Forklift ETL
Moving heavy databases around.
What?
Forklift is a ruby gem that makes it easy for you to move your data around. Forklift can be an integral part of your datawarehouse pipeline or a backup tool. Forklift can collect and collapse data from multiple sources or across a single source. In forklift's first version, it was only a MySQL tool but now, you can create transports to deal with the data of your choice.
Set up
Make a new directory with a Gemfile
like this:
source 'http://rubygems.org'
gem 'forklift_etl'
Then bundle
Use the generator by doing (bundle exec) forklift --generate
Make your plan.rb
using the examples below.
Run your plan forklift plan.rb
You can run specific parts of your plan like forklift plan.rb step1 step5
Directory structure
Forklift expects your project to be arranged like:
├── config/
| ├── email.yml
├── connections/
| ├── mysql/
| ├── (DB).yml
| ├── elasticsearch/
| ├── (DB).yml
| ├── csv/
| ├── (file).yml
├── log/
├── pid/
├── template/
├── patterns/
├── transformations/
├── Gemfile
├── Gemfile.lock
├── plan.rb
To enable a foklift connection, all you need to do is place the yml config file for it within /config/connections/(type)/(name).yml
Files you place within /patterns/
or connections/(type)/
will be loaded automatically.
Examples
Example Project
Visit the /example
directory to see a whole forklift project.
Simple extract and load (no transformations)
If you have multiple databases and want to consolidate into one, this plan should suffice.
plan = Forklift::Plan.new
plan.do! do
# ==> Connections
service1 = plan.connections[:mysql][:service1]
service2 = plan.connections[:mysql][:service2]
analytics_working = plan.connections[:mysql][:analytics_working]
analytics = plan.connections[:mysql][:analytics]
# ==> Extract
# Load data from your services into your working database
# If you want every table: service1.tables.each do |table|
# Data will be extracted in 1000 row collections
%w(users organizations).each do |table|
service1.read("select * from `#{table}`") { |data| analytics_working.write(data, table) }
end
%w(orders line_items).each do |table|
service2.read("select * from `#{table}`") { |data| analytics_working.write(data, table) }
end
# ==> Load
# Load data from the working database to the final database
analytics_working.tables.each do |table|
# will attempt to do an incremental pipe, will fall back to a full table copy
# by default, incremental updates happen off of the `updated_at` column, but you can modify this by setting the `matcher` in the options
# If you want a full pipe instead of incremental, then just use `pipe` instead of `optimistic_pipe`
# The `pipe pattern` works within the same database. To copy across databases, try the `mysql_optimistic_import` method
# This example show the options with their default values.
Forklift::Patterns::Mysql.optimistic_pipe(analytics_working.current_database, table, analytics.current_database, table, matcher: 'updated_at', primary_key: 'id')
end
end
Simple MySQL ETL
plan = Forklift::Plan.new
plan.do! do
# Do some SQL transformations
# SQL transformations are done exactly as they are written
destination = plan.connections[:mysql][:destination]
destination.exec!("./transformations/combined_name.sql")
# Do some Ruby transformations
# Ruby transformations expect `do!(connection, forklift)` to be defined
destination = plan.connections[:mysql][:destination]
destination.exec!("./transformations/email_suffix.rb")
# mySQL Dump the destination
destination = plan.connections[:mysql][:destination]
destination.dump('/tmp/destination.sql.gz')
end
Elasticsearch to MySQL
plan = Forklift::Plan.new
plan.do! do
source = plan.connections[:elasticsearch][:source]
destination = plan.connections[:mysql][:destination]
table = 'es_import'
index = 'aaa'
query = { query: { match_all: {} } } # pagination will happen automatically
destination.truncate!(table) if destination.tables.include? table
source.read(index, query) {|data| destination.write(data, table) }
end
MySQL to Elasticsearch
plan = Forklift::Plan.new
plan.do! do
source = plan.connections[:mysql][:source]
destination = plan.connections[:elasticsearch][:source]
table = 'users'
index = 'users'
query = "select * from users" # pagination will happen automatically
source.read(query) {|data| destination.write(data, table, true, 'user') }
end
Forklift Emails
Setup
Put this at the end of your plan inside the do!
block.
# ==> Email
# Let your team know the outcome. Attaches the log.
email_args = {
to: "team@yourcompany.com",
from: "Forklift",
subject: "Forklift has moved your database @ #{Time.new}",
body: "So much data!"
}
plan.mailer.send(email_args, plan.logger.messages)
ERB templates
You can get fancy by using an ERB template for your email and SQL variables:
# ==> Email
# Let your team know the outcome. Attaches the log.
email_args = {
to: "team@yourcompany.com",
from: "Forklift",
subject: "Forklift has moved your database @ #{Time.new}"
}
email_variables = {
total_users_count: service1.read('select count(1) as "count" from users')[0][:count]
}
email_template = "./template/email.erb"
plan.mailer.send_template(email_args, email_template, email_variables, plan.logger.messages)
Then in template/email.erb
:
<h1>Your forklift email</h1>
<ul>
<li><strong>Total Users</strong>: <%= @total_users_count %></li>
</ul>
Config
When you run forklift --generate
, we create config/email.yml
for you:
# Configuration is passed to Pony (https://github.com/benprew/pony)
# ==> SMTP
# If testing locally, mailcatcher (https://github.com/sj26/mailcatcher) is a helpful gem
via: smtp
via_options:
address: localhost
port: 1025
# user_name: user
# password: password
# authentication: :plain # :plain, :login, :cram_md5, no auth by default
# domain: "localhost.localdomain" # the HELO domain provided by the client to the server
# ==> Sendmail
# via: sendmail
# via_options:
# location: /usr/sbin/sendmail
# arguments: '-t -i'
Workflow
# do! is a wrapper around common setup methods (pidfile locking, setting up the logger, etc)
# you don't need to use do! if you want finer control
def do!
# you can use `plan.logger.log` in your plan for logging
self.logger.log "Starting forklift"
# use a pidfile to ensure that only one instance of forklift is running at a time; store the file if OK
self.pid.safe_to_run?
self.pid.store!
# this will load all connections in /config/connections/#{type}/#{name}.yml into the plan.connections hash
# and build all the connection objects (and try to connect in some cases)
self.connect!
yield # your stuff here!
# remove the pidfile
self.logger.log "Completed forklift"
self.pid.delete!
end
Steps
You can optionally divide up your forklift plan into steps:
plan = Forklift::Plan.new
plan.do! do
plan.step('Mysql Import'){
source = plan.connections[:mysql][:source]
destination = plan.connections[:mysql][:destination]
source.tables.each do |table|
Forklift::Patterns::Mysql.optimistic_pipe(source, table, destination, table)
end
}
plan.step('Elasticsearch Import'){
source = plan.connections[:elasticsearch][:source]
destination = plan.connections[:mysql][:destination]
table = 'es_import'
index = 'aaa'
query = { query: { match_all: {} } } # pagination will happen automatically
destination.truncate!(table) if destination.tables.include? table
source.read(index, query) {|data| destination.write(data, table) }
}
end
When you use steps, you can run your whole plan, or just part if it with command line arguments. For example, forklift plan.rb "Elasticsearch Import"
would just run that single portion of the plan. Note that any parts of your plan not within a step will be run each time.
Error Handling
By default, exceptions within your plan will raise and crash your application. However, you can pass an optional error_handler
lambda to your step about how to handle the error. the error_handler
will be passed (step_name
,exception
). If you don't re-raise within your error handler, your plan will continue to excecute. For example:
error_handler = lambda { |name, exception|
if exception.class =~ /connection/
# I can't connect, I should halt
raise e
elsif exception.class =~ /SoftError/
# this type of error is OK
else
raise e
end
}
plan.step('a_complex_step', error_handler){
# ...
}
Transports
Transports are how you interact with your data. Every transport defines read
and write
methods which handle arrays of data objects (and the helper methods required).
Each transport should have a config file in ./config/connections/#{transport}/
. It will be loaded at boot.
Transports optionally define helper methods which are a shortcut to copy data within a transport, like the mysql pipe
methods (i.e.: insert into #{to_db}.#{to_table}; select * from #{from_db}.#{from_table})
. A transport may also define other helpers (like how to create a MySQL dump). These should be defined in /patterns/#{type}.rb
within the Forklift::Patterns::#{type}
namespace.
Creating your own transport
In the /connections
directory in your project, create a file that defines at least the following:
module Forklift
module Connection
class Mixpanel < Forklift::Base::Connection
def initialize(config, forklift)
@config = config
@forklift = forklift
end
def config
@config
end
def forklift
@forklift
end
def read(index, query, args)
# ...
data = [] # data is an array of hashes
# ...
if block_given?
yield data
else
return data
end
end
def write(data, table)
# data is an array of hashes
# "table" can be any argument(s) you need to know where/how to write
# ...
end
def pipe(from_table, from_db, to_table, to_db)
# ...
end
private
#/private
end
end
end
Existing transports and patterns for them are documented here
MySQL
Elasticsearch
Csv
Transformations
Forklift allows you to create both Ruby transformations and script transformations.
- It is up to the transport to define
exec_script
, and not all transports will support it. Mysql can run.sql
files, but there is not an equivalent for elasticsearch. Mysql scripts evaluate statement by statement. The delimeter (by default;
) can be redefined using thedelimeter
command as described here -
.exec
runs and logs exceptions, while.exec!
will raise on an error. For example,destination.exec("./transformations/cleanup.rb")
will run cleanup.rb on the destination database. - Script files are run as-is, but ruby transformations must define a
do!
method in their class and are passeddef do!(connection, forklift)
- args is optional, and can be passed in from your plan
# Example transformation to count users
# count_users.rb
class CountUsers
def do!(connection, forklift, args)
forklift.logger.log "counting users"
count = connection.count('users')
forklift.logger.log "[#{args.name}] found #{count} users"
end
end
# in your plan.rb
plan = Forklift::Plan.new
plan.do! do
destination = plan.connections[:mysql][:destination]
destination.exec!("./transformations/combined_name.sql", {name: 'user counter'})
end
Options & Notes
- Thanks to @rahilsondhi, @rgarver and Looksharp for all their help
- email_options is a hash consumed by the Pony mail gem
- Forklift's logger is Lumberjack with a wrapper to also echo the log lines to stdout and save them to an array to be accessed later by the email system.
- The mysql connections hash will be passed directly to a mysql2 connection.
- The elasticsearch connections hash will be passed directly to a elasticsearch connection.
- Your databases must exist. Forklift will not create them for you.
- Ensure your databases have the right encoding (eg utf8) or you will get errors like
#<Mysql2::Error: Incorrect string value: '\xEF\xBF\xBDFal...' for column 'YOURCOLUMN’ at row 1>
- If testing locally, mailcatcher (https://github.com/sj26/mailcatcher) is a helpful gem to test your email sending
Contributing and Testing
See: CONTRIBUTING
Alternatives
If you want something similar for Node.js try Empujar