Redistat

A Redis-backed statistics storage and querying library written in Ruby.

Redistat was originally created to replace a small hacked together statistics collection solution which was MySQL-based. When I started I had a short list of requirements:

• Store and increment/decrement integer values (counters, etc)
• Up to the second statistics available at all times
• Screamingly fast

Redis fits perfectly with all of these requirements. It has atomic operations like increment, and it's lightning fast, meaning if the data is structured well, the initial stats reporting call will store data in a format that's instantly retrievable just as fast.

gem install redistat

If you are using Ruby 1.8.x, it's recommended you also install the SystemTimer gem, as the Redis gem will otherwise complain.

view_stats.rb:

require 'redistat'

class ViewStats
  include Redistat::Model
end

# if using Redistat in multiple threads set this
# somewhere in the beginning of the execution stack
Redistat.thread_safe = true

Store:

ViewStats.store('hello', {:world => 4})
ViewStats.store('hello', {:world => 2}, 2.hours.ago)

Fetch:

ViewStats.find('hello', 1.hour.ago, 1.hour.from_now).all
  #=> [{'world' => 4}]
ViewStats.find('hello', 1.hour.ago, 1.hour.from_now).total
  #=> {'world' => 4}
ViewStats.find('hello', 3.hour.ago, 1.hour.from_now).total
  #=> {'world' => 6}

Store page view on product #44 from Chrome 11:

ViewStats.store('views/product/44', {'count/chrome/11' => 1})

Fetch product #44 stats:

ViewStats.find('views/product/44', 23.hours.ago, 1.hour.from_now).total
  #=> { 'count' => 1, 'count/chrome' => 1, 'count/chrome/11' => 1 }

Store a page view on product #32 from Firefox 3:

ViewStats.store('views/product/32', {'count/firefox/3' => 1})

Fetch product #32 stats:

ViewStats.find('views/product/32', 23.hours.ago, 1.hour.from_now).total
  #=> { 'count' => 1, 'count/firefox' => 1, 'count/firefox/3' => 1 }

Fetch stats for all products:

ViewStats.find('views/product', 23.hours.ago, 1.hour.from_now).total
  #=> { 'count'           => 2,
  #     'count/chrome'    => 1,
  #     'count/chrome/11' => 1,
  #     'count/firefox'   => 1,
  #     'count/firefox/3' => 1 }

Store a 404 error view:

ViewStats.store('views/error/404', {'count/chrome/9' => 1})

Fetch stats for all views across the board:

ViewStats.find('views', 23.hours.ago, 1.hour.from_now).total
  #=> { 'count'           => 3,
  #     'count/chrome'    => 2,
  #     'count/chrome/9'  => 1,
  #     'count/chrome/11' => 1,
  #     'count/firefox'   => 1,
  #     'count/firefox/3' => 1 }

Fetch list of products known to Redistat:

finder = ViewStats.find('views/product', 23.hours.ago, 1.hour.from_now)
finder.children.map { |child| child.label.me }
  #=> [ "32", "44" ]
finder.children.map { |child| child.label.to_s }
  #=> [ "views/products/32", "views/products/44" ]
finder.children.map { |child| child.total }
  #=> [ { "count" => 1, "count/firefox" => 1, "count/firefox/3" => 1 },
  #     { "count" => 1, "count/chrome"  => 1, "count/chrome/11" => 1 } ]

A type of global-namespace for storing data. When using the Redistat::Model wrapper, the scope is automatically set to the class name. In the examples above, the scope is ViewStats. Can be overridden by calling the #scope class method on your model class.

Identifier string to separate different types and groups of statistics from each other. The first argument of the #store, #find, and #fetch methods is the label that you're storing to, or fetching from.

Labels support multiple grouping levels by splitting the label string with / and storing the same stats for each level. For example, when storing data to a label called views/product/44, the data is stored for the label you specify, and also for views/product and views. You may also configure a different group separator using the Redistat.group_separator= method. For example:

Redistat.group_separator = '|'

A word of caution: Don't use a crazy number of group levels. As two levels causes twice as many hincrby calls to Redis as not using the grouping feature. Hence using 10 grouping levels, causes 10 times as many write calls to Redis.

You provide Redistat with the data you want to store using a Ruby Hash. This data is then stored in a corresponding Redis hash with identical key/field names.

Key names in the hash also support grouping features similar to those available for Labels. Again, the more levels you use, the more write calls to Redis, so avoid using 10-15 levels.

Define how accurately data should be stored, and how accurately it's looked up when fetching it again. By default Redistat uses a depth value of :hour, which means it's impossible to separate two events which were stored at 10:18 and 10:23. In Redis they are both stored within a date key of 2011031610.

You can set depth within your model using the #depth class method. Available depths are: :year, :month, :day, :hour, :min, :sec

When you fetch data, you need to specify a start and an end time. The selection behavior can seem a bit weird at first when, but makes sense when you understand how Redistat works internally.

For example, if we are using a Depth value of :hour, and we trigger a fetch call starting at 1.hour.ago (13:34), till Time.now (14:34), only stats from 13:00:00 till 13:59:59 are returned, as they were all stored within the key for the 13th hour. If both 13:00 and 14:00 was returned, you would get results from two whole hours. Hence if you want up to the second data, use an end time of 1.hour.from_now.

Calling the #find method on a Redistat model class returns a Redistat::Finder object. The finder is a lazy-loaded gateway to your data. Meaning you can create a new finder, and modify instantiated finder's label, scope, dates, and more. It does not call Redis and fetch the data until you call #total, #all, #map, #each, or #each_with_index on the finder.

This section does need further expanding as there's a lot to cover when it comes to the finder.

Support for expiring keys from Redis is available, allowing you too keep varying levels of details for X period of time. This allows you easily keep things nice and tidy by only storing varying levels detailed stats only for as long as you need.

In the below example we define how long Redis keys for varying depths are stored. Second by second stats are available for 10 minutes, minute by minute stats for 6 hours, hourly stats for 3 months, daily stats for 2 years, and yearly stats are retained forever.

class ViewStats
  include Redistat::Model

  depth :sec

  expire \
    :sec => 10.minutes.to_i,
    :min => 6.hours.to_i,
    :hour => 3.months.to_i,
    :day => 2.years.to_i
end

Keep in mind that when storing stats for a custom date in the past for example, the expiry time for the keys will be relative to now. The values you specify are simply passed to the Redis#expire method.

Redistat stores all data into a Redis hash keys. The Redis key name the used consists of three parts. The scope, label, and datetime:

{scope}/{label}:{datetime}

For example, this...

ViewStats.store('views/product/44', {'count/chrome/11' => 1})

...would store the follow hash of data...

{ 'count' => 1, 'count/chrome' => 1, 'count/chrome/11' => 1 }

...to all 12 of these Redis hash keys...

ViewStats/views:2011
ViewStats/views:201103
ViewStats/views:20110315
ViewStats/views:2011031510
ViewStats/views/product:2011
ViewStats/views/product:201103
ViewStats/views/product:20110315
ViewStats/views/product:2011031510
ViewStats/views/product/44:2011
ViewStats/views/product/44:201103
ViewStats/views/product/44:20110315
ViewStats/views/product/44:2011031510

...by creating the Redis key, and/or hash field if needed, otherwise it simply increments the already existing data.

It would also create the following Redis sets to keep track of which child labels are available:

ViewStats.label_index:
ViewStats.label_index:views
ViewStats.label_index:views/product

It should now be more obvious to you why you should think about how you use the grouping capabilities so you don't go crazy and use 10-15 levels. Storing is done through Redis' hincrby call, which only supports a single key/field combo. Meaning the above example would call hincrby a total of 36 times to store the data, and sadd a total of 3 times to ensure the label index is accurate. 39 calls is however not a problem for Redis, most calls happen in less than 0.15ms (0.00015 seconds) on my local machine.

By default when fetching statistics, Redistat will figure out how to do the least number of reads from Redis. First it checks how long range you're fetching. If whole days, months or years for example fit within the start and end dates specified, it will fetch the one key for the day/month/year in question. It further drills down to the smaller units.

It is also intelligent enough to not fetch each day from 3-31 of a month, instead it would fetch the data for the whole month and the first two days, which are then removed from the summary of the whole month. This means three calls to hgetall instead of 29 if each whole day was fetched.

The buffer is a new, still semi-beta, feature aimed to reduce the number of Redis hincrby that Redistat sends. This should only really be useful when you're hitting north of 30,000 Redis requests per second, if your Redis server has limited resources, or against my recommendation you've opted to use 10, 20, or more label grouping levels.

Buffering tries to fold together multiple store calls into as few as possible by merging the statistics hashes from all calls and groups them based on scope, label, date depth, and more. You configure the the buffer by setting Redistat.buffer_size to an integer higher than 1. This basically tells Redistat how many store calls to buffer in memory before writing all data to Redis.

• More details in Readme.
• Documentation.
• Anything else that becomes apparent after real-world use.

Global Personals deserves a thank you. Currently the primary user of Redistat, they've allowed me to spend some company time to further develop the project.

• Fork the project.
• Make your feature addition or bug fix.
• Add tests for it. This is important so I don't break it in a future version unintentionally.
• Commit, do not mess with rakefile, version, or history. (if you want to have your own version, that is fine but bump version in a commit by itself I can ignore when I pull)
• Send me a pull request. Bonus points for topic branches.

Copyright (c) 2011 Jim Myhrberg.

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.