Skip to content

cwru-xlab/hat-py-sdk

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

hat-py-sdk

Unofficial Python SDK for the Dataswift API.

Features

  • Installation flags for minimal dependency overhead:
    • sync: synchronous client backed by the requests package
    • async: asynchronous client backed by the aiohttp package
    • sync-cache: HTTP caching backed by the requests-cache package
    • async-cache: HTTP caching backed by the aiohttp-client-cache package
    • orjson: fast JSON (de)serialization with the orjson package
    • ulid: use ULID identifiers instead of UUIDs
  • Authentication with owner tokens (via API or web auth) and application tokens
  • Automatic token refreshing and verification
  • Supports any keyring credential for API owner token authentication
  • All Direct Data API operations:
    • POST: groups records by endpoint to minimize request bandwidth
    • GET: supports single-endpoint requests and options
    • PUT: supports multi-record requests
    • DELETE: supports multi-record requests
  • Response streaming and caching to minimize latency and bandwidth
  • Meaningful exception types
  • Lazy token initialization for efficiency
  • Session-based requests
  • Powerful model validation and parsing for records and API tokens with pydantic
  • Encouraged immutability to avoid subtle bugs

Usage

Domain modeling

Top-level domain objects should inherit from HatModel, which is a special kind of pydantic BaseModel. All other domain objects can be either pydantic BaseModel instances or any other kind of JSON-serializable object. A HatModel has a record ID and an endpoint that relate to the Dataswift API. The record ID uniquely identifies the record in the PDA. The endpoint is a path that describes where in the PDA the record is located. In this way, a PDA is like an object database (e.g., Amazon S3) where all the objects are accessible via the endpoint at which they are located. A PDA is also like a document database in that each record stored at an endpoint can be arbitrary JSON.

Note: This SDK assumes that each endpoint contains homogenous data (i.e., all records have the same JSON schema). The HatModel allows for arbitrary fields, so it is possible to retrieve any JSON record from an endpoint, but fields that represent JSON objects will remain as Python dicts.

from pydantic import BaseModel
from hat.model import HatModel


class Nested(BaseModel):
    prop1: int
    prop2: str


class MyModel(HatModel):
    nested: Nested
    prop3: bytes

Creating a client

This SDK provides both synchronous and asynchronous HTTP support and are very similar. The two noticeable differences in their usage is that asynchronous class names have a prefix of "Async" and require the async/await syntax when using the client.

Synchronous

from keyring.credentials import SimpleCredential

from hat.client import HttpClient, HatClient, CredentialOwnerToken, AppToken

http_client = HttpClient()
credential = SimpleCredential("username", "password")
token = CredentialOwnerToken(http_client, credential)
token = AppToken(http_client, token, "application-id")
# Application namespace is only required for endpoint-specific requests.
client = HatClient(http_client, token, "namespace")

Asynchronous

from keyring.credentials import SimpleCredential

from hat.aioclient import (
    AsyncHttpClient, AsyncHatClient, AsyncCredentialOwnerToken, AsyncAppToken
)

http_client = AsyncHttpClient()
credential = SimpleCredential("username", "password")
token = AsyncCredentialOwnerToken(http_client, credential)
token = AsyncAppToken(http_client, token, "application-id")
# Application namespace is only required for endpoint-specific requests.
client = AsyncHatClient(http_client, token, "namespace")

CRUD API

Synchronous

from hat.client import HatClient
from hat.model import HatModel, GetOpts, Ordering

client = HatClient(...)

# GET requests accept objects with an endpoint attribute...
models: list[MyModel] = client.get(
    mtype=MyModel,
    endpoint=HatModel(endpoint="endpoint"),
    # GET request options are also validated using pydantic:
    options=GetOpts(order_by="id", ordering=Ordering.ASCENDING, skip=3, take=5))
# ...or just specify the endpoints.
models = client.get(mtype=MyModel, endpoint="endpoint", ...)

# Models are grouped by endpoint for efficient mixed-endpoint POST requests.
models: list[MyModel] = client.post(my_model, ...)

models: list[MyModel] = client.put(my_model, ...)

# Similar to GET requests, DELETE requests can specify an object...
client.delete(HatModel(record_id="record_id"), ...)
# ...or just the record IDs.
client.delete("record_id", ...)

Asynchronous

from hat.aioclient import AsyncHatClient
from hat.model import HatModel, GetOpts, Ordering

client = AsyncHatClient(...)

# GET requests accept objects with an endpoint attribute...
models: list[MyModel] = await client.get(
    mtype=MyModel,
    endpoint=HatModel(endpoint="endpoint"),
    # GET request options are also validated using pydantic:
    options=GetOpts(order_by="id", ordering=Ordering.ASCENDING, skip=3, take=5))
# ...or just specify the endpoints.
models = await client.get(mtype=MyModel, endpoint="endpoint", ...)

# Models are grouped by endpoint for efficient mixed-endpoint POST requests.
models: list[MyModel] = await client.post(my_model, ...)

models: list[MyModel] = await client.put(my_model, ...)

# Similar to GET requests, DELETE requests can specify an object...
await client.delete(HatModel(record_id="record_id"), ...)
# ...or just the record IDs.
await client.delete("record_id", ...)

Active-record API

This SDK also provides an alternative usage of the CRUD API with the active-record pattern. It provides a simpler interface and offers a more object-centric experience. Domain modeling is the same as before, except that the top-level object must inherit from either ActiveHatModel or AsyncActiveHatModel. The Create and Update operations are provided by a single save() operation.

Note: By its very nature, this API does not have the advantage of efficient bulk POST and PUT operations that the standard CRUD API offers. However, because this API is merely a thin wrapper around the CRUD API, it is easy to switch between them when most appropriate.

Synchronous

from hat import client

# Assign the client as a class attribute.
client.set_client(client.HatClient(...))


# Model your data using the special active-record pydantic model.
class MyModel(client.ActiveHatModel):
    value: int


# Retrieve models from their endpoint with automatic data binding from JSON.
model: MyModel = MyModel.get("endpoint")[0]
# Modify their attributes,...
model.value += 1
# ...easily persist the changes,...
model.save()
# ...or delete the model.
model.delete()

# It is also possible to delete multiple records...
MyModel.delete_all(HatModel(record_id="record_id"), ...)
# ...or just with the record IDs.
MyModel.delete_all("record_id", ...)

Asynchronous

from hat import aioclient

# Assign the client as a class attribute.
aioclient.set_async_client(aioclient.AsyncHatClient(...))


# Model your data using the special active-record pydantic model.
class MyModel(aioclient.AsyncActiveHatModel):
    value: int


# Retrieve models from their endpoint with automatic data binding from JSON.
model: MyModel = await MyModel.get("endpoint")[0]
# Modify their attributes,...
model.value += 1
# ...easily persist the changes,...
await model.save()
# ...or delete the model.
await model.delete()

# It is also possible to delete multiple records...
await MyModel.delete_all(HatModel(record_id="record_id"), ...)
# ...or just with the record IDs.
await MyModel.delete_all("record_id", ...)