Skip to content

Latest commit

 

History

History
368 lines (268 loc) · 12.4 KB

README.md

File metadata and controls

368 lines (268 loc) · 12.4 KB
Raw

Upsert

Make it easy to upsert on traditional RDBMS like MySQL, PostgreSQL, and SQLite3—hey look NoSQL!. Transparently creates (and re-uses) stored procedures/functions when necessary.

You pass it a bare-metal connection to the database like Mysql2::Client (from mysql2 gem on MRI) or Java::OrgPostgresqlJdbc4::Jdbc4Connection (from jdbc-postgres on Jruby).

As databases start to natively support SQL MERGE (which is basically upsert), this library will take advantage (but you won't have to change your code).

Does not depend on ActiveRecord.

Does not use INSERT ON DUPLICATE KEY UPDATE on MySQL as this only works if you are very careful about creating unique indexes.

70–90%+ faster than emulating upsert with ActiveRecord.

Supports MRI and JRuby.

Usage

You pass a selector that uniquely identifies a row, whether it exists or not. You also pass a setter, attributes that should be set on that row.

Syntax inspired by mongo-ruby-driver's update method.

Basic

connection = Mysql2::Client.new([...])
table_name = :pets
upsert = Upsert.new connection, table_name
# N times...
upsert.row({:name => 'Jerry'}, :breed => 'beagle', :created_at => Time.now)

The created_at and created_on columns are used for inserts, but ignored on updates.

So just to reiterate you've got a selector and a setter:

selector = { :name => 'Jerry' }
setter = { :breed => 'beagle' }
upsert.row(selector, setter)

Batch mode

By organizing your upserts into a batch, we can do work behind the scenes to make them faster.

connection = Mysql2::Client.new([...])
Upsert.batch(connection, :pets) do |upsert|
  # N times...
  upsert.row({:name => 'Jerry'}, :breed => 'beagle')
  upsert.row({:name => 'Pierre'}, :breed => 'tabby')
end

Batch mode is tested to be about 80% faster on PostgreSQL, MySQL, and SQLite3 than other ways to emulate upsert (see the tests, which fail if they are not faster).

ActiveRecord helper method

require 'upsert/active_record_upsert'
# N times...
Pet.upsert({:name => 'Jerry'}, :breed => 'beagle')

Wishlist

Pull requests for any of these would be greatly appreciated:

  1. Cache JDBC PreparedStatement objects.
  2. Sanity check my three benchmarks (four if you include activerecord-import on MySQL). Do they accurately represent optimized alternatives?
  3. Provide require 'upsert/debug' that will make sure you are selecting on columns that have unique indexes
  4. Test that Upsert instances accept arbitrary columns, even within a batch, which is what people probably expect.
  5. @antage's idea for "true" upserting: (from seamusabshere#17)
selector = { id: 15 }
update_setter = { count: Upsert.sql('count + 1') }
insert_setter = { count: 1 }
upsert.row_with_two_setter(update_setter, insert_setter, selector)

Real-world usage

Faraday logo

We use upsert for big data at Faraday. Originally written to speed up the data_miner data mining library.

Supported databases/drivers

* MySQL PostgreSQL SQLite3
MRI mysql2 pg sqlite3
JRuby jdbc-mysql jdbc-postgres jdbc-sqlite3

See below for details about what SQL MERGE trick (emulation of upsert) is used, performance, code examples, etc.

Rails / ActiveRecord

(assuming that one of the other three supported drivers is being used under the covers)

Upsert.new Pet.connection, Pet.table_name

Speed

Depends on the driver being used!

SQL MERGE trick

Depends on the driver being used!

MySQL

On MRI, use the mysql2 driver.

require 'mysql2'
connection = Mysql2::Connection.new(:username => 'root', :password => 'password', :database => 'upsert_test')
table_name = :pets
upsert = Upsert.new(connection, table_name)

On JRuby, use the jdbc-mysql driver.

require 'jdbc/mysql'
java.sql.DriverManager.register_driver com.mysql.jdbc.Driver.new
connection = java.sql.DriverManager.get_connection "jdbc:mysql://127.0.0.1/mydatabase?user=root&password=password"

Speed

From the tests (updated 11/7/12):

Upsert was 82% faster than find + new/set/save
Upsert was 85% faster than find_or_create + update_attributes
Upsert was 90% faster than create + rescue/find/update
Upsert was 46% faster than faking upserts with activerecord-import

SQL MERGE trick

Thanks to Dennis Hennen's StackOverflow response!!

CREATE PROCEDURE upsert_pets_SEL_name_A_tag_number_SET_name_A_tag_number(`name_sel` varchar(255), `tag_number_sel` int(11), `name_set` varchar(255), `tag_number_set` int(11))
BEGIN
  DECLARE done BOOLEAN;
  REPEAT
    BEGIN
      -- If there is a unique key constraint error then 
      -- someone made a concurrent insert. Reset the sentinel
      -- and try again.
      DECLARE ER_DUP_UNIQUE CONDITION FOR 23000;
      DECLARE ER_INTEG CONDITION FOR 1062;
      DECLARE CONTINUE HANDLER FOR ER_DUP_UNIQUE BEGIN
        SET done = FALSE;
      END;
      
      DECLARE CONTINUE HANDLER FOR ER_INTEG BEGIN
        SET done = TRUE;
      END;

      SET done = TRUE;
      SELECT COUNT(*) INTO @count FROM `pets` WHERE `name` = `name_sel` AND `tag_number` = `tag_number_sel`;
      -- Race condition here. If a concurrent INSERT is made after
      -- the SELECT but before the INSERT below we'll get a duplicate
      -- key error. But the handler above will take care of that.
      IF @count > 0 THEN 
        -- UPDATE table_name SET b = b_SET WHERE a = a_SEL;
        UPDATE `pets` SET `name` = `name_set`, `tag_number` = `tag_number_set` WHERE `name` = `name_sel` AND `tag_number` = `tag_number_sel`;
      ELSE
        -- INSERT INTO table_name (a, b) VALUES (k, data);
        INSERT INTO `pets` (`name`, `tag_number`) VALUES (`name_set`, `tag_number_set`);
      END IF;
    END;
  UNTIL done END REPEAT;
END

PostgreSQL

On MRI, use the pg driver.

require 'pg'
connection = PG.connect(:dbname => 'upsert_test')
table_name = :pets
upsert = Upsert.new(connection, table_name)

On JRuby, use the jdbc-postgres driver.

require 'jdbc/postgres'
java.sql.DriverManager.register_driver org.postgresql.Driver.new
connection = java.sql.DriverManager.get_connection "jdbc:postgresql://127.0.0.1/mydatabase?user=root&password=password"

If you want to use HStore, make the pg-hstore gem available and pass a Hash in setters:

gem 'pg-hstore'
require 'pg_hstore'
upsert.row({:name => 'Bill'}, :mydata => {:a => 1, :b => 2})

Speed

From the tests (updated 9/21/12):

Upsert was 72% faster than find + new/set/save
Upsert was 79% faster than find_or_create + update_attributes
Upsert was 83% faster than create + rescue/find/update
# (can't compare to activerecord-import because you can't fake it on pg)

SQL MERGE trick

Adapted from the canonical PostgreSQL upsert example:

CREATE OR REPLACE FUNCTION upsert_pets_SEL_name_A_tag_number_SET_name_A_tag_number("name_sel" character varying(255), "tag_number_sel" integer, "name_set" character varying(255), "tag_number_set" integer) RETURNS VOID AS
$$
DECLARE
  first_try INTEGER := 1;
BEGIN
  LOOP
    -- first try to update the key
    UPDATE "pets" SET "name" = "name_set", "tag_number" = "tag_number_set"
      WHERE "name" = "name_sel" AND "tag_number" = "tag_number_sel";
    IF found THEN
      RETURN;
    END IF;
    -- not there, so try to insert the key
    -- if someone else inserts the same key concurrently,
    -- we could get a unique-key failure
    BEGIN
      INSERT INTO "pets"("name", "tag_number") VALUES ("name_set", "tag_number_set");
      RETURN;
    EXCEPTION WHEN unique_violation THEN
      -- seamusabshere 9/20/12 only retry once
      IF (first_try = 1) THEN
        first_try := 0;
      ELSE
        RETURN;
      END IF;
      -- Do nothing, and loop to try the UPDATE again.
    END;
  END LOOP;
END;
$$
LANGUAGE plpgsql;

I slightly modified it so that it only retries once - don't want infinite loops.

Sqlite

On MRI, use the sqlite3 driver.

require 'sqlite3'
connection = SQLite3::Database.open(':memory:')
table_name = :pets
upsert = Upsert.new(connection, table_name)

On JRuby, use the jdbc-sqlite3 driver.

# TODO somebody please verify
require 'jdbc/sqlite3'
java.sql.DriverManager.register_driver org.sqlite.Driver.new
connection = java.sql.DriverManager.get_connection "jdbc:sqlite://127.0.0.1/mydatabase?user=root&password=password"

Speed

From the tests (updated 9/21/12):

Upsert was 77% faster than find + new/set/save
Upsert was 80% faster than find_or_create + update_attributes
Upsert was 85% faster than create + rescue/find/update
# (can't compare to activerecord-import because you can't fake it on sqlite3)

SQL MERGE trick

Thanks to @dan04's answer on StackOverflow:

Please note! This will only work properly on Sqlite if one of the columns being used as the "selector" are a primary key or unique index

INSERT OR IGNORE INTO visits VALUES (127.0.0.1, 1);
UPDATE visits SET visits = 1 WHERE ip LIKE 127.0.0.1;

Features

Tested to be fast and portable

In addition to correctness, the library's tests check that it is

  1. Faster than comparable upsert techniques
  2. Compatible with supported databases

Not dependent on ActiveRecord

As below, all you need is a raw database connection like a Mysql2::Connection, PG::Connection or a SQLite3::Database. These are equivalent:

# with activerecord
Upsert.new ActiveRecord::Base.connection, :pets
# with activerecord, prettier
Upsert.new Pet.connection, Pet.table_name
# without activerecord
Upsert.new Mysql2::Connection.new([...]), :pets

For a specific use case, faster and more portable than activerecord-import

You could also use activerecord-import to upsert:

Pet.import columns, all_values, :timestamps => false, :on_duplicate_key_update => columns

activerecord-import, however, only works on MySQL and requires ActiveRecord—and if all you are doing is upserts, upsert is tested to be 40% faster. And you don't have to put all of the rows to be upserted into a single huge array - you can batch them using Upsert.batch.

Gotchas

No automatic typecasting beyond what the adapter/driver provides

We don't have any logic to convert integers into strings, strings into integers, etc. in order to satisfy PostgreSQL/etc.'s strictness on this issue.

So if you try to upsert a blank string ('') into an integer field in PostgreSQL, you will get an error.

Dates and times are converted to UTC

Datetimes are immediately converted to UTC and sent to the database as ISO8601 strings.

If you're using MySQL, make sure server/connection timezone is UTC. If you're using Rails and/or ActiveRecord, you might want to check ActiveRecord::Base.default_timezone... it should probably be :utc.

In general, run some upserts and make sure datetimes get persisted like you expect.

Doesn't work with transactional fixtures

Per seamusabshere#23 you might have issues if you try to use transactional fixtures and this library.

Copyright

Copyright 2014 Seamus Abshere