Permalink
257 lines (186 sloc) 11.5 KB

This document will soon be retired, and you probably shouldn't be reading it. It may mislead you.

It documents an approach we experiemented with early, before realising it was probably wrong, but we haven't yet transfered the knowledge it contains to some other, less central place. So here is still sits.

The RIGHT WAY in think can be found in FAQs like:

  1. https://github.com/Day8/re-frame/blob/master/docs/FAQs/LoadOnMount.md
  2. https://github.com/Day8/re-frame/blob/master/docs/FAQs/PollADatabaseEvery60.md
  3. https://github.com/Day8/re-frame-http-fx

Subscribing to External Data

In Talking To Servers we learned how to communicate with servers using both pure and effectful handlers. This is great, but what if you want to query external data using subscriptions the same way you query data stored in app-db? This tutorial will show you how.

There Can Be Only One!!

re-frame apps have a single source of data called app-db.

The re-frame README asks you to imagine app-db as something of an in-memory database. You query it (via subscriptions) and transactionally update it (via event handlers).

Components Don't Know, Don't Care

Components never know the structure of your app-db, much less its existence.

Instead, they subscribe, declaratively, to data, like this (subscribe [:something "blah"]), and that allows Components to obtain a stream of updates to "something", while knowing nothing about the source of the data.

A 2nd Source

All good but ... SPAs are seldom completely self contained data-wise.

There's a continuum between apps which are 100% standalone data-wise, and those where remote data is utterly central to the app's function. In this page, we're exploring the remote-data-centric end of this continuum.

And just to be clear, when I'm talking about remote data, I'm thinking of data luxuriating in remote databases like firebase, rethinkdb, PostgreSQL, Datomic, etc

  • data sources that an app must query and mutate.

So, the question is: how would we integrate this kind of remote data into an app when re-frame seems to have only one source of data: app-db?
How do we introduce a second or even third source of data? How should we subscribe to this remote data, and how would we update it?

By way of explanation, let's make the question specific: how could we wire up a Component which displays a collection of items, when those items come from a remote database?

In your mind's eye, imagine this kind of query against that remote database: select id, price, description from items where type="see through".

Via A Subscription

In re-frame, Components always obtain data via a subscription. Always.

So, our Component which shows items is going to

(let [items (re-frame/subscribe [:items "see through"]) ...

and the subscription handler will deliver them.

Which, in turn, means our code must have a subscription handler defined:

(re-frame/reg-sub
  :items
  (fn [db [_ item-type]
    ...))

Which is fine ... except we haven't really solved this problem yet, have we?
We've just transferred the problem away from the Component and into the subscription handler?

Well, yes, we have, and isn't that a fine thing!! That's precisely what we want from our subscription handlers ... to manage how the data is sourced ... to hide that from the Component.

The Subscription Handler's Job

Right, so let's write the subscription handler.

There'll be code in a minute but, first, let's describe how the subscription handler will work:

  1. Upon being required to provide items, it has to issue a query to the remote database. Perhaps this will be done via a RESTful GET. Or via a firebase connection. Or by pushing a JSON representation of the query down a websocket. Something. And it is the subscription handler's job to know how it is done.

  2. This query be async - with the results arriving sometime "later". And when they eventually arrive, the handler must organise for the query results to be placed into app-db, at some known, particular path. In the meantime, the handler might want to ensure that the absence of results is also communicated to the Component, allowing it to display "Loading ...". The Nine States of Design has some useful information on designing your application for different states that your data might be in.

  3. The subscription handler must return something to the Component. It should give back a reaction to that known, particular path within app-db, so that when the query results eventually arrive, they will flow through into the Component for display.

  4. The subscription handler will detect when the Component is destroyed and no longer requires the subscription. It will then clean up, getting rid of those now-unneeded items, and sorting out any stateful database connection issues.

Notice what's happening here. In many respects, app-db is still acting as the single source of data. The subscription handler is organising for the right remote data to "flow" into app-db at a known, particular path, when it is needed by a Component. And, equally, for this data to be cleaned up when it is no longer required.

Some Code

Enough fluffing about with words, here's a code sketch for our subscription handler:

(re-frame/reg-sub-raw
  :items
  (fn [app-db [_ type]]
       (let  [query-token (issue-items-query!
                            type
                            :on-success #(re-frame/dispatch [:write-to  [:some :path]]))]
         (reagent.ratom/make-reaction
           (fn [] (get-in @app-db [:some :path]))
           :on-dispose #(do (terminate-items-query! query-token)
                            (re-frame/dispatch [:cleanup [:some :path]]))))))

A few things to notice:

  1. We are using the low level reg-sub-raw to register our handler (and not the more normal reg-sub) so we can get an :on-dispose callback when the subscription is no longer needed. See the reg-sub-raw docs at the end of this tutorial

  2. You have to write issue-items-query!. Are you making a Restful GET? Are you writing JSON packets down a websocket? The query has to be made.

  3. We do not issue the query via a dispatch because, to me, it isn't an event. But we most certainly do handle the arrival of query results via a dispatch and associated event handler. That to me is an external event happening to the system. The event handler can curate the arriving data in whatever way makes sense. Maybe it does nothing more than to assoc into an app-db path, or maybe this is a rethinkdb changefeed subscription and your event handler will have to collate the newly arriving data with what has previously been returned. Do what needs to be done in that event handler, so that the right data will be put into the right path.

  4. We use Reagent's make-reaction function to create a reaction which will return that known, particular path within app-db where the query results are to be placed.

  5. We use the on-dispose callback on this reaction to do any cleanup work when the subscription is no longer needed. Clean up app-db? Clean up the database connection?

Any Good?

It turns out that this is a surprisingly flexible and clean approach. And pretty damn obvious once someone points it out to you (which is a good sign). There's a lot to like about it.

For example, if you are using rethinkdb, which supports queries which yield "change feeds" over time, rather than a one-off query result, you have to actively close such queries when they are no longer needed. That's easy to do in our cleanup code.

We can source some data from both PostgreSQL and firebase in the one app, using the same pattern. All remote data access is done in the same way.

Because query results are dispatched to an event handler, you have a lot of flexibility about how you process them.

The whole set of pieces can be arranged and tweaked in many ways. For example, with a bit of work, we could keep a register of all currently used queries. And then, if ever we noticed that the app had gone offline, and then back online, we could organise to reissue all the queries again (with results flowing back into the same known paths), avoiding stale results.

Also, notice that putting ALL interesting data into app-db has nice flow on effects. In particular, it means it is available to event handlers, should they need it when servicing events (event handlers get db as a parameter, right?).
If this item data was held in a separate place, other than app-db, it wouldn't be available in this useful way.

Warning: Undo/Redo

This technique caches remote data in app-db. Be sure to exclude this cache area from any undo/redo operations using the available configuration options

Query De-duplication

In v0.8.0 of re-frame onwards, subscriptions are automatically de-duplicated.

In prior versions, in cases where the same query is simultaneously issued from multiple places, you'd want to de-duplicate the queries. One possibility is to do this duplication in issue-items-query! itself. You can count the duplicate queries and only clear the data when that count goes to 0.

Thanks To

@nidu for his valuable review comments and insights

The Alternative Approach

Event handlers do most of the heavy lifting within re-frame apps.

When buttons get clicked, or items get dragged 'n dropped, or tabs get chosen, they know how to transition the app from one state to the next. That's their job. And, when they make such a transition, it is quite reasonable to expect them to ALSO source the data needed in the new state.

So there's definitely a case for NOT using the approach outlined above and, instead, making event handlers source data and plonk it into a certain part of app-db for use by subscriptions.

In effect, there's definitely an argument that subscriptions should only ever source from app-db BUT that it is event handlers which start and stop the sourcing of data from remote places.

Sorry, but you'll have to work out which of these two variations works best for you.

Within this document the first alternative has been given more word count only because there's a few more tricks to make it work, not because it is necessarily preferred.

Absolutely Never Do This

Sometimes, because of their background with other JS frameworks, new re-framians feel like the Components themselves (the views) should have the responsibility of sourcing the data they need.

They then use React lifecycle methods like :component-did-mount to load remote data.

I believe this is absolutely the wrong way to do it.

In re-frame, we want views to be as simple and dumb as possible. They turn data into HTML and nothing more. they absolutely do not do imperative stuff.

Use one of the two alternatives described above.


Previous: Talking to Servers       Up: Index      

Table Of Contents