An elegant, opinionated framework for deploying BrightHive Data Resources with zero coding.
BrightHive is in the business of building Data Trusts, which are legal, technical, and governance frameworks that enable networks of organizations to securely and responsibly share and collaborate with data, generating new insights and increasing their combined impact.
Data Resources are a core element of Data Trusts and may be loosely defined as data that is owned by or stewarded over by members of Data Trusts.
From the technical perspective, a BrightHive Data Resource is an entity comprised of the following elements:
- Data Model - The data model consists of one or more database tables and associated Object Relational Mapping (ORM) objects for communicating with these tables.
- RESTful API - Data managed by Data Resources are accessed via RESTful API endpoints. These endpoints support standard operations (i.e. GET, POST, PUT, PATCH, DELETE).
- Data Resource Schemas - Data Resource Schemas are JSON documents that define the data model, API endpoints, and security constraints placed on the specific Data Resource. These schemas are what allow for new Data Resources to be created without new code being written.
Spinning up a fresh, out-of-the-box Data Resource API requires minimal setup. Follow these steps:
1. Download install docker-cli and docker-compose based on your machine and preferences (e.g., you may choose to run docker-compose
in a virtualenv).
2. Clone or download the project to your local machine
git clone [email protected]:brighthive/data-resource-api.git
3. Build the Docker image with specified tag
# run this command in the root of the data-resource-api repo
docker build -t brighthive/data-resource-api:1.0.0-alpha .
4. Launch Docker containers
# run this command in the root of the data-resource-api repo
docker-compose up
Alternatively, you can run the data resource api in the background:
docker-compose up -d
Visit http://0.0.0.0:8000/programs
to test the API URL (though expect an "Access Denied" message).
Note! The default docker-compose.yml
launches three containers: postgres
, data-model-manager
, and data-resource-api
. Expect the postgres container to raise an error, at first, something like: ERROR: relation "checksums" does not exist at character 171
. This occurs temporarily, as Docker spins up the data-model-manager
(i.e., the service that creates and runs migrations). Once the model manager has done its business, the data-resource-manager
can ease-fully listen for incoming schema files. A successful launch should continuously log the following happy messages:
data-resource-api_1 | 2019-06-21 21:31:21,392 - data-resource-manager - INFO - Data Resource Manager Running...
data-resource-api_1 | 2019-06-21 21:31:21,393 - data-resource-manager - INFO - Checking data resources...
data-resource-api_1 | 2019-06-21 21:31:21,413 - data-resource-manager - INFO - Completed check of data resources
data-resource-api_1 | 2019-06-21 21:31:21,414 - data-resource-manager - INFO - Data Resource Manager Sleeping for 60 seconds...
The Data Resource API auto-magically instantiates data stores and corresponding endpoints without any coding.
Each JSON file in the schema
directory delineates a resource, namely: its RESTful paths ("api"), and its data model ("datastore"). Create a new resource by adding another JSON file to the schema
directory. (Visit the schema
directory for example JSON files.)
A resource JSON blob defines RESTful methods. These can be enabled or disabled.
{
"api": {
"resource": "name_of_endpoint",
"methods": [
{
"get": {
// enable the method
"enabled": true,
"secured": true,
"grants": ["get:users"]
},
"post": {
// disable the method
"enabled": false,
"secured": true,
"grants": ["get:users"]
},
...
The Data Resource API makes use of BrightHive authlib for adding authentication to endpoints. Authentication can be toggled on or off – on a per method basis.
{
"api": {
"resource": "name_of_endpoint",
"methods": [
{
"get": {
"enabled": true,
// do not require authentication
"secured": false,
"grants": ["get:users"]
},
"post": {
"enabled": false,
// require authentication
"secured": true,
"grants": ["get:users"]
},
...
The Data Resource API utilizes the Table Schema from fictionless data. The Table Schema is represented by a "descriptor", or a JSON object with particular attributes. In a Data Resource schema, the descriptor occupies the value of "datastore" >> "schema". A schema can have up to four properties, among them: primaryKey
, foreignKeys
, and fields
.
The fields
must be an array of JSON objects, and each object must define a field on the data model. A field definition should include, at minimum, a name
, type
(defaults to String), and required
(defaults to false
).
Frictionless data and, correspondingly, the Data Resource API can be particular about what goes inside a field descriptor. TABLESCHEMA_TO_SQLALCHEMY_TYPES
defines the available types in the frictionless schema and the corresponding types in sqlalchemy. Stick to these types, or you can anticipate unexpected behavior in your API! (See data-resource-api/data_resource_api/factories/table_schema_types.py
for more context.)
TABLESCHEMA_TO_SQLALCHEMY_TYPES = {
'string': String,
'number': Float,
'integer': Integer,
'boolean': Boolean,
'object': String,
'array': String, # Not supported yet
'date': Date,
'time': DateTime, # Not supported yet
'datetime': DateTime,
'year': Integer,
'yearmonth': Integer, # Not supported yet
'duration': Integer,
'geopoint': String,
'geojson': String, # Not supported yet
'any': String
}
Creating a foreign key involves making two additions to the JSON.
First, add the foreign key as a field in the schema
:
"datastore": {
"tablename": "programs",
"restricted_fields": [],
"schema": {
"fields": [
{
"name": "location_id",
"title": "Provider ID",
"type": "integer",
"description": "Foreign key for provider",
"required": false
},
...
Second, define the foreign key in the foreignKeys
array:
"datastore": {
"tablename": "programs",
"restricted_fields": [],
"schema": {
"fields": [
// Descriptor for location_id field, plus other fields
...
],
"foreignKeys": [
{
"fields": ["location_id"],
"reference": {
"resource": "locations",
"fields": ["id"]
}
}
]
}
}
Sometimes, a data resource has fields that contain important information (e.g., a source URL, a UUID) that should not be visible in the API. Hide these fields by adding the field name
to the list of restricted_fields
.
"datastore": {
"tablename": "programs",
"restricted_fields" : ["location_id"],
...
To create a many to many resource add the relationship to the API section.
{
"api": {
"resource": "programs",
"methods": [
{
"custom": [
{
"resource": "/programs/credentials"
}
]
}
]
},
...
This will generate a many to many relationship.
To add a resource you would POST and in the body include the child field as a parameter. ]
To query the relationship you have to go to /programs/1/credentials
. The relationship will not currently show up if you simply query /programs/1
.
,4
To replace the relationship perform a PUT
to /programs/1/credentials
with the full list of primary keys.
{
"credentials": [2,3]
}
To append a primary key to the relationship list perform a PATCH
. If you currently have a list of "credentials": [1]
and you perform PATCH
with "credentials": [2,3]
it will return "credentials":[1,2,3]
You can perform a DELETE
that will remove the given items from the list. If you wish to remove all items perform a PUT
with an empty list.
Given that you have a list of relationships, [1,2,3,4,5]
:
Performing a DELETE
with
{
"credentials": [2]
}
Results in [1,3,4,5]
.
You can also give more than one item. Given that you have a list of relationships, [1,2,3,4,5]
and we perform the following with DELETE
{
"credentials": [1,3,5]
}
Results in [2,4]
.
The following parameters can be adjusted to serve testing, development, or particular deployment needs.
RELATIVE_PATH
ABSOLUTE_PATH
ROOT_PATH
MIGRATION_HOME
DATA_RESOURCE_SLEEP_INTERVAL
DATA_MODEL_SLEEP_INTERVAL
SQLALCHEMY_TRACK_MODIFICATIONS
PROPAGATE_EXCEPTIONS
POSTGRES_USER
POSTGRES_PASSWORD
POSTGRES_DATABASE
POSTGRES_HOSTNAME
POSTGRES_PORT
SQLALCHEMY_DATABASE_URI
OAUTH2_PROVIDER
OAUTH2_URL
OAUTH2_JWKS_URL
OAUTH2_AUDIENCE
OAUTH2_ALGORITHMS
SECRET_MANAGER
=======
A few tests within the test suite currently require a docker image to be spun up in order to pass the tests. In order for the tests to pass you simply need to have docker running.
For developers to run a test use the following,
- First install the requirements
pipenv install --dev
- Run the tests with the following command,
pipenv run pytest
To leave the database up after running,
DR_LEAVE_DB=true pipenv run pytest
If you have trouble starting the application with docker-compose up
the problem may lie in your postgres container.
Try running docker rm postgres-container-id
to remove any prevoiusly saved data in the postgres container. This will allow the application to rebuild the database from scratch and should start successfully.
Or try running docker system prune
.