Skip to content

Latest commit

 

History

History
203 lines (149 loc) · 7.27 KB

README.md

File metadata and controls

203 lines (149 loc) · 7.27 KB
Raw

About

Lamer news is an implementation of a Reddit / Hacker News style news web site written using Ruby, Sinatra, Redis and jQuery.

The goal is to have a system that is very simple to understand and modify and that is able to handle a very high load using a small virtual server, ensuring at the same time a very low latency user experience.

This project was created in order to run http://lamernews.com but is free for everybody to use, fork, and have fun with.

We believe it is also a good programming example for Redis as a sole DB of a nontrivial, real world, web application.

Installation

Lamer news is a Ruby/Sinatra/Redis/jQuery application. You need to install Redis and Ruby 1.9.2+ with the following gems:

  • redis 3.0 or greater
  • hiredis
  • sinatra
  • json
  • ruby-hmac
  • net/smtp
  • openssl (not needed but will speedup the authentication if available).

Please note that Redis uses port 6379 by default, so you should either change the port number in the configuration file (which is set to use port 10000), or configure Redis to use a matching port.

How to contribute

I plan to hack on Lamer News in my free time as it is interesting to have a non trivial open source example for Redis that is also an useful application. However contributions are welcomed. Just make sure to:

  • Keep it simple. No complex code, no extreme ruby programming. Ideally non ruby people should understand the code without much efforts.
  • Don't use templates, they suck.
  • If your code slows down significantly the page generation time it will not get merged.
  • Do everything you can to avoid depending on new ruby gems.
  • Open an issue on github before firing your editor to see if there are good chances that your changes will be merged.
  • If you don't want to follow all this rules, forking the code is encouraged! The license is two clause BSD, do with this code what you want. Run your site, turn it into a blog, hack it to the extreme consequences. Have fun :)

Web sites using this code

Data Layout

Users

Every user is represented by the following fields:

A Redis hash named user:<user id> with the following fields:

id -> user ID
username -> The username
password -> Hashed password, PBKDF2(salt|password) note: | means concatenation
ctime -> Registration time (unix time)
karma -> User karma, earned visiting the site and posting good stuff
about -> Some optional info about the user
email -> Optional, used to show gravatars
auth -> authentication token
apisecret -> api POST requests secret code, to prevent CSRF attacks.
flags -> user flags. "a" enables administrative privileges.
karma_incr_time -> last time karma was incremented
pwd_reset -> unix time of the last password reset requested.
replies -> number of unread replies

Additionally for every user there is the following key:

`username.to.id:<lowercase_username>` -> User ID

This is used to lookup users by name.

Frequency of user posting is limited by a key named user:<user_id>:submitted_recently with TTL 15 minutes. If a user attempts to post before that key has expired an error message notifies the user of the amount of time until posting is permitted.

Account creation is rate limited by IP address with a key named limit:create_user:<ip_address> with TTL 15 hours.

Authentication

Users receive an authentication token after a valid pair of username/password is received. This token is in the form of a SHA1-sized hex number. The representation is a simple Redis key in the form:

`auth:<lowercase_token>` -> User ID

News

News are represented as an hash with key name news:<news id>. The hash has the following fields:

id -> News id
title -> News title
url -> News url
user_id => The User ID that posted the news
ctime -> News creation time. Unix time.
score -> News score. See source to check how this is computed.
rank -> News score adjusted by age: RANK = SCORE / AGE^ALPHA
up -> Counter with number of upvotes
down -> Counter with number of downvotes
comments -> number of comments

Note: up, down, comments fields are also available in other ways but we denormalize for speed.

Also recently posted urls have a key named url:<actual full url> with TTL 48 hours and set to the news ID of a recently posted news having this url.

So if another user will try to post a given content again within 48 hours the system will simply redirect it to the previous news.

News is never deleted, but just marked as deleted adding the "del" field with value 1 to the news object. However when the post is rendered into HTML, it is displayed as [deleted news] text.

News votes

Every news has a sorted set with user upvotes and downvotes. The keys are named respectively news.up:<news id> and news.down:<news id>.

In the sorted sets the score is the unix time of the vote creation, the element is the user ID of the voting user.

Posting a news will automatically register an up vote from the user posting the news.

Saved news

The system stores a list of upvoted news for every user using a sorted set named user.saved:<user id>, index by unix time. The value of the sorted set elements is the <news id>.

Submitted news

Like saved news every user has an associated sorted set with news he posted. The key is called user.posted:<user id>. Again the score is the unix time and the element is the news id.

Top and Latest news

news.cron is used to generate the "Latest News" page. It is a sorted set where the score is the Unix time the news was posted, and the value is the news ID.

news.top is used to generate the "Top News" page. It is a sorted set where the score is the "RANK" of the news, and the value is the news ID.

Comments

Comments are represented using a very memory efficient pattern. The system is implemented in the comments.rb file.

In short every thread (that is a collection of comments for a given news) is represented by an hash. Every hash entry represents a single comment:

  • The hash field is the comment ID.
  • The hash value is a JSON representation of the "comment object".

The comment object has many fields, like ctime (creation time), body, user_id, and so forth. In order to render all the comments for a thread we simply do an HGETALL to fetch everything. Then we run the list of returned comments and build a graph of comments, calling a recursive function with this graph as input.

Comments are never deleted, but just marked as deleted adding the "del" field with value 1 to the comment object. However when the thread is rendered into HTML deleted comments without childs are not displayed. Deleted comments with childs are displayed as [deleted comment] text.

Please check comments.rb for details, it is trivial to read.

User comments

All the comments posted by a given user are also taken into a sorted set of comments, keyed by creation time. The key name is: user.comments:<userid>.

In this sorted set the score is the unix time and the value is a string composed in this way: <newsid>-<commentid>. So a unique comment is referenced by the news id and the id of the comment inside the hash of comments for this news. Example of an actual comment pointer: 882-15.