Skip to content

Data modeling & verification for the City of Amsterdam's open data - based on JSON Schema

Notifications You must be signed in to change notification settings

Amsterdam/amsterdam-schema

Repository files navigation

Amsterdam logo

Amsterdam Schema

Amsterdam Schema aims to describe and validate open data published by the City of Amsterdam, in order to make the storing and publishing of different datasets more structured, simpler and better documented.

This repository contains:

  1. JSON documents that describe the structure and metadata of datasets (i.e.: dataset schemas not to be confused with JSON-schemas);
  2. JSON documents that describe the structure and metadata of tables (i.e.: table schemas not to be confused with JSON-schemas);
  3. A JSON-Schema metaschema to validate the documents mentioned under 1) and 2).

More specifically, metaschemas are JSON-Schemas that can make sure every dataset published by the City of Amsterdam always contains the right metadata and is of the right form.

This is done by running structural and semantic validation. The structural part is handled by the metaschema defined in this repository. The logic for semantic validation is defined in the schematools repository.

Amsterdam Schema Specification

Apart from the technical description an in-depth textual specification of the Amsterdam Schema can be found at https://schemas.data.amsterdam.nl/docs/ams-schema-spec.html.

The Amsterdam Schema is chosen to be delimited in such a way that it can interoperate with as many systems as possible. The results of this analysis can be found at the Grootst Gemene Deler page.

Amsterdam Schema Registry

Each instance of Amsterdam Schema exists of:

  1. Metadata about a single dataset;
  2. Metadata about each table in this single dataset;
  3. For each table, a table-schema to describe and validate the data in these tables.

An overview of the current schemas can be found at https://github.com/Amsterdam/amsterdam-schema/tree/master/datasets.

Concepts

In Amsterdam Schema, we're using the following concepts:

Type Description
Dataset A single dataset, with contents and metadata
Table A single table with objects of a single class/type
Row A row in such a table (a single object, a row in a source CSV file or feature in a source Shapefile, for example)
Field A property of a single object

For example:

  • The dataset bag contains data for each building and address in the city;
  • This dataset contains two tables: buildings and addresses;
  • To describe this dataset according to Amsterdam Schema, we first describe the metadata of the dataset (such as its identifier, title, description and DCAT fields) in a dataset.json file;
  • For each table in this dataset, we describe the table metadata in a separate JSON file. We can also choose to combine the dataset and table JSON data in a single JSON file;
  • For each table, we create a table-schema to describe its contents. This JSON Schema describes all the fields in a single table row, and the types of these fields;
  • Amsterdam Schema is used to validate the dataset and table JSON data
  • Amsterdam Schema is used to validate the table row JSON Schema, with a meta-schema (a JSON Schema to verify a JSON Schema).

Versioning

You can find all historical versions of the Amsterdam Schema definition in this repository. Version numbers are shown as '@1.0.0' where we follow SchemaVer for versioning. This will allow for a gradual evolution of capabilities.

See also

For more information, see (some of these pages are in Dutch):

Publishing

In order to publish the Amsterdam Schema to the object store install the Python package included in this repository:

% python3.8 -m venv venv
% pip install -U pip setuptools
% pip install '.[tests,dev]'

The extra options tests and dev are not strictly necessary for publishing, but are handy to have installed while working on the schema definitions. Once installed publishing could be as simple as running:

% publish

but it likely isn't.

See, the publish tool expects a number of environment variables to be set. These are:

DATAPUNT_ENVIRONMENT=[acceptance|production|...]  # default is 'acceptance'
OS_USERNAME=dataservices
OS_TENANT_NAME=...
OS_PASSWORD=...
OS_AUTH_URL=https://identity.stack.cloudvps.com/v2.0

Where the OS prefix stands for Object Store, and the ... for values that you should provide.

For development purposes, it can be convenient to publish schemas to an isolated development location on the objectstore. The schema:$ref attributes will be set correctly during the publishing process. This is essential for the validator in schema-tools to follow the references to the metaschema during validation.

This development location is a container on the dataservices objectstore.

To create a new container, the swift commandline client can be used (has been installed as part of python-swiftclient) that is a dependency.

Create new container with:

% swift post <schemas-yourname>  # example name, remove <>

Now make this location read-accessible over HTTP with:

swift post --read-acl ".r:*,.rlistings" <schemas-yourname>

Change the SCHEMA_BASE_URL environment variable to the http address of the container you just created.

SCHEMA_BASE_URL=https://<OS_TENANT_NAME>.objectstore.eu/<schemas-yourname>

The name of the objectstore container is constructed from 2 environment variables: $CONTAINER_PREFIX-$DATAPUNT_ENVIRONMENT

The default value for CONTAINER_PREFIX is schemas-.

Developing a new metaschema

In order to develop a new metaschema version locally and run structural and semantic validation against it:

Install the package from the repository root dir
0) pip install -e .[dev]

Create a new schema that we will develop

  1. cp -R schema@<latest-version> schema@<your-version>

Replace the internal references of the metaschema with the new version
2) sed -i s/<latest-version>/<your-version>/g schema@<your-version>/{,**/}*.json

Point the references in the new schema to the devserver
3) sed -i 's/https:\/\/schemas\.data\.amsterdam\.nl/http:\/\/localhost:8000/g' schema@<your-version>/{,**/}*.json

Generate the index expected by schematools
4) generate-index > datasets/index.json

Point the references in the dataset that we will use for development to the devserver
5) sed -i 's/https://schemas.data.amsterdam.nl/http://localhost:8000/g' datasets//{,**/}*.json

Start an nginx server with the source mounted and which rewrites URIs so
that it supports the URL structure expected by the schema references.
5) docker-compose up devserver

Validate a dataset
6) schema validate --schema-url='http://localhost:8000/datasets' <some-dataset> 'http://localhost:8000/schema@<your-version>'

And of course; after the metaschema is finished, the references in the new metaschema and the dataset used for development need to be be reset to the online URL.