/gnaf-loader

A quick way to get started with PSMA's open GNAF & Admin Boundaries

Primary LanguagePythonApache License 2.0Apache-2.0

Change Log - November 2016 Release

  • Logging is now written to load-gnaf.log in your local repo directory as well as to the console
  • Added --psma-version to the parameters. Represents the PSMA version number in YYYYMM format and is used to add a suffix to the default schema names. Defaults to current year and latest release month. e.g. 201611. Valid values are <year>02 <year>05 <year>08 <year>11, and is based on the PSMA quarterly release months
  • All default schema names are now suffixed with --psma-version to avoid clashes with previous versions. e.g. gnaf_201611
  • Postgres 9.6 dump files for the November 2016 PSMA release are available
  • load-gnaf.py now works with Python 2.7 and Python 3.5
  • load-gnaf.py has been successfully tested on Postgres 9.6 and PostGIS 2.3
    • Note: Limited performance testing on Postgres 9.6 has shown setting the maximum number of parallel processes --max-processes to 3 is the most efficient value on non-SSD machines
  • Final row counts are stored in a new 'qa' table in the gnaf and admin_bdys schemas for checking the results
  • Code has been refactored to simplify it a bit and move some common functions to a new psma.py file

gnaf-loader

A quick way to load the complete Geocoded National Address File of Australia (GNAF) and Australian Administrative Boundaries into Postgres, simplified and ready to use as reference data for geocoding, analysis, visualisation and aggregation.

What's GNAF?

Have a look at these intro slides (PDF), as well as the data.gov.au page.

There are 3 options for loading the data

  1. Run the load-gnaf Python script and build the database in a single step
  2. Build the database in a docker environment
  3. Download the GNAF and/or Admin Bdys Postgres dump files & restore them in your database

Option 1 - Run load.gnaf.py

Running the Python script takes 30-120 minutes on a Postgres server configured for performance.

My benchmarks are:

  • 3 year old, 32 core Windows server with SSDs = ~45 mins
  • MacBook Pro = ~75 mins
  • 3 year old, 8 core commodity PC = ~75 mins.

Performance

To get a good load time you'll need to configure your Postgres server for performance. There's a good guide here, noting it's a few years old and some of the memory parameters can be beefed up if you have the RAM.

Pre-requisites

  • Postgres 9.3+ with PostGIS 2.2 (tested on 9.3, 9.4 & 9.5 on Windows and 9.5, 9.6 on macOS)
  • Add the Postgres bin directory to your system PATH
  • Python 2.7 or Python 3.5 with Psycopg2 2.6

Process

  1. Download PSMA GNAF from data.gov.au
  2. Download PSMA Administrative Boundaries from data.gov.au (download the ESRI Shapefile version)
  3. Unzip GNAF to a directory on your Postgres server
  4. Alter security on the directory to grant Postgres read access
  5. Unzip Admin Bdys to a local directory
  6. Create the target database (if required)
  7. Check the available and required arguments by running load-gnaf.py with the -h argument (see command line examples below)
  8. Run the script, come back in 30-120 minutes and enjoy!

Command Line Options

The behaviour of gnaf-loader can be controlled by specifying various command line options to the script. Supported arguments are:

Required Arguments

  • --gnaf-tables-path specifies the path to the extracted source GNAF tables (eg *.psv files). This should match the extracted directory which contains the subfolders Authority Code and Standard. This directory must be accessible by the Postgres server, and the corresponding local path for the server to this directory may need to be set via the local-server-dir argument
  • --local-server-dir specifies the local path on the Postgres server corresponding to gnaf-tables-path. If the server is running locally this argument can be omitted.
  • --admin-bdys-path specifies the path to the extracted source admin boundary files. This path should contain a subfolder named Administrative Boundaries. Unlike gnaf-tables-path, this path does not necessarily have to be accessible to the remote Postgres server.

Postgres Parameters

  • --pghost the host name for the Postgres server. This defaults to the PGHOST environment variable if set, otherwise defaults to localhost.
  • --pgport the port number for the Postgres server. This defaults to the PGPORT environment variable if set, otherwise 5432.
  • --pgdb the database name for Postgres server. This defaults to the PGDATABASE environment variable if set, otherwise psma_201602.
  • --pguser the username for accessing the Postgres server. This defaults to the PGUSER environment variable if set, otherwise postgres.
  • --pgpassword password for accessing the Postgres server. This defaults to the PGPASSWORD environment variable if set, otherwise password.

Optional Arguments

  • --psma-version PSMA version number in YYYYMM format. Defaults to current year and last release month. e.g. 201611.
  • --raw-gnaf-schema schema name to store raw GNAF tables in. Defaults to raw_gnaf_<psma_version>.
  • --raw-admin-schema schema name to store raw admin boundary tables in. Defaults to raw_admin_bdys_<psma_version>.
  • --gnaf-schema destination schema name to store final GNAF tables in. Defaults to gnaf_<psma_version>.
  • --admin-schema destination schema name to store final admin boundary tables in. Defaults to admin_bdys_<psma_version>.
  • --states space separated list of states to load, eg --states VIC TAS. Defaults to loading all states.
  • --prevacuum forces the database to be vacuumed after dropping tables. Defaults to off, and specifying this option will slow the import process.
  • --raw-fk creates both primary & foreign keys for the raw GNAF tables. Defaults to off, and will slow the import process if specified. Use this option if you intend to utilise the raw GNAF tables as anything more then a temporary import step. Note that the final processed tables will always have appropriate primary and foreign keys set.
  • --raw-unlogged creates unlogged raw GNAF tables, speeding up the import. Defaults to off. Only specify this option if you don't care about the raw data tables after the import - they will be lost if the server crashes!
  • --max-processes specifies the maximum number of parallel processes to use for the data load. Set this to the number of cores on the Postgres server minus 2, but limit to 12 if 16+ cores - there is minimal benefit beyond 12. Defaults to 6.
  • --boundary-tag tags all addresses with some of the key admin boundary IDs for creating aggregates and choropleth maps.

Example Command Line Arguments

  • Local Postgres server: python load-gnaf.py --gnaf-tables-path="C:\temp\psma_201602\G-NAF" --admin-bdys-path="C:\temp\psma_201602\Administrative Boundaries" Loads the GNAF tables to a Postgres server running locally. GNAF archives have been extracted to the folder C:\temp\psma_201602\G-NAF, and admin boundaries have been extracted to the C:\temp\psma_201602\Administrative Boundaries folder.
  • Remote Postgres server: python load-gnaf.py --gnaf-tables-path="\\svr\shared\gnaf" --local-server-dir="f:\shared\gnaf" --admin-bdys-path="c:\temp\unzipped\AdminBounds_ESRI" Loads the GNAF tables which have been extracted to the shared folder \\svr\shared\gnaf. This shared folder corresponds to the local f:\shared\gnaf folder on the Postgres server. Admin boundaries have been extracted to the c:\temp\unzipped\AdminBounds_ESRI folder.
  • Loading only selected states: python load-gnaf.py --states VIC TAS NT ... Loads only the data for Victoria, Tasmania and Northern Territory

Advanced

You can load the Admin Boundaries without GNAF. To do this: comment out steps 1, 3 & 4 in def main.

Note: you can't load GNAF without the Admin Bdys due to dependencies required to split Melbourne and to fix non-boundary locality_pids on addresses.

Attribution

When using the resulting data from this process - you will need to adhere to the attribution requirements on the data.gov.au pages for GNAF and the Admin Bdys, as part of the open data licensing requirements.

WARNING:

  • The scripts will DROP ALL TABLES and recreate them using CASCADE; meaning you'll LOSE YOUR VIEWS if you have created any! If you want to keep the existing data - you'll need to change the schema names in the script or use a different database
  • All raw GNAF tables can be created UNLOGGED to speed up the data load. This will make them UNRECOVERABLE if your database is corrupted. You can run these scripts again to recreate them. If you think this sounds ok - set the unlogged_tables flag to True for a slightly faster load
  • Boundary tagging (on by default) will add 15-60 minutes to the process if you have PostGIS 2.2. If you have PostGIS 2.1 or lower - it can take HOURS as the boundary tables can't be optimised!

IMPORTANT:

  • Whilst you can choose which 4 schemas to load the data into, I haven't QA'd every permutation. Stick with the defaults if you have limited Postgres experience
  • If you're not running the Python script on the Postgres server, you'll need to have access to a network path to the GNAF files on the database server (to create the list of files to process). The alternative is to have a local copy of the raw files
  • The 'create tables' sql script will add the PostGIS extension to the database in the public schema, you don't need to add it to your database
  • There is an option to VACUUM the database at the start after dropping the existing GNAF/Admin Bdy tables - this doesn't really do anything outside of repeated testing. (I was too lazy to take it out of the code as it meant renumbering all the SQL files and I'd like to go to bed now)

Option 2 - Build the database in a docker environment

Create a Docker container with GNAF and the Admin Bdys ready to go, so they can be deployed anywhere.

Process

  1. Download PSMA GNAF from data.gov.au
  2. Download PSMA Administrative Boundaries from data.gov.au (download the ESRI Shapefile version)
  3. Unzip GNAF and the Admin Bdys in the data/ directory of this repository
  4. Run docker-compose: docker-compose up. The database will be built.
  5. Use the constructed database as you wish.

Option 3 - Load PG_DUMP Files

Download Postgres dump files and restore them in your database.

Should take 15-60 minutes.

Pre-requisites

Process

  1. Download gnaf-201611.dmp (~1.6Gb)
  2. Download admin-bdys-201611.dmp (~2.0Gb)
  3. Edit the restore-gnaf-admin-bdys.bat or .sh script in the supporting-files folder for your database parameters and for the location of pg_restore
  4. Run the script, come back in 15-60 minutes and enjoy!

Data Licenses

Incorporates or developed using G-NAF ©PSMA Australia Limited licensed by the Commonwealth of Australia under the Open Geo-coded National Address File (G-NAF) End User Licence Agreement.

Incorporates or developed using Administrative Boundaries ©PSMA Australia Limited licensed by the Commonwealth of Australia under Creative Commons Attribution 4.0 International licence (CC BY 4.0).

DATA CUSTOMISATION

GNAF and the Admin Bdys have been customised to remove some of the known, minor limitations with the data. The most notable are:

  • All addresses link to a gazetted locality that has a boundary. Those small number of addresses that don't in raw GNAF have had their locality_pid changed to a gazetted equivalent
  • Localities have had address and street counts added to them
  • Suburb-Locality bdys have been flattened into a single continuous layer of localities - South Australian Hundreds have been removed and ACT districts have been added where there are no gazetted localities
  • The Melbourne, VIC locality has been split into Melbourne, 3000 and Melbourne 3004 localities (the new locality PIDs are VIC 1634_1 & VIC 1634_2). The split occurs at the Yarra River (based on the postcodes in the Melbourne addresses)
  • A postcode boundaries layer has been created using the postcodes in the address tables. Whilst this closely emulates the official PSMA postcode boundaries, there are several hundred addresses that are in the wrong postcode bdy. Do not treat this data as authoritative

KNOWN ISSUES

  • If you only want to load certain states - you may get an error in Part 2 stating a table doesn't exist - this can be ignored. e.g. loading ACT only will return an error saying "local_government_wards" does not exist.

TO DO

  • Output reference tables to PSV & SHP
  • List final record counts
  • Build QA into the Python script
  • Script the creation of pg_dump files in Python
  • Script the copying of pg_dump, PSV & SHP file to Amazon S3, in Python