Meteor is awesome! But after 3 years, it's starting to show its age. This project is designed to showcase the exact same functionality as Meteor, but without the monolithic structure. It trades a little simplicity for a lot of flexibility.
Some of my chief complaints with Meteor
- Built on Node 0.10, and that ain't changing anytime soon
- Build system doesn't allow for code splitting (the opposite, in fact)
- Global scope (namespacing doesn't count)
- Goes Oprah-Christmas-special with websockets (not every person/page needs one)
- Can't handle css-modules (CSS is all handled behind the scenes)
- Tied to MongoDB for official support
Problem | Meteor's solution | My solution | Result |
---|---|---|---|
Database | MongoDB | RethinkDB | Built in reactivity, but you can use any DB you like |
Database schema | Simple Schema | GraphQL | Can't have a hipster webapp without GraphQL! |
Client validation | Simple Schema | Joi | Clean API for client validation, although the package is HUGE |
Database hooks | Collections2 | GraphQL | GraphQL is overkill for small apps (then again, so is meatier) |
Forms | AutoForm | redux-form | state tracking awesomeness that works beautifully with react |
Client-side cache | Minimongo | redux | Bonus logging, time traveling, and undo functionality |
Socket server | DDP-server | socketcluster | super easy scaling, pubsub, auth, middleware |
Authentication | Meteor accounts | JWTs | JWTs can also serve to authorize actions, too |
Auth-transport | DDP | GraphQL (via HTTP) | Don't use sockets until you need to |
Front-end | Blaze | React | Vdom, server-side rendering, async router, etc. |
Build system | meteor | webpack | using webpack inside meteor is very limited |
CSS | magically bundle & serve | css-modules | component-scoped css with variables available in a file or embedded |
Optimistic UI | latency compensation | redux-optimistic-ui | written by yours truly |
Testing | Velocity (or nothing at all) | AVA | awesome es2016 concurrent testing |
Linting | Your choice | xo | no dotfiles, fixes errors |
Routing | FlowRouter | react-router-redux | stick the route in the state, react-router SSR, async routes |
Server | Node 0.10.41 | Node 5 | Faster, maintained, not a dinosaur... |
##Installation
brew install rethinkdb
- make sure you are using [email protected] (not a v1 installed globally)
rethinkdb
(in second terminal window)git clone
this repocd meatier
npm install
npm run quickstart
##Client-side development
npm start
- https://localhost:3000
Rebuilds the client code in-memory & uses hot module reload so you can develop super fast! On my 2013 MBA an initial build takes about 8 seconds and updates usually take 800ms
##Server-side development
npm run prod
- https://localhost:3000
- If you edit any client or universal files, run
npm run bs
to rebuild & serve the bundle
This mode is great because you can make changes to the server without having to recompile the client code That means you only wait for the server to restart! GAME CHANGER!
##Database development
- https://localhost:8080 for RethinkDB
- All tables are managed in
./src/server/setupDB.js
. Just add your tables & indices to that file and rerun - A standard ORM would check for tables & ensure indices at least once per build, doing it this way keeps your build times down
- https://localhost:3000/graphql for testing out new queries/mutations
##Webpack configs ####Development config When the page is opened, a basic HTML layout is sent to the client along with a stringified redux store and a request for the common chunk of the JS. The client then injects the redux store & router to create the page. The redux devtools & logger are also loaded so you track your every state-changing action. The routes are loaded async, check your networks tab in chrome devtools and you'll see funny js files load now & again. If this isn't crazy amazing to you, then go away.
####Production config
Builds the website & saves it to the build
folder.
Maps the styles to the components, but uses the prerendered CSS from the server config (below)
Separates the vendor
packages and the app
packages for a super quick, cachable second visit.
Creates a webpack manifest to enable longterm caching (eg can push new vendor.js without pushing a new app.js)
Optimizes the number of chunks, sometimes it's better to have the modules of 2 routes in the same chunk if they're small
####Server config
A webpack config builds the entire contents of the routes on the server side.
This is required because node doesn't know how to require .css
.
When a request is sent to the server, react-router matches the url to the correct route & sends it to the client.
Any browser dependency is ignored & uglified away.
To test this, disable javascript in the browser. You'll see the site & css loads without a FOUC.
##How it works
When the page loads, it checks your localStorage for Meatier.token
& will automatically log you in if the token is legit.
If not, just head to the 'Sign up' page. The 'Sign up' page uses redux-form, which handles all errors, schema validation,
and submissions. Your credentials are set as variables in a GraphQL mutation & sent to the GraphQL endpoint and a user document (similar to Meteor's) and authToken is returned to your state.
The 'Kanban' app requires a login & websocket, so when you enter, your token will be used to authenticate a websocket.
That token is stored on the server so it is only sent during the handshake (very similar to DDP). Socket state is managed
by redux-socket-cluster
, just clicking socket
in the devtools let's you explore its current state.
When you enter the route, reducers are lazily loaded to the redux store and the redux-optimistic-ui
reducer enhancer is applied to the store to enable an optimistic UI. To work, it requires some middleware that scans each redux action for an isOptimistic
prop and reverts actions that fail server side.
When the kanban component loads, it subscribes to lanes
& notes
, which starts your personalized changefeed.
When you do something that changes the persisted state (eg add a kanban lane) that action is executed
optimistically on the client & emitted to the server where it is validated & sent to the database.
The database then emits a changefeed doc to all subscribed viewers.
Since the DB doesn't know which client made the mutation, it always sends a changefeed to the server.
The server is smart enough to ignore sending that document back to the originator, but it does send an acknowledgement.
The kanban lane titles & notes are really basic, you click them & they turn into input fields. The notes can be dragged from lane to lane. This is to showcase a local state change that doesn't affect the persisted state. When the note is dropped to its new location, the change is persisted.
##Tutorials (not for beginners...but then again, neither is meatier)
##Similar Projects
- https://github.com/erikras/react-redux-universal-hot-example (Really nice, but no auth or DB)
- https://github.com/kriasoft/react-starter-kit (nice, I borrowed their layout, but no sockets, no DB)
- https://github.com/GordyD/3ree (uses RethinkDB, but no optimistic UI)
- https://survivejs.com/ (A nice alt-flux & react tutorial for a kanban)
##In Action I don't know of any place that hosts RethinkDB for free...so here's a gif.
##Contributing
- Pull requests welcomed!
- Use the gitter for any questions
- No donations necessary (but if you know of any jobs that'll let me move back to San Diego, let me know 😉)
##Changelog
-
0.10
-
Use the redux devtools chrome extension
-
Update just about all the deps
-
Make deployments more configurable (deployment strategy coming soon)
-
0.9
-
Upgraded to
[email protected]
-
Now you can do cool things like time travel through routes!
-
0.8
-
Move auth & mutations to GraphQL (changefeeds still go through WS subs)
-
Make the rest of the
state.auth
immutable -
Add graphiql (https://localhost:3000/graphql) as a component & pattern to create an admin site
-
Break out auth, landing page, kanban page, and admin into 4 separate modules in the folder hierarchy
##License MIT