requests in which you must guarantee success before proceeding, use blocking connections for all the request and sync flows that may be involved.

Last but not least, it's recommended that you use a Coordinator instead of manually configuring event listeners. Read on to understand why ...