Skip to content

Production-ready, scaling-ready Node + Express on Heroku with a message broker (node-redis-queue), sessions (redis), and storage (mongo).

Notifications You must be signed in to change notification settings

cwjohan/node-articles-nlp-redis-queue

 
 

Repository files navigation

Node-Articles-NLP-Redis-Queue

Running Node all the way from development to production on Heroku.

Check it out at http://node-articles-nlp-redis-queue.herokuapp.com/.

Deploy

Local dependencies

Installing

$ brew install redis mongodb
$ brew services start mongodb
$ brew services start redis
$ npm install

Running

  1. npm start
  2. http://localhost:3000

Deploying

Deploying is easy - just use the Heroku Button:

Deploy

If you'd rather clone locally and then deploy through the CLI, you can do that too:

git clone https://github.com/cwjohan/node-articles-nlp-redis-queue.git
cd node-articles-nlp

heroku create

heroku addons:add mongohq
heroku addons:add rediscloud

heroku config:set NODE_ENV=production
heroku config:set VIEW_CACHE=true
heroku config:set THRIFTY=true
heroku config:set QUEUE_CONFIG_FILE=/app/redis-queue-config.json

git push heroku master
heroku open

Config

Environment variables are mapped to a config object in lib/config.js. This provides reasonable defaults as well as a layer of generalization (process.env.REDISCLOUD_URL => config.redis_url).

You can locally override the defaults by adding variables to a .env file.

Scaling

The app is separated into two tiers:

This enables horizontally scaling both web traffic and long-running jobs.

On Heroku

The default deploy configuration includes THRIFTY=true, which starts the app in single-dyno mode (free!). With THRIFTY=true, the web process handles both http requests and queued jobs.

Similarly, the default configuration does not include a CONCURRENCY value, which means only one Cluster worker will be created per process. This is to keep under free levels of addon connection limits (like redis).

Of course, a production app should never run in a single instance or make users wait for worker processes. Additionally, allowing Cluster to take advantage of all the CPUs on a dyno can improve performance. When you're ready to test in staging or deploy to production, you can scale beyond single-dyno mode:

heroku config:unset THRIFTY
heroku config:set CONCURRENCY=4
heroku ps:scale web=2X:2 worker=2X:1

Change Procfile to be the same as Procfile2.

note: You'll need to upgrade your addons in order to scale your processes. The free addons from this app restrict the number of concurrent connections you can maintain.

Locally

npm start runs node-foreman, which will check the Procfile and start a single web process and a single worker process.

To test that your app behaves correctly when clustered in multiple processes, you can specify process scales to node-forman and set CONCURRENCY=4 in a local .env file.

Architecture

Writing maintainable Node apps is all about separating concerns into small, well-defined modules. This barebones app has three distinct components with their own responsibilities:

App

The business logic is all in lib/app. This module orchestrates and provides a facade for the underlying MongoDB database and the node-redis-queue job queue.

Web

The user-facing portion of the project lies in lib/web. This module is responsible for providing an http interface and routing requests. It shows things and relies on an App instance to do things.

Worker

The background processes run through lib/worker. This module is tiny - it just instantiates an App instance to process the job queue.

Change Log

v0.0.3: First cut at replacing amqp with node-redis-queue, but with vote queue disabled, due to a queue stall issue when the vote queue is empty upon starting the app.

v0.0.7: Fixes the stall issue; vote queue no longer disabled; works fine in local test environment; not working properly when deployed to Heroku; reason unknown but under investigation.

v0.0.8: Appears to fix issue with RedisCloud queues not working, but requires a separate worker process. Not working properly in single process thrifty mode when using remote RedisCloud. Works fine in a local test environment with a local instance of redis-server.

v0.0.9: Works in thrifty mode by using node-foreman to run separate web and worker sub-processes in a single Heroku web process.

v0.0.10: Upgraded node-redis-queue to version 0.1.10.

v0.0.11: Upgraded node-redis-queue to version 0.2.1. This required some class and variable name changes.

v0.0.12: Upgraded node-redis-queue to version 0.2.2. Re-enabled startScraping in thrifty mode in the web server. In connections.js, we now open the WorkQueueMgr connection in 'full-duplex' mode if config.thrifty is true. The Procfile now runs only the web dyno by default. That is, we assume thrifty mode.

About

Production-ready, scaling-ready Node + Express on Heroku with a message broker (node-redis-queue), sessions (redis), and storage (mongo).

Resources

Stars

Watchers

Forks

Packages

No packages published

Languages

  • JavaScript 91.2%
  • CSS 7.8%
  • Shell 1.0%