Ode to Sequel
I’ve used and loved ActiveRecord for most of my Ruby life. While I was in Rails, I couldn’t imagine why I would want to use anything else. When I moved away from Rails, I was still using ActiveRecord at first, but some things started to bother me:
- limited query interface, you very quickly have to switch to SQL strings
- no good low-level query interface (Arel is not writable)
- very fuzzy LEFT JOIN support
- it’s pretty difficult to set up ActiveRecord in a non-Rails project
- dependency on ActiveSupport and its core extensions (you may not want tons of monkey patches if you’re using ActiveRecord outside of Rails)
I wanted to try another ORM. I’ve thought about ROM, but I felt like it required a completely different mindset. I wanted a gem which also implements the ActiveRecord pattern, but with a better implementation than the ActiveRecord gem. I’ve heard about Sequel before, and I decided it was finally high time to try it out.
Since then I’ve had a fair amount of experience with Sequel and its community, and it’s been amazing. Sequel has all the features I wanted from ActiveRecord, and so much more, even features I didn’t know I wanted. Jeremy Evans, the author of Sequel, keeps Sequel at 0 issues, and also keeps a mailing list where you can get help with anything.
I would like to give you some of my personal highlights of Sequel’s amazing features.
NOTE: “Sequel” should not be confused with “Squeel”. Squeel is an extension for ActiveRecord’s query DSL, while Sequel is a complete ORM and an alternative to ActiveRecord.
The plugin system
While ActiveRecord is one monolithic gem, Sequel utilizes a plugin system. Sequel consists of a relatively thin core, which gives you the most common behaviour, and you can then choose to add additional functionality via plugins. Each plugin corresponds to a single file included in the gem, but which is required only when the plugin is loaded.
Because of this design vanilla Sequel loads 5 times faster than ActiveRecord.
Query interface
Sequel’s query interface is very similar to ActiveRecord’s, but much more advanced.
Regular expressions
Did you know that PostgresSQL and MySQL have support for POSIX regular expressions? Neither did I. If you’re using one of these two databases, Sequel will transform Ruby regular expressions to SQL:
Movie.where(name: /future/)
# SELECT * FROM movies WHERE (name ~ 'future')
Movie.where(name: /future/i) # Case insensitive match
# SELECT * FROM movies WHERE (name ~* 'future')
This means that you can simply replace all your ugly LIKE queries with beautiful regular expressions! Just note that regex matches are usually slower than LIKE queries (in my benchmarks they were twice as slow), so be sure to measure in your application how this impacts the performance. If it does, LIKE queries are still nicer to write in Sequel (see below).
Virtual row blocks
Most of Sequel’s query methods, in addition to arguments, also support blocks (so-called “virtual row blocks”) which gives you a DSL for more advanced queries.
Movie.where{year >= 2010} # inequality operators
# WHERE (year >= 2010)
Movie.where{(title =~ "Batman") | (year < 2010)} # OR query
# WHERE ((title = 'Batman') OR (year < 2010))
Movie.where{rating >= avg(rating)} # functions get translated to SQL
# WHERE (rating >= avg(rating)
Movie.where{title.like("%Future%")} # special methods
# WHERE (title LIKE '%FUTURE%')
You may be familiar with this syntax if you’ve ever used the Squeel gem. This is because Squeel originally borrowed this syntax from Sequel (hence the play of characters in the name). But the problem with Squeel is that it’s essentially an ActiveRecord hack, so it breaks with every ActiveRecord update. Virtual row blocks are a part of Sequel core, so they will always remain fully stable.
While in ActiveRecord you often have to switch to SQL strings (OR query, LIKE query, any non-canonic JOIN etc.), with Sequel’s virtual row blocks you essentially never have to write SQL strings.
Low-level usage
When working with large amounts of data, the time it takes to allocate all these ActiveRecord objects can very quickly surpass the time it takes for actual queries to execute. This is often the case in ActiveRecod migrations, where you’re operating on whole production tables. Ideally you want to work instead with light data structures, like hashes. However, as far as I know that’s not possible when querying through ActiveRecord models.
Some more advanced ActiveRecord users might be thinking that Arel, a library that ActiveRecord uses underneath to build SQL queries, is a good fit here. Performance-wise it probably is, but Arel is very difficult to use and the resulting code is often very unreadable (even with arel-helpers).
On the other hand, Sequel allows you to write low-level queries using the
exact same query interface you use for models! Instead of going through
models, You can go through the Sequel::Database
object directly, and the
records will be returned as simple Ruby hashes.
DB = Sequel.connect("postgres:///my_database") #=> #<Sequel::Database>
Movie.where(title: "Matrix").first #=> #<Movie title="Matrix" year=1999 ...>
DB[:movies].where(title: "Matrix").first #=> {title: "Matrix", year: 1999, ...}
# DB[:movies].sql #=> "SELECT * FROM movies"
This means that with Sequel you can write very readable migrations (because you don’t have to redefine your models inside migrations), and have them be blazing-fast!
Model design
Where ActiveRecord uses class-level DSL, Sequel instead prefers simple OO
design. For example, the idiomatic way to write validations in Sequel is by
overriding the instance method #validate
:
class Movie < Sequel::Model
plugin :validation_helpers
def validate
validates_presence [:name, :year]
validates_includes 1..10, :rating
if genre == "Horror"
validates_presence :rated
end
end
end
I personally find this way of writing validations much more natural than
ActiveRecord’s class-level DSL, since I’m not constrained to :if
and
:unless
options. As a bonus, if validations become more complex, I can pull
them out of the model into a service object, and still be able to use the
convenient helper methods (since they’re instance-level).
JOINs
I’m not an SQL guru, but LEFT JOINs are really common in SQL. ActiveRecord
unfortunately doesn’t directly support LEFT JOINs. The #join
method only does
an INNER JOIN by default, and while you can use it to write a custom JOIN
statement, it’s really verbose as you have to write the full SQL string with
all the column-joining logic.
# LEFT JOINs in ActiveRecord
Movie.joins("LEFT JOIN on directors ON directors.movie_id = movies.id")
You could also use includes(:directors).refrences(:directors)
, which does a
LEFT JOIN, but this also eager loads your directors into memory, which is
unfortunate if you don’t need that.
Sequel, on the other hand, has support for ALL types of JOINs. You can do joins through associations, or write them manually:
Movie.association_left_join(:directors) # association_(left|right|inner|cross|...)_join
Movie.left_join(:directors, movie_id: :id) # (left|right|inner|cross|...)_join
Postgres-specific support
Jeremy Evans really loves Postgres (as everyone should), and he put a lot of effort into supporting as many Postgres features as possible in Sequel. And Postgres has a LOT of features.
JSON
Sequel supports reading and writing to JSON columns, but so does ActiveRecord, so what’s the big deal? What ActiveRecord doesn’t have is an API for querying JSON columns, which is the reason you’re using JSON columns in the first place. The problem is that Postgres’ JSON operators can be quite cryptic, which hurts your codebase. Luckily, Sequel provides a nice, readable API to help you with that:
Sequel.extension :pg_json_ops # we load the plugin ("ops" stands for "operations")
# Let's say that the `movies` table has an "info" JSON column.
info = Sequel.pg_json_op(:info) # we create a "JSON operation" object
Movie.where(info.has_key?('rated')) # WHERE (info ? 'rated')
Movie.where(info.get_text('rated') => 'PG-13') # WHERE ((info ->> 'rated') = 'PG-13')
Movie.order(info.get_text(['directors', 1, 'name'])) # ORDER BY (info #>> ARRAY['directors', 1, 'name'])
Views
Sequel makes it very simple and intuitive to write database views – just use the query interface!
DB.create_view :recent_ruby_items, DB[:items].where(category: "ruby").limit(5)
# CREATE VIEW recent_ruby_items AS
# SELECT * from items WHERE category = 'ruby' LIMIT 5
Database views are very helpful for DRYing up some common queries, as you can
query them as tables (Thoughtbot used views to implement multi-table full-text
search in Postgres). Moreover, unlike other databases,
Postgres has materialized views, which transforms views
into sort of temporary tables by caching them, which can really help you speed
up complex queries. Sequel supports materialized views by
accepting materialized: true
in DB.create_view
.
Cursors
Sequel has a #paged_each
method, which is an equivalent of ActiveRecord’s
#find_each
. This method is used to iterate over large datasets
without having to load all records into memory. By default these methods use a
separate query for each iteration, changing LIMIT and OFFSET to mimic paging.
Sequel, if it detects you’re using Postgres, will instead change #paged_each
to use Postgres cursors under the hood, which are faster than additional
queries and work with unordered datasets.
Movie.paged_each { |row| ... }
# BEGIN;
# DECLARE sequel_cursor NO SCROLL CURSOR WITHOUT HOLD FOR SELECT * FROM "table";
# FETCH FORWARD 1000 FROM sequel_cursor
# FETCH FORWARD 1000 FROM sequel_cursor
# ...
# FETCH FORWARD 1000 FROM sequel_cursor
# CLOSE sequel_cursor
# COMMIT
sequel_pg
sequel_pg is a gem that provides a C extension which optimizes the fetching of rows, generally resulting in a 2-6x speedup. So, you just add the gem to your Gemfile and get free performance.
In addition to optimization, sequel_pg also adds streaming support if used on
PostgreSQL 9.2. Steaming support is similar to using a cursor, but it is faster
and more transparent. #paged_each
will automatically use streaming to
implement paging if enabled.
Other goodies
- Support for PostgresSQL’s LISTEN/NOTIFY commands (e.g. queue_classic uses this feature):
DB.loose_count(:users)
for fast approximate counts using Postgres’ system tables (COUNT queries can be slow on larger tables)- “pg_array_associations” plugin which enables you to avoid an additional join table in “has and belongs to many” associations by keeping foreign keys in a Postgres array column instead
- and many more…
Switching from ActiveRecord
Sequel has a very exhaustive guide “Sequel for ActiveRecord Users”, which is aimed at helping ActiveRecord users transition to Sequel. The guide first explains how each ActiveRecord feature is implemented in Sequel, and mentions some Sequel plugins you could use in your transition to make Sequel more similar to ActiveRecord. Then it lists how each and every ActiveRecord’s method and option correspond to Sequel. Pretty good, huh?
There is a sequel-rails gem which is actively maintained, which helps you keep the same development workflow as you had with Rails. It provides:
- Database Rake tasks (for migrations and such)
- Migration/model generators
Sequel::NoMatchingRow
error is returned as 404 (some other errors are mapped as well)- Logging is integrated into Rails logs
- And more…
Conclusion
Even after all of this, I have only scratched the surface of Sequel’s amazing features. ActiveRecord was long my ORM of choice only because it’s part of Rails, not because it was the best. After using Sequel for a period of time, I have found it to be much more stable (0 issues maintained), better designed, more performant (benchmark), and more advanced than ActiveRecord. It encourages you to make the most out of your database. I urge you to give it a try!