/ht-etl

Anidata 1.0: ETL and algorithm code.

Primary LanguagePython

Human Trafficking ETL

Build Status Coverage Status

This project will hold all the ETL code required to transform the raw HTML web pages into clean, normalized data that can be used for analysis.

Getting started

You will need the following installed:

  • Python 2.7.x (where x means any number) download here
    • Alternative: make a Python 2.7 environment with Anaconda - see example below.

Then click on the 'Fork' button above to make your own copy of the project, so that you run

git clone https://gitlab.com/yourUsername/ht-etl.git

where yourUsername is your actual user name (e.g. for Bryant Menn it would be bmenn, yours may vary). The SSH protocol currently does NOT work with git lfs with these instructions.

Do NOT run the following:

git clone https://gitlab.com/anidata/ht-etl.git

We are going to use what is called a fork-merge model for git.

Getting the raw data

A sample of the raw can be accessing by using the lfs plugin for git. Instructions to install the lfs plugin can be found here.

After installing the lfs plugin, setup the plugin by running

git lfs install

and get the file with

git lfs fetch

Module Installation

pip install -e .
  • Alternative: see instructions below for Windows 10 and Anaconda.

Hacking

Pick an issue off the issue list and get started! If you need help just ping the anidata1_1 slack channel for help.

When you're done hacking, run

git add .
git commit -m "An explanation of what you did goes here"
git push origin

And then open a merge request and make sure it's mentioned in the issue's comments.

Running ETL

ETL batch uses Luigi (http://luigi.readthedocs.io/en/stable/index.html) under the hood.

To configure Luigi, rename luigi.cfg.example to luigi.cfg and add the password to that file.

To run all the jobs excute:

luigi --module htetl.main_jobs LoadEntityIds --local-scheduler
  • Alternative: see instructions below for running email parser with Windows 10 and Anaconda.

Example: Getting ht-etl email parser to run on Windows 10 using Anaconda from scratch

Example is for running on local machine instead of Docker if you have trouble getting Docker to work

Install ht-etl

  • Install Anaconda
  • In Anaconda Navigator (help)
    • Make & activate a Python 2.7 virtual environment, call it something like python_27
    • Open terminal - the path on the left should end in your environment name like python_27 and python --version should say it's 2.7.
    • pip install -e C:\\your\\path\\to\\ht-etl

Create PostgreSQL database running on your local machine (i.e. not in a Docker container)

  • Download and unzip crawler_er.tar.gz from (https://github.com/anidata/ht-archive)

    • To unzip, first install 7-Zip, then right-click on the file and select 7-Zip.
    • You have to first unzip the .gz, then the .tar.
    • Result should be a crawler.sql file.

  • Install PostgreSQL

    • Save the password it makes you enter during installation process! We'll call this your_password below

  • In pgAdmin

    • Servers -> PostgreSQL 9.6 (top left in PGAdmin window)
    • Password: your_password
    • PostgreSQL 9.6 -> Databases -> Create -> Database, call it "crawler"

  • Open psql shell (it's under Windows Start menu -> PostgreSQL 9.6 -> SQL Shell (psql))

  • In psql shell

    • Server = localhost
    • Database = crawler (NOT postgres)
    • Port = 5432 (Default Postgres port is 5432 - you can see the server's port in pgAdmin)
    • Username = postgres
    • Password=your_password
    • Prompt should now be crawler=#
    • CREATE ROLE dbadmin WITH SUPERUSER LOGIN PASSWORD '1234';
    • \i 'C:/your/path/to/crawler.sql';
    • in pgAdmin you should see some tables as below

    pgAdmin_crawler

Run Luigi tasks

  • In a Python 2.7 terminal (open it from Anaconda Navigator as described above)

    • cd C:\your\path\to\ht-etl (otherwise it errors when it can't find things like data/flat_post.csv)

  • luigi --module htetl.main_jobs EmailsToPostgres --host localhost --database crawler --user postgres --password your_password --local-scheduler

  • If it worked, it should say something like this at the end

    execution_success

  • You should see two new files in your ht-etl/data folder: flat_post.csv and parsed_email.csv

  • You should see two new tables in the crawler database (in pgAdmin, right-click the server icon & "Refresh"): emailaddress and table_updates

A strange error I encountered sometimes was "ValueError: need more than 1 value to unpack", originating in lock.py. I don't know why, but fixed it by deleting luigi's .pid files in C:\Users\Lukas\AppData\Local\Temp\luigi

What just happened? (step-by-step for Luigi newbies)

  • See also Luigi documentation Follow along in main_jobs.py, get_data.py and util.py)
  • First, EmailsToPostgres is PENDING, but before it can run it requires ParseEmails to run and produce output.
    • Your database parameters (host, database, user, password) were passed into EmailsToPostgres via luigi.Parameter in its superclass LoadPostgres. From there they are passed to subsequent classes, as you can see in the code.
  • Now ParseEmails is PENDING, but before it can run it requires RawHTMLPostData to run and produce output.
  • RawHTMLPostData (which inherits from QueryPostgres) requires nothing (no requires() method), so it runs (its run() method). It gets some columns from the database and makes the 1st CSV, as defined by its output() method.
  • Now RawHTMLPostData is DONE, satisfying the requirement (i.e. the output file exists). Luigi can now run ParseEmails.
  • ParseEmails loads the 1st CSV (via self.input()), parses out some email addresses and saves it in the 2nd CSV through its run() method.
  • Now ParseEmails is DONE, satisfying the requirement (the output file exists). Luigi can now run EmailsToPostgres.
  • EmailsToPostgres inherits from LoadPostgres, which inherits from Luigi's helper class luigi.postgres.CopyToTable.
    • NB: that documentation is for Luigi 2.6 which moved it to luigi.contrib.postgres but I expect it works similarly.
  • luigi.postgres.CopyToTable invokes run(), which eventually calls rows() in LoadPostgres which loads the 2nd CSV file (via self.input()) and returns the appropriate generator (yielding a tuple for each row).
  • luigi.postgres.CopyToTable now writes each row into the Postgres database crawler. It knows which table, columns and datatypes to write because of table and columns variables in EmailsToPostgres.
  • The "output" of EmailsToPostgres is not a file but a PostgresTarget representing the inserted dataset (it keeps track using the Postgres table table_updates)

What if I want to re-run the pipeline to make a new set of output files and new Postgres table?

  • Luigi tasks won't execute a second time if the output target already exists. Therefore, you must manually delete the .csv files.
  • To re-run the last task, EmailsToPostgres, you should manually delete the Postgres tables emailaddress and table_updates
    • See comment in main_jobs.py for why table_updates has to be deleted too.
  • Once the files & tables are deleted you can re-run all 3 tasks using the same luigi command.
  • If you only want to run ParseEmails and EmailsToPostgres you just have to delete the CSV file and database tables created by those tasks. You can leave alone the CSV made by RawHTMLPostData. This prevents having to re-run time-consuming steps.
  • If you only want to run ParseEmails but not EmailsToPostgres (remember RawHTMLPostData will run first if its output does not exist): luigi --module htetl.main_jobs ParseEmails --host localhost --database crawler --user postgres --password your_password --local-scheduler