We Keep Coding

React/Flux and xhr/routing/caching

This is more of a "whats your opinion/Am I correct in thinking this?" question.
Trying to be as strict as possible while understanding Flux, I was trying to figure out where XHR calls are made, websockets/external stimuli handled, routing takes places, etc.
From what I read across articles, interviews and looking through facebook examples there are a few ways of handling these things. Following flux strictly, Action creators are the ones that do all the XHR calls with the possibility of a PENDING/SUCCESS/FAILURE Actions being fired before and after the request completes.
Another was, coming from facebook's Ian Obermiller, all the READ(GETs) requests are handled directly by the Stores(without involvement of an Action creator/dispatcher) and WRITE(POSTs) requests are handled by the Action Creators going through the entire action>dispatcher>store flow.
Some understandings/conclusions we drew/would like to stick to:
Ideally, anything going in/out of the system happens only through Actions.
Async calls leaving/entering the system will have PENDING/PROGRESS(think file uploads)/SUCCESS/FAILURE Actions.
Single dispatcher across the entire App.
Action>Dispatcher>Store calls are strictly synchronous to stick to the dispatches not being able to start another dispatch internally to avoid chaining events/actions.
Stores are persisted across Views(considering its a single page app, you want to be able to reuse data)
A few questions that we came to some conclusion with, but I'm not entirely satisfied with:
If you take the approach where Stores do Reads, and Actions to Writes, how do you handle situations where multiple Stores might be able to use data from a single XHR call?
Example: API calls issued by TeamStore to /api/teams/{id} which returns something like:
{
entities: {
teams: [{
name: ...,
description: ...,
members: [1, 2, 4],
version: ...
}],
users: [{
id: 1
name: ...,
role: ...,
version: ...
},
{
id: 2
name: ...,
role: ...,
version: ...
},
{
id: 3
name: ...,
role: ...,
version: ...
}]
}
}
Ideally, I'd also like to update the MemberStore with the information returned in this API. We maintain a version number for every entity which is updated on updates to the record, which is what we use internally do reject calls on stale data, etc. Using this, I could have an internal logic, where if I as a side effect of some other API call, I know my data is stale, I trigger a refresh on that record.
The solution, it would seem, is that you'd need the store to trigger an action(which would effectively update the other dependent stores). This short circuits the Store>View>Action to Store>Action and I'm not sure if its a good idea. We already have one thing out of sync with Stores doing their own XHR calls. Concessions like these would start creeping into the entire system eventually.
Or Stores that are aware of other stores and be able to communicate with them. But this breaks the Stores have no Setters rule.
A simple solution to the above problem would be that you stick to Actions being the ONLY place external incoming/outgoing stimulus happens. This simplifies the logic of multiple Stores getting updated.
But now, where and how do you handle caching? We came to the conclusion that the caching would happen at the API Utils/DAO level. (if you look at the flux diagram).
But this introduces other problems. To better understand/explain what I mean by example:
/api/teams returns a list of all the teams with which I display a list of all the teams.
On clicking on a team's link, I go its details view which requires data from /api/teams/{id} if it isn't already present in the Store.
If Actions handle all the XHRs, the View would do something like TeamActions.get([id]) which does TeamDAO.get([id]). To be able to return this call immediately(since we have it cached) the DAO would have to do caching but also maintain the relation between collections/items. This logic, by design, is already present in Stores.
Here come the questions:
Do you duplicate this logic in DAOs and Stores?
Do you make DAO's aware of Stores and they can ask the Store if they already have some data and just return a 302 saying, you're good you have the latest data.
How do you handle validation that involves XHR APIs? Something simple like duplicate Team names.
Views directly hit DAOs and do something like TeamDAO.validateName([name]) which returns a promise or do you do you create an Action? If you create an Action through which Store does Valid/Invalid flow back to the View considering its mostly transient data?
How do you handle Routing? I looked through react-router and I'm not sure I like it. I don't necessarily think forcing a react-ish JSX way of providing route mappings/configs are needed at all. Also, apparently, it employs a RouteDispatcher of its own, which ondoes the single dispatcher rule.
The solution I prefer came from some blog posts/SO answers where you have a the route mappings are stored in the RouteStore.
RouteStore also maintains CURRENT_VIEW. The react AppContainer component is registered with RouteStore and replaces its child views with the CURRENT_VIEW on change. Current Views inform the AppContainer when they're fully loaded and AppContainer fires RouteActions.pending/success/failure, possibly with some context, to inform other components of reaching a stable state, show/hide busy/loading indications.
Something that I have not been able to design cleanly was if you were to design routing similar to Gmail, how would you do it? Some observations of Gmail that I'm a big fan of:
URLs don't change until the page is ready to load. It stays on the current URL while its 'Loading' and moves to the new one once the loading has finished. This makes it so that...
On failure, you don't lose you current page at all. So if you're on compose, and the 'Send' fails, you don't lose your mail (i.e. you don't lose your current stable view/state). (they don't do this because auto saving is le pwn, but you get the idea) You have the option of copy/pasting the mail somewhere for safe keeping till you can send again.
Some references:
https://github.com/gaearon/flux-react-router-example
http://ianobermiller.com/blog/2014/09/15/react-and-flux-interview/
https://github.com/facebook/flux

It's my implementation using facebook Flux and Immutable.js that I think responds to many of your concerns, based on few rules of thumb :
STORES
Stores are responsible for maintaining data state through Immutable.Record and maintaining cache through a global Immutable.OrderedMap referencing Record instance via ids.
Stores directly call WebAPIUtils for read operations and trigger actions for write operations.
Relationship between RecordA and FooRecordB are resolved from a RecordA instance through a foo_id params and retrieved via a call such as FooStore.get(this.foo_id)
Stores only expose getters methods such as get(id), getAll(), etc.
APIUTILS
I use SuperAgent for ajax calls. Each request is wrapped in Promise
I use a map of read request Promise indexed by the hash of url + params
I trigger action through ActionCreators such as fooReceived or fooError when Promise is resolved or rejected.
fooError action should certainly contains payloads with validation errors returned by the server.
COMPONENTS
The controller-view component listen for changes in store(s).
All my components, other than controller-view component, are 'pure', so I use ImmutableRenderMixin to only re-render what it's really needed (meaning that if you print Perf.printWasted time, it should be very low, few ms.
Since Relay and GraphQL are not yet open sourced, I enforce to keep my component props as explicit as possible via propsType.
Parent component should only passes down the necessary props. If my parent component holds an object such as var fooRecord = { foo:1, bar: 2, baz: 3}; (I'm not using Immutable.Record here for the sake of simplicity of this example) and my child component need to display fooRecord.foo and fooRecord.bar, I do not pass the entire foo object but only fooRecordFoo and fooRecordBar as props to my child component because an other component could edit the foo.baz value, making the child component re-render while this component doesn't need at all this value !
ROUTING
- I simply use ReactRouter
IMPLEMENTATION
Here is a basic example :
api
apiUtils/Request.js
var request = require('superagent');
//based on http://stackoverflow.com/a/7616484/1836434
var hashUrl = function(url, params) {
var string = url + JSON.stringify(params);
var hash = 0, i, chr, len;
if (string.length == 0) return hash;
for (i = 0, len = string.length; i < len; i++) {
chr = string.charCodeAt(i);
hash = ((hash << 5) - hash) + chr;
hash |= 0; // Convert to 32bit integer
}
return hash;
}
var _promises = {};
module.exports = {
get: function(url, params) {
var params = params || {};
var hash = hashUrl(url, params);
var promise = _promises[hash];
if (promise == undefined) {
promise = new Promise(function(resolve, reject) {
request.get(url).query(params).end( function(err, res) {
if (err) {
reject(err);
} else {
resolve(res);
}
});
});
_promises[hash] = promise;
}
return promise;
},
post: function(url, data) {
return new Promise(function(resolve, reject) {
var req = request
.post(url)
.send(data)
.end( function(err, res) {
if (err) {
reject(err);
} else {
resolve(res);
}
});
});
}
};
apiUtils/FooAPI.js
var Request = require('./Request');
var FooActionCreators = require('../actions/FooActionCreators');
var _endpoint = 'http://localhost:8888/api/foos/';
module.exports = {
getAll: function() {
FooActionCreators.receiveAllPending();
Request.get(_endpoint).then( function(res) {
FooActionCreators.receiveAllSuccess(res.body);
}).catch( function(err) {
FooActionCreators.receiveAllError(err);
});
},
get: function(id) {
FooActionCreators.receivePending();
Request.get(_endpoint + id+'/').then( function(res) {
FooActionCreators.receiveSuccess(res.body);
}).catch( function(err) {
FooActionCreators.receiveError(err);
});
},
post: function(fooData) {
FooActionCreators.savePending();
Request.post(_endpoint, fooData).then (function(res) {
if (res.badRequest) { //i.e response return code 400 due to validation errors for example
FooActionCreators.saveInvalidated(res.body);
}
FooActionCreators.saved(res.body);
}).catch( function(err) { //server errors
FooActionCreators.savedError(err);
});
}
//others foos relative endpoints helper methods...
};
stores
stores/BarStore.js
var assign = require('object-assign');
var EventEmitter = require('events').EventEmitter;
var Immutable = require('immutable');
var AppDispatcher = require('../dispatcher/AppDispatcher');
var ActionTypes = require('../constants/BarConstants').ActionTypes;
var BarAPI = require('../APIUtils/BarAPI')
var CHANGE_EVENT = 'change';
var _bars = Immutable.OrderedMap();
class Bar extends Immutable.Record({
'id': undefined,
'name': undefined,
'description': undefined,
}) {
isReady() {
return this.id != undefined //usefull to know if we can display a spinner when the Bar is loading or the Bar's data if it is ready.
}
getBar() {
return BarStore.get(this.bar_id);
}
}
function _rehydrate(barId, field, value) {
//Since _bars is an Immutable, we need to return the new Immutable map. Immutable.js is smart, if we update with the save values, the same reference is returned.
_bars = _bars.updateIn([barId, field], function() {
return value;
});
}
var BarStore = assign({}, EventEmitter.prototype, {
get: function(id) {
if (!_bars.has(id)) {
BarAPI.get(id);
return new Bar(); //we return an empty Bar record for consistency
}
return _bars.get(id)
},
getAll: function() {
return _bars.toList() //we want to get rid of keys and just keep the values
},
Bar: Bar,
emitChange: function() {
this.emit(CHANGE_EVENT);
},
addChangeListener: function(callback) {
this.on(CHANGE_EVENT, callback);
},
removeChangeListener: function(callback) {
this.removeListener(CHANGE_EVENT, callback);
},
});
var _setBar = function(barData) {
_bars = _bars.set(barData.id, new Bar(barData));
};
var _setBars = function(barList) {
barList.forEach(function (barData) {
_setbar(barData);
});
};
BarStore.dispatchToken = AppDispatcher.register(function(action) {
switch (action.type)
{
case ActionTypes.BAR_LIST_RECEIVED_SUCESS:
_setBars(action.barList);
BarStore.emitChange();
break;
case ActionTypes.BAR_RECEIVED_SUCCESS:
_setBar(action.bar);
BarStore.emitChange();
break;
case ActionTypes.BAR_REHYDRATED:
_rehydrate(
action.barId,
action.field,
action.value
);
BarStore.emitChange();
break;
}
});
module.exports = BarStore;
stores/FooStore.js
var assign = require('object-assign');
var EventEmitter = require('events').EventEmitter;
var Immutable = require('immutable');
var AppDispatcher = require('../dispatcher/AppDispatcher');
var ActionTypes = require('../constants/FooConstants').ActionTypes;
var BarStore = require('./BarStore');
var FooAPI = require('../APIUtils/FooAPI')
var CHANGE_EVENT = 'change';
var _foos = Immutable.OrderedMap();
class Foo extends Immutable.Record({
'id': undefined,
'bar_id': undefined, //relation to Bar record
'baz': undefined,
}) {
isReady() {
return this.id != undefined;
}
getBar() {
// The whole point to store an id reference to Bar
// is to delegate the Bar retrieval to the BarStore,
// if the BarStore does not have this Bar object in
// its cache, the BarStore will trigger a GET request
return BarStore.get(this.bar_id);
}
}
function _rehydrate(fooId, field, value) {
_foos = _foos.updateIn([voucherId, field], function() {
return value;
});
}
var _setFoo = function(fooData) {
_foos = _foos.set(fooData.id, new Foo(fooData));
};
var _setFoos = function(fooList) {
fooList.forEach(function (foo) {
_setFoo(foo);
});
};
var FooStore = assign({}, EventEmitter.prototype, {
get: function(id) {
if (!_foos.has(id)) {
FooAPI.get(id);
return new Foo();
}
return _foos.get(id)
},
getAll: function() {
if (_foos.size == 0) {
FooAPI.getAll();
}
return _foos.toList()
},
Foo: Foo,
emitChange: function() {
this.emit(CHANGE_EVENT);
},
addChangeListener: function(callback) {
this.on(CHANGE_EVENT, callback);
},
removeChangeListener: function(callback) {
this.removeListener(CHANGE_EVENT, callback);
},
});
FooStore.dispatchToken = AppDispatcher.register(function(action) {
switch (action.type)
{
case ActionTypes.FOO_LIST_RECEIVED_SUCCESS:
_setFoos(action.fooList);
FooStore.emitChange();
break;
case ActionTypes.FOO_RECEIVED_SUCCESS:
_setFoo(action.foo);
FooStore.emitChange();
break;
case ActionTypes.FOO_REHYDRATED:
_rehydrate(
action.fooId,
action.field,
action.value
);
FooStore.emitChange();
break;
}
});
module.exports = FooStore;
components
components/BarList.react.js (controller-view component)
var React = require('react/addons');
var Immutable = require('immutable');
var BarListItem = require('./BarListItem.react');
var BarStore = require('../stores/BarStore');
function getStateFromStore() {
return {
barList: BarStore.getAll(),
};
}
module.exports = React.createClass({
getInitialState: function() {
return getStateFromStore();
},
componentDidMount: function() {
BarStore.addChangeListener(this._onChange);
},
componentWillUnmount: function() {
BarStore.removeChangeListener(this._onChange);
},
render: function() {
var barItems = this.state.barList.toJS().map(function (bar) {
// We could pass the entire Bar object here
// but I tend to keep the component not tightly coupled
// with store data, the BarItem can be seen as a standalone
// component that only need specific data
return <BarItem
key={bar.get('id')}
id={bar.get('id')}
name={bar.get('name')}
description={bar.get('description')}/>
});
if (barItems.length == 0) {
return (
<p>Loading...</p>
)
}
return (
<div>
{barItems}
</div>
)
},
_onChange: function() {
this.setState(getStateFromStore();
}
});
components/BarListItem.react.js
var React = require('react/addons');
var ImmutableRenderMixin = require('react-immutable-render-mixin')
var Immutable = require('immutable');
module.exports = React.createClass({
mixins: [ImmutableRenderMixin],
// I use propTypes to explicitly telling
// what data this component need. This
// component is a standalone component
// and we could have passed an entire
// object such as {id: ..., name, ..., description, ...}
// since we use all the datas (and when we use all the data it's
// a better approach since we don't want to write dozens of propTypes)
// but let's do that for the example's sake
propTypes: {
id: React.PropTypes.number.isRequired,
name: React.PropTypes.string.isRequired,
description: React.PropTypes.string.isRequired
}
render: function() {
return (
<li>
<p>{this.props.id}</p>
<p>{this.props.name}</p>
<p>{this.props.description}</p>
</li>
)
}
});
components/BarDetail.react.js
var React = require('react/addons');
var ImmutableRenderMixin = require('react-immutable-render-mixin')
var Immutable = require('immutable');
var BarActionCreators = require('../actions/BarActionCreators');
module.exports = React.createClass({
mixins: [ImmutableRenderMixin],
propTypes: {
id: React.PropTypes.number.isRequired,
name: React.PropTypes.string.isRequired,
description: React.PropTypes.string.isRequired
},
handleSubmit: function(event) {
//Since we keep the Bar data up to date with user input
//we can simply save the actual object in Store.
//If the user goes back without saving, we could display a
//"Warning : item not saved"
BarActionCreators.save(this.props.id);
},
handleChange: function(event) {
BarActionCreators.rehydrate(
this.props.id,
event.target.name, //the field we want to rehydrate
event.target.value //the updated value
);
},
render: function() {
return (
<form onSubmit={this.handleSumit}>
<input
type="text"
name="name"
value={this.props.name}
onChange={this.handleChange}/>
<textarea
name="description"
value={this.props.description}
onChange={this.handleChange}/>
<input
type="submit"
defaultValue="Submit"/>
</form>
)
},
});
components/FooList.react.js (controller-view component)
var React = require('react/addons');
var FooStore = require('../stores/FooStore');
var BarStore = require('../stores/BarStore');
function getStateFromStore() {
return {
fooList: FooStore.getAll(),
};
}
module.exports = React.createClass({
getInitialState: function() {
return getStateFromStore();
},
componentDidMount: function() {
FooStore.addChangeListener(this._onChange);
BarStore.addChangeListener(this._onChange);
},
componentWillUnmount: function() {
FooStore.removeChangeListener(this._onChange);
BarStore.removeChangeListener(this._onChange);
},
render: function() {
if (this.state.fooList.size == 0) {
return <p>Loading...</p>
}
return this.state.fooList.toJS().map(function (foo) {
<FooListItem
fooId={foo.get('id')}
fooBar={foo.getBar()}
fooBaz={foo.get('baz')}/>
});
},
_onChange: function() {
this.setState(getStateFromStore();
}
});
components/FooListItem.react.js
var React = require('react/addons');
var ImmutableRenderMixin = require('react-immutable-render-mixin')
var Bar = require('../stores/BarStore').Bar;
module.exports = React.createClass({
mixins: [ImmutableRenderMixin],
propTypes: {
fooId: React.PropTypes.number.isRequired,
fooBar: React.PropTypes.instanceOf(Bar).isRequired,
fooBaz: React.PropTypes.string.isRequired
}
render: function() {
//we could (should) use a component here but this answer is already too long...
var bar = <p>Loading...</p>;
if (bar.isReady()) {
bar = (
<div>
<p>{bar.get('name')}</p>
<p>{bar.get('description')}</p>
</div>
);
}
return (
<div>
<p>{this.props.fooId}</p>
<p>{this.props.fooBaz}</p>
{bar}
</div>
)
},
});
Let's go through an entire loop for FooList:
State 1:
User hits the page /foos/ listing the Foos via the FooListcontroller-view component
FooListcontroller-view component calls FooStore.getAll()
_foos map is empty in FooStore so FooStore performs a request via FooAPI.getAll()
The FooList controller-view component renders itself as loading state since its state.fooList.size == 0.
Here's the actual look of our list :
++++++++++++++++++++++++
+ +
+ "loading..." +
+ +
++++++++++++++++++++++++
FooAPI.getAll() request resolves and triggers the FooActionCreators.receiveAllSuccess action
FooStore receive this action, updates its internal state, and emits change.
State 2:
FooList controller-view component receive change event and update its state to get the list from the FooStore
this.state.fooList.size is no longer == 0 so the list can actually renders itself (note that we use toJS() to explicitly get a raw javascript object since React does not handle correctly mapping on not raw object yet).
We're passing needed props to the FooListItem component.
By calling foo.getBar() we're telling to the FooStore that we want the Bar record back.
getBar() method of Foo record retrieve the Bar record through the BarStore
BarStore does not have this Bar record in its _bars cache, so it triggers a request through BarAPI to retrieve it.
The same happens for all Foo in this.sate.fooList of FooList controller-view component
The page now looks something like this:
++++++++++++++++++++++++
+ +
+ Foo1 "name1" +
+ Foo1 "baz1" +
+ Foo1 bar: +
+ "loading..." +
+ +
+ Foo2 "name2" +
+ Foo2 "baz2" +
+ Foo2 bar: +
+ "loading..." +
+ +
+ Foo3 "name3" +
+ Foo3 "baz3" +
+ Foo3 bar: +
+ "loading..." +
+ +
++++++++++++++++++++++++
-Now let's say the BarAPI.get(2) (requested by Foo2) resolves before BarAPI.get(1) (request by Foo1). Since it's asynchronous it's totally plausible.
- The BarAPI triggers the BAR_RECEIVED_SUCCESS' action via theBarActionCreators.
- TheBarStore` responds to this action by updating its internal store and emits change. That's the now the fun part...
State 3:
The FooList controller-view component responds to the BarStore change by updating its state.
The render method is called
The foo.getBar() call now retrieve a real Bar record from BarStore. Since this Bar record has been effectively retrieved, the ImmutablePureRenderMixin will compare old props with current props and determine that the Bar objects has changed ! Bingo, we could re-render the FooListItem component (a better approach here would be to create a separate FooListBarDetail component to let only this component to re-render, here we also re-rendering the Foo's details that have not changed but for the sake of simplicity let's just do that).
The page now looks like this :
++++++++++++++++++++++++
+ +
+ Foo1 "name1" +
+ Foo1 "baz1" +
+ Foo1 bar: +
+ "loading..." +
+ +
+ Foo2 "name2" +
+ Foo2 "baz2" +
+ Foo2 bar: +
+ "bar name" +
+ "bar description" +
+ +
+ Foo3 "name3" +
+ Foo3 "baz3" +
+ Foo3 bar: +
+ "loading..." +
+ +
++++++++++++++++++++++++
If you want me to add more details from a non detailed part (such as action creators, constants, routing, etc., use of BarListDetail component with form, POST, etc.) just tell me in the comments :).

A few differences in my implementation:
I like stores employing a flyweight pattern. That is, unless forced
to, all operations are "getOrRetrieveOrCreate"
I've had to forgo promise heavy development in favor of
events/state. Async communication should still use promises, that
is, things in actions use them otherwise communication occurs using
events. If a view always renders the current state, then you need a
state like "isLoading" to render a spinner. Or you need an event to
get fired then update a state on a view. I think responding from an
action with a promise may be an anti-pattern (not entirely sure).
URL changes fire the appropriate action. GET should work and be
idempotent so a URL change should generally not result in a failure.
It may however result in a redirect. I have an "authRequired"
decorator for some actions. If you aren't authenticated then we
redirect you to the login page with the target URL listed as a
redirect path.
For validation we are thinking about starting from an action, firing a "xyzModel:willSaveData", before we start; then firing either "xyzModel:didSaveData" or "xyzModel:failedSaveData" events. The store listening to these events will indicate "saving" to the views that care. It may also indicate "hasValidationError" to views that care. If you want to dismiss an error. You can fire an action from a view that indicates that the error "wasReceived", which removes the "hasValidationError" flag or optionally could do something else like clear out all validation errors. Validations are interesting because of the different styles of validation. Ideally, you could create an app that would accept most any input due the limitations imposed by your input elements. Then again, servers may disagree with those choices :/.

Meteor - how do I make this "reactive" using Deps?

On my client side, I display a list of users and a small chart for each user's points stored in the DB (using jQuery plugin called sparklines).
Drawing the chart is done on Template.rendered method
// client/main.js
Template.listItem.rendered = function() {
var arr = this.data.userPoints // user points is an array of integers
$(this.find(".chart")).sparkline(arr);
}
Now I have a Meteor method on the server side, that is called on a regular basis to update the the user points.
Meteor.methods({
"getUserPoints" : function getUserPoints(id) {
// access some API and fetch the latest user points
}
});
Now I would like the chart to be automatically updated whenever Meteor method is called. I have a method on the template that goes and calls this Meteor method.
Template.listItem.events({
"click a.fetchData": function(e) {
e.preventDefault();
Meteor.call("getUserPoints", this._id);
}
});
How do I turn this code into a "reactive" one?

You need to use reactive data source ( Session, ReactiveVar ) together with Tracker.
Using ReactiveVar:
if (Meteor.isClient) {
Template.listItem.events({
"click a.fetchData": function(e) {
e.preventDefault();
var instance = Template.instance();
Meteor.call("getUserPoints", this._id, function(error, result) {
instance.userPoints.set(result)
});
}
});
Template.listItem.created = function() {
this.userPoints = new ReactiveVar([]);
};
Template.listItem.rendered = function() {
var self = this;
Tracker.autorun(function() {
var arr = self.userPoints.get();
$(self.find(".chart")).sparkline(arr);
})
}
}
Using Session:
if (Meteor.isClient) {
Template.listItem.events({
"click a.fetchData": function(e) {
e.preventDefault();
Meteor.call("getUserPoints", this._id, function(error, result) {
Session.set("userPoints", result);
});
}
});
Template.listItem.rendered = function() {
var self = this;
Tracker.autorun(function() {
var arr = Session.get("userPoints");
$(self.find(".chart")).sparkline(arr);
})
}
}
Difference between those implementation :
A ReactiveVar is similar to a Session variable, with a few
differences:
ReactiveVars don't have global names, like the "foo" in
Session.get("foo"). Instead, they may be created and used locally, for
example attached to a template instance, as in: this.foo.get().
ReactiveVars are not automatically migrated across hot code pushes,
whereas Session state is.
ReactiveVars can hold any value, while Session variables are limited
to JSON or EJSON.
Source
Deps is deprecated, but still can be used.

The most easily scalable solution is to store the data in a local collection - by passing a null name, the collection will be both local and sessional and so you can put what you want in it and still achieve all the benefits of reactivity. If you upsert the results of getUserPoints into this collection, you can just write a helper to get the appropriate value for each user and it will update automatically.
userData = new Meteor.Collection(null);
// whenever you need to call "getUserPoints" use:
Meteor.call("getUserPoints", this._id, function(err, res) {
userData.upsert({userId: this._id}, {$set: {userId: this._id, points: res}});
});
Template.listItem.helpers({
userPoints: function() {
var pointsDoc = userData.findOne({userId: this._id});
return pointsDoc && pointsDoc.points;
}
});
There is an alternative way using the Tracker package (formerly Deps), which would be quick to implement here, but fiddly to scale. Essentially, you could set up a new Tracker.Dependency to track changes in user points:
var pointsDep = new Tracker.Dependency();
// whenever you call "getUserPoints":
Meteor.call("getUserPoints", this._id, function(err, res) {
...
pointsDep.changed();
});
Then just add a dummy helper to your listItem template (i.e. a helper that doesn't return anything by design):
<template name="listItem">
...
{{pointsCheck}}
</template>
Template.listItem.helpers({
pointsCheck: function() {
pointsDep.depend();
}
});
Whilst that won't return anything, it will force the template to rerender when pointsDep.changed() is called (which will be when new user points data is received).

Meteor "counts-by-room" example causing template not to load

I used that "counts-by-room" example from Meteor's docs to count the number of "reviews" for an "entry" on my project. Now, when I click through to an entry, the template isn't loading. It requires a refresh. Did I mess the example up somewhere? Here's my code. I don't really understand the "initializing" part of this, and don't know what to play with to get it to work correctly.
Server:
Meteor.publish("counts-by-entry", function (entryId) {
var self = this;
check(entryId, String);
var count = 0;
var initializing = true;
var handle = Reviews.find({entry: entryId}).observeChanges({
added: function (id) {
count++;
if (!initializing)
self.changed("counts", entryId, {count: count});
},
removed: function (id) {
count--;
self.changed("counts", entryId, {count: count});
}
});
initializing = false;
self.added("counts", entryId, {count: count});
self.ready();
self.onStop(function () {
handle.stop();
});
});
Client (iron-router):
this.route('entry', {
path: '/entry/:_id',
layoutTemplate: 'layout',
yieldTemplates: {
'navigation': {to: 'navigation'},
'pt_entry': {to: 'content'}
},
waitOn: function () {
Meteor.subscribe('singleEntry', this.params._id),
Meteor.subscribe('entryReviews', this.params._id),
Meteor.subscribe('counts-by-entry', this.params._id);
},
data: {
singleEntry: function () {return Entries.findOne()},
reviews: function () {return Reviews.find({entry: Session.get('entryId')}, {sort: {date: -1}})},
count: function () {return Counts.findOne(Session.get("entryId")).count + " reviews."}
},
before: function() {
Session.set("activeNav", Router.current().route.name),
Session.set("entryId", this.params._id);
},
notFoundTemplate: 'notFoundTemplate'
});
and also:
Counts = new Meteor.Collection("counts");

I just tried recreating the minimum amount required of your example to get it to work and I don't have the same issue as you.
Do you have the javascript console open in your browser? I would look for an error there, usually this kind of stuff happens when a helper is called upon when the collection data isn't available to the client. This is what iron-router's waitOn fixes which you have made use of.
In my reproduction I only have the one subscription (counts-by-entry) so maybe there is a issue with the other ones.
As for what the initializing part does:
The publish block is a piece of code that will be run for each client subscription. It does 2 things, it provides the client with the initial payload of data which in a traditional publication would be all the documents from a collection query then it reacts to changes that affects the result of the query and sends just those changes to the client.
Here is the most common publication you will see:
Meteor.publish("all-reviews", function() {
return Reviews.find({});
});
Meteor hides the complexities of what is really going on in this publication. This is closer to what is really going on:
Meteor.publish("all-reviews", function() {
var self = this;
//this is the query we want realtime updates for
//when relevant additions, removals or changes happen the correct callbacks will fire...
var handle = Reviews.find({}).observeChanges({
added: function(id, fields) {
//when a document is added it gets sent to the client.
//Note: the initial payload of data happens here, lets say you had 5 reviews
//this would get called 5 times as soon as a user subscribes.
self.added("reviews", id, fields);
},
removed: function(id) {
//when a document is removed the client is told which one
self.removed("reviews", id);
},
changed: function(id, fields) {
//when a document has a change made to its fields the changes get sent
self.changed("reviews", id, fields);
}
});
//letting the client know that the initial payload of data has been sent.
//Stuff like iron-routers waitOn would rely on this message before loading a template
//that relies on this publication
self.ready();
//stops the publication running forever. This will fire when a client no longer needs a
//subscription or when they close the page.
self.onStop(function() {
handle.stop();
});
});
As for what it going on in the docs example with the initializing flag. The initializing flag is used as a way of simply counting all the initial payload of existing reviews in your case then after the observeChanges call telling the client how many there are. This is an optimisation on the other way of doing it which would be to send the client several self.changed messages during the initial payload.
Maybe it will make more sense if I show how it would be done without the flag:
Meteor.publish("counts-by-entry", function (entryId) {
var self = this;
check(entryId, String);
var count = 0;
//we need to initially add the document we are going to increment
self.added("counts", entryId, {count: 0});
var handle = Reviews.find({entry: entryId}).observeChanges({
added: function (id) {
count++;
//So for example if there were 100 reviews for this entry when the user
//subscribes this self.changed would get called 100 times:
//self.changed("counts", entryId, {count: 1})
//self.changed("counts", entryId, {count: 2})
//self.changed("counts", entryId, {count: 3})
//self.changed("counts", entryId, {count: 4}) etc...
//which is a waste when we can just initially tell the client how
//many reviews there are at the point of them subscribing
self.changed("counts", entryId, {count: count});
},
removed: function (id) {
count--;
self.changed("counts", entryId, {count: count});
}
});
//remove the self.added(...) which handles the optimiation as explained above
self.ready();
self.onStop(function () {
handle.stop();
});
});
Regardless it doesn't look like that particular publish is the problem. I would expect the console to make it clear what the issue is

waitOn should return an array with the subscription handles.

Meteor observe changes added callback on server fires on all item

Tracker.autorun(function() {
DATA.find().observeChanges({
added: function(id, doc) {
console.log(doc);
}
});
});
This code is being called on the server. Every time the meteor server starts, the added function fires for every single item in the database. Is there a way to have the added callback fire only when new items are added?

added will be called for every document in the result set when observeChanges is first run. The trick is to ignore the callback during this initialization period. I have an expanded example in my answer to this question, but this code should work for you:
(function() {
var initializing = true;
DATA.find().observeChanges({
added: function(id, doc) {
if (!initializing) {
console.log(doc);
}
}
});
initializing = false;
})();
Note that Tracker.autorun is a client-only function. On the server I think it only ever executes once.

I struggled with this for a long time. For some reason, David's answer did not work for me - it was firing after the initializing variable was set to false.
This pattern from Avi was successful for me:
var usersLoaded = false;
Meteor.subscribe("profiles", function () {
// at this point all new users sent down are legitimately new ones
usersLoaded = true;
});
Meteor.users.find().observe({
added: function(user) {
if (usersLoaded) {
console.log("New user created: ", user);
}
}
});

Since it is initialization issue, you can do this.
var observerOfMessages = Messages.find({}).observe({
added: function(doc){
if(!observerOfMessages) return;
console.log(doc)
}
});
This is more elegant actually.

Provide a selector for the query which does not match old items. If using mongo ObjectID as _id you could query for items that have _id greater than the latest item's:
const latest = DATA.findOne({}, {sort: {_id: -1}})
DATA.find({_id: {$gt: latest._id}}).observeChanges({
added: function() { ... }
})
Or with createdAt timestamp:
const currentTime = new Date()
DATA.find({createdAt: {$gt: currentTime}}).observeChanges({
added: function() { ... }
})

Here's another way to solve this:
Meteor.subscribe('messages', function() {
var messages = Messages.find();
var msgCount = messages.count();
messages.observe({
addedAt: function(doc, atIndex) {
if(atIndex > (msgCount - 1)) console.log('added');
}
});
});
Should only fire for docs added after the existing amount is delivered. It's important that this goes in an onReady callback for Meteor.subscribe so that the msgCount changes as your subscription does... if for example, you're paginating your subscriptions.
cursor.observe() documentation

Ember get not getting certain attribute

When running the following from the UserController on Google Chrome, with ember-couchdb-kit-0.9, Ember Data v1.0.0-beta.3-56-g8367aa5, Ember v1.0.0, and this couchdb adapter:
customerSignUp: function () {
var model = this.get('model');
var customer = this.get('store').createRecord('customer', {
description: 'Why hello sir',
user: model
});
customer.save().then(function() {
model.set('customer', customer);
model.save();
});
}
with these models:
App.User = App.Person.extend({
name: DS.attr('string'),
customer: DS.belongsTo('customer', {async: true })
App.Customer = DS.Model.extend({
user: DS.belongsTo('user', {async: true}),
description: DS.attr('string')
});
neither the user nor the customer has their relationship set properly (in the Ember Debugger the user has null and the customer has <computed>, rather than some sort of <EmberPromiseObject> which is what they have when it works).
This only happens when the object in question is persisted. If the save() calls are omitted, both have correctly set relationships, but of course the database hasn't been updated with this information. Whenever the saves happen, the relationships are overwritten with empty entries.
I found that the problem was in the adapter's serializeBelongsTo function, which I've now changed my copy to the following:
serializeBelongsTo: function(record, json, relationship) {
console.log("serializeBelongsTo");
console.log(record.get('user'));
console.log(json);
console.log(relationship);
var attribute, belongsTo, key;
attribute = relationship.options.attribute || "id";
console.log(attribute);
key = relationship.key;
console.log(key);
belongsTo = Ember.get(record, key);
console.log(belongsTo);
if (Ember.isNone(belongsTo)) {
return;
}
json[key] = Ember.get(belongsTo, attribute);
console.log(Ember.get(belongsTo, attribute));
console.log(json);
if (relationship.options.polymorphic) {
return json[key + "_type"] = belongsTo.constructor.typeKey;
}
else {
return json;
}
}
attribute, belongsTo, and key all log as correct, but
console.log(Ember.get(belongsTo, attribute)); returns undefined,
which I've tried to change to
console.log(Ember.get(Ember.get(belongsTo, 'content'), attribute));
since console.log(belongsTo); told me the id attribute was hidden inside a content object. Attached is a screenshot showing what I mean.
The change doesn't fix the problem though, and I keep getting undefined. No matter what method I use to try to get the id out of the belongsTo object, I always get either null or undefined. Here are some examples of things I've tried to get content out of the object:
var content = belongsTo.content;
var content = Ember.get(belongsTo, 'content');
var content = belongsTo.get('content');
console.log(json); returns Object {description: "Why hello sir", user: undefined}
Here's a pastebin showing relevant output: http://pastebin.com/v4mb3PJ2
Update
A very confusing update!
When I save the model from a different function:
saveModel: function() {
this.get('model').save().then(
function( data, textStatus, jqXHR ) {
console.log('Saved successfully.');
},
function( jqXHR, textStatus, errorThrown ) {
console.log(jqXHR);
console.log(errorThrown);
console.log(textStatus);
}
);
}
The model is correctly saved. Everything in serializeBelongsto works exactly as expected.
Here's a different pastebin showing output for this case: http://pastebin.com/Vawur8Q0

I figured out the problem. Basically the belongsTo object in serializeBelongsTo wasn't really resolved by the time it was being referenced, which I found out by querying isFulfilled. So I implemented by saving side this way:
function saveOn (target, attribute) {
target.addObserver(attribute, function () {
if (target.get(attribute)) {
console.log("Inside with %#".fmt(attribute));
target.removeObserver(attribute);
Ember.run.once(target, function() {
target.save();
});
}
});
};
customerSignUp: function () {
var model = this.get('model');
var customer = this.get('store').createRecord('customer', {
description: 'Why hello sir'
});
customer.save().then(function () {
model.set('customer', customer);
customer.set('user', model);
saveOn(customer, 'user.isFulfilled');
saveOn(model, 'customer.isFulfilled');
});
}
Now everything works like a charm. It might be a good idea for serializeBelongsTo to take this into account though. This line: console.log(Ember.get(belongsTo, 'isFulfilled')); was coming up false in my case. There was just a race condition of some sort between the creation of the record and it's serialization!
I'd like to make my saveOn function return a promise though, which I could then use to chain multiple saveOns together. That way I wouldn't have to do a customer.save() to make sure the id's were populated.