Easy maintenance of collections of processed data in MongoDB with Mongoid 3, 4 and 5.
Suppose that you have a Mongoid model called Artwork
, stored in a MongoDB collection called artworks
and the underlying documents look something like:
{ name: 'Flowers', artist: 'Andy Warhol', price: 3000000 }
From time to time, your system runs a map/reduce job to compute the average price of each artist's works, resulting in a collection called artist_average_price
that contains documents that look like:
{ _id: { artist: 'Andy Warhol' }, value: { price: 1500000 } }
If your system wants to maintain and use this average price data, it has to do so at the level of raw MongoDB operations, since map/reduce result documents don't map well to models in Mongoid.
Furthermore, even though map/reduce jobs can take some time to run, you probably want the entire artist_average_price
collection populated atomically from the point of view of your system, since otherwise you don't ever know the state of the data in the collection - you could access it in the middle of a map/reduce and get partial, incorrect results.
A mongoid_collection_snapshot solves this problem by providing an atomic view of collections of data like map/reduce results that live outside of Mongoid.
In the example above, we'd set up our average artist price collection like:
class AverageArtistPrice
include Mongoid::CollectionSnapshot
def build
map = <<-EOS
function() {
emit({ artist_id: this['artist_id']}, { count: 1, sum: this['price'] })
}
EOS
reduce = <<-EOS
function(key, values) {
var sum = 0;
var count = 0;
values.forEach(function(value) {
sum += value['sum'];
count += value['count'];
});
return({ count: count, sum: sum });
}
EOS
Artwork.map_reduce(map, reduce).out(inline: 1).each do |doc|
collection_snapshot.insert_one(
artist_id: doc['_id']['artist_id'],
count: doc['value']['count'],
sum: doc['value']['sum']
)
end
end
end
Now, if you want to schedule a recomputation, just call AverageArtistPrice.create
. You can define other methods on collection snapshots.
class AverageArtistPrice
...
def average_price(artist_name)
artist = Artist.where(name: artist_name).first
doc = collection_snapshot.where(artist_id: artist.id).first
doc['sum'] / doc['count']
end
end
The latest snapshot is always available as AverageArtistPrice.latest
, so you can write code like:
warhol_expected_price = AverageArtistPrice.latest.average_price('Andy Warhol')
And always be sure that you'll never be looking at partial results. The only thing you need to do to hook into mongoid_collection_snapshot is implement the method build
, which populates the collection snapshot and any indexes you need.
By default, mongoid_collection_snapshot maintains the most recent two snapshots computed any given time.
You can do better than the average price example above and define first-class models for your collection snapshot data, then access them as any other Mongoid collection via collection snapshot's .documents
method.
class AverageArtistPrice
document do
belongs_to :artist, inverse_of: nil
field :sum, type: Integer
field :count, type: Integer
end
def average_price(artist_name)
artist = Artist.where(name: artist_name).first
doc = documents.where(artist: artist).first
doc.sum / doc.count
end
end
Another example iterates through all latest artist price averages.
AverageArtistPrice.latest.documents.each do |doc|
puts "#{doc.artist.name}: #{doc.sum / doc.count}"
end
You can maintain multiple collections atomically within the same snapshot by passing unique collection identifiers to collection_snaphot
when you call it in your build or query methods:
class ArtistStats
include Mongoid::CollectionSnapshot
def build
# ...
# define map/reduce for average and max aggregations
# ...
Mongoid.default_session.command('mapreduce' => 'artworks', map: map_avg, reduce: reduce_avg, out: collection_snapshot('average'))
Mongoid.default_session.command('mapreduce' => 'artworks', map: map_max, reduce: reduce_max, out: collection_snapshot('max'))
end
def average_price(artist)
doc = collection_snapshot('average').find('_id.artist' => artist).first
doc['value']['sum'] / doc['value']['count']
end
def max_price(artist)
doc = collection_snapshot('max').find('_id.artist' => artist).first
doc['value']['max']
end
end
Specify the name of the collection to define first class Mongoid models.
class ArtistStats
document('average') do
field :value, type: Hash
end
document('max') do
field :value, type: Hash
end
end
Access these by name.
ArtistStats.latest.documents('average')
ArtistStats.latest.documents('max')
If fields across multiple collection snapshots are identical, a single default document
is sufficient.
class ArtistStats
document do
field :value, type: Hash
end
end
Your class can specify a custom database for storage of collection snapshots by overriding the snapshot_session
instance method. In this example, we memoize the connection at the class level to avoid creating many separate connection instances.
class ArtistStats
include Mongoid::CollectionSnapshot
def build
# ...
end
def snapshot_session
self.class.snapshot_session
end
def self.snapshot_session
@@snapshot_session ||= Mongo::Client.new('mongodb://localhost:27017').tap do |c|
c.use :alternate_db
end
end
end
Another common way of configuring this is through mongoid.yml.
development:
sessions:
default:
database: dev_data
imports:
database: dev_imports
def snapshot_session
Mongoid.session('imports')
end
MIT License, see LICENSE.txt for details.