Skip to content
Switch branches/tags


Failed to load latest commit information.
Latest commit message
Commit time


A boilerplate implementation of Luigi at Groupon


  • Luigi is a Python package that helps you build complex pipelines of batch jobs. It handles dependency resolution, workflow management, visualization, handling failures, command line integration, and much more

  • Luigi-Warehouse adds

  • example workflows (i.e. replicating postgresql tables to redshift)

  • more data sources

  • variable data sources that do not rely on default luigi behavior/configs (i.e. VariableS3Client)

Install / Setup

  • Install python3 - This repo has been tested against python 3.4+

python install

Developers - if you're wanting to modify/use the workflows with your custom logic
  • Clone this repo
  • pip3 install -r requirements.txt if you want full functionality of all data sources
  • mkdir your-path-to/data
  • Put your credentials and settings in luigi.cfg. luigi.cfg-example shows some possible options. You can also $ export LUIGI_CONFIG_PATH=/path/to/your/luigi.cfg && python...
  • You're ready to replicate or move data around...

Getting Started

  • Some example workflows are included. Assumptions, Args & Comments are in the File
File Description Main Class(es) replicates all data from a google sheet to a redshift table (full copy/replace) Run replicates all data from a google sheet to a hadoop hive table via spark (full copy/replace) main replicates postgres tables to redshift (incrementally or full copy/replace) Run - PerformIncrementalImport PerformFullImport spark app that replicates postgres tables to hadoop(hive) (incrementally or copy/replace) Run - RunIncremental RunFromScratch replicates a salesforce report or SOQL to a redshift table(full copy/replace) SOQLtoRedshift ReporttoRedshift replicates given teradata SQL to redshift table (incrementally or full copy/replace) Run replicates all data from typeform responses to a redshift table (full copy/replace) Run extracts users,orgs,tickets,ticket_events from zendesk to redshift (partially incremental) Run generic class to extract from zendesk API and load to hadoop hive via spark (incrementally or full copy/replace) ZendeskSpark
  • Example to start the luigi scheduler daemon
$ ./start_luigi_server.bash
  • Example to run a workflow with multiple workers in parallel
$ LUIGI_CONFIG_PATH=/path/to/your/luigi.cfg && python3 luigi_warehouse/ Run --params here --workers 50

Data Sources

Dependent python packages required & API reference

Luigi - Spotify/Luigi
Postgres / Redshift - psycopg2
MySQL - pymysql
Adwords - googleads : API Reference
Googlesheets - gspread : API Reference
Slack - slackclient : API Reference
Five9 - suds : API Reference
Twilio - twilio : API Reference
Livechat - API Reference
Zendesk - zdesk : API Reference
Shiftplanning - API Reference
Kochava - API Reference
Teradata - teradata
  • requires some configuring to install. We typically have to do
$ mv ~/.odbc.ini ~/.odbc.ini.orig 
$ cp /opt/teradata/client/15.10/odbc_64/odbcinst.ini ~/.odbcinst.ini 
$ cp /opt/teradata/client/15.10/odbc_64/odbc.ini ~/.odbc.ini
OnboardIQ - API Reference
AppBoy - API Reference
Salesforce - simple-salesforce : API Reference
  • Props to cghall for the capability to query salesforce reports directly using the analytics API

  • Also available are SalesforceBulk and SalesforceBulkJob classes which use the Salesforce bulk API

Braintree - braintree : API Reference
Typeform - API Reference
Checkr - API Reference
AWS - boto : boto3


  • We currently use slack or email for job status notifications which can easily be added

  • luigi-slack

from luigi_slack import SlackBot, notify
slack_channel = 'luigi-status-messages'

if __name__ == '__main__':
  slack_channel = 'luigi-status-messages'
  slacker = SlackBot(token=luigi.configuration.get_config().get('slackbots', 'BOWSER_SLACK_API_KEY'),
  with notify(slacker): 
import boto3

class Email:
  def __init__(self, region, aws_access_key_id, aws_secret_access_key):
    self.client = boto3.client('ses',region_name=region,aws_access_key_id=aws_access_key_id, aws_secret_access_key=aws_secret_access_key)

  def send(self, from_, to_list, subject, body):
    return self.client.send_email(Source=from_,
                                  Destination={'ToAddresses': to_list},
                                                     {'Data': subject},
                                                             {'Data': body},
                                                              {'Data':' '}

Data Validation

  • Targeted towards ensuring successful replication of data to Redshift (see modules/
  • if the same number of columns in the csv are in the target table
  • if the columns have the same datatypes in the same order (VARCHAR is acceptable for any python datatype)
    • uses python_redshift_dtypes to convert
  • Checks for load errors for the target:schema:table provided since the load_start provided timestamp
  • Use the wrapper class RunAnywayTarget if you want to make it easier as we make each validation scheme better

  • pass in the taskobj with the following attributes

    • type = ['LoadError', 'Structure']
    • target = Redshift
    • table =
    • schema =
    • local_file = local csv file path
    • load_start = when you started to copy the records from S3
  • doing RunAnywayTarget(self).done() will not do validation

  • doing RunAnywayTarget(self).validation() will do the validation and if successful also say we're done the task

  • Takes the following args
  1. target_cols : a list of columns ordered for how you want your dataframe to be structured
  2. df : your dataframe you want restructured
  • example: I my dataframe to have columns in this order ['one','two','three','four','five','six']
>>> from validation import OrderedDF
>>> import pandas as pd
>>> test = [[None,'',1,7,8],[None,'',2,5,6]]
>>> test = pd.DataFrame(test,columns=['one','two','four','five','three'])
>>> test
    one two  four  five  three
0  None         1     7      8
1  None         2     5      6
>>> result = OrderedDF(['one','two','three','four','five','six'],t)
>>> result.df
    one two  three  four  five   six
0  None          8     1     7  None
1  None          6     2     5  None
  • This class will fix tables for you
  1. Check for copy errors
  2. Handle the copy errors
  • Add column(s) if needed
  • Change dtype(s) if needed
  1. Get orig table's schema
  2. Craft new table's schema with changes from errors
  3. Make the change and retry the copy and remove duplicate * records
  4. While there are copy errors
  • handle the errors

  • attempt to fix

  • retry copy

  • remove duplicate * records

  • To run use

StructureDynamic(target_schema=  ,# redshift schema your table is in
                 target_table=    # your table
                      add_cols=  ,# True or False for if you want columns added in attempting to fix
                      change_dtypes=  ,# True or False if you want column data types changed in attempting to fix
                      copy=           ,# copy command you attempted
                      load_start=      # when you started the copy command, '%Y-%m-%d %H:%M:$S
  • Example usage:
    • sql prep: create the table
CREATE TABLE public.test(id INT, col VARCHAR);
INSERT INTO test VALUES (1,'2');
INSERT INTO test VALUES (2, 'two');
  • test.csv: create the csv you want to attempt to copy
  • we attempt to copy normally but we get load errors because one of the columns isn't right
COPY public.test FROM 's3://luigi-godata/test.csv' 
CREDENTIALS 'aws_access_key_id=XXXX;aws_secret_access_key=XXXX'
  • we run ValidationDynamic
from validation import StructureDynamic
copy = '''COPY public.test FROM 's3://luigi-godata/test.csv' 
          CREDENTIALS 'aws_access_key_id=XXXX;aws_secret_access_key=XXXX'
StructureDynamic(target_schema='public',target_table='test').run(add_cols=True,change_dtypes=True,copy=copy,load_start='2016-10-6 10:15:00')
  • our table is fixed and called public.test
  • our original table is kept as public.test_orig_backup
  • stdout lists the stl_load_errors
  • the changes made to the table's ddl is printed to stdout