A Python DB API and SQLAlchemy dialect for querying non-SQL resources like APIs

shillelagh

Shillelagh is a Python DB API and SQLAlchemy dialect for querying non-SQL resources like APIs and files. You can use it to write queries like this:

INSERT INTO "csv:///path/to/file.csv"
    SELECT time, chance_of_rain
    FROM "https://api.weatherapi.com/v1/history.json?key=XXX&q=London"
    WHERE strftime('%Y-%m-%d', time) IN (
      SELECT day
      FROM "https://docs.google.com/spreadsheets/d/1_rN3lm0R_bU3NemO0s9pbFkY5LQPcuy1pscv8ZXPtg8/edit#gid=2064361835"
    )

The query above reads holidays from a Google Sheet, uses the days to get weather data from WeatherAPI, and writes the change of rain at each hour of the holidays into a (pre-existing) CSV file.

Each of these resources is implemented via an adapter, and writing adapters is relatively straightforward.

Writing a new adapter

Let's say we want to fetch data from WeatherAPI using SQL. Their API is pretty straightforward — to fetch data for a given day in a given location all we need is an HTTP request:

https://api.weatherapi.com/v1/history.json?key=XXX&q=94158&dt=2020-01-01

This will return data for 2020-01-01 in the ZIP code 94158 as a JSON payload.

The response contains many different variables, but let's assume we're only interested in timestamp and temperature for the sake of this example. Of those two, timestamp is special because it can be used to filter data coming from the API, reducing the amount that needs to be downloaded.

We start by defining an "adapter" class, with the columns we're interested in:

from shillelagh.adapters.base import Adapter

class WeatherAPI(Adapter):

    ts = DateTime(filters=[Range], order=Order.ASCENDING, exact=False)
    temperature = Float()

The ts (timestamp) column has the type DateTime, and can be filtered with a desired range (for example, WHERE ts >= '2020-01-01' AND ts <= '2020-01-07'). We know that the values will be returned in ascending order by the API, so we annotate that to help the SQL engine. If a query has ORDER BY ts ASC we know that we don't need to sort the payload.

In addition, we declare that the results from filtering ts are not exact. This is because the API returns data for every hour of a given day. To make our lives easier we're going to filter the data down to the daily granularity, and let the SQL engine filter the rest. For example, imagine this query:

SELECT * FROM weatherapi WHERE ts > '2020-01-01T12:00:00' AND ts < '2020-01-02T12:00:00'

In this case, the adapter is going to download all data for the days 2020-01-01 and 2020-01-02, and pass them to the SQL engine to narrow it down to between noon in each day. We could do that filtering ourselves in the adapter, but since we're not discarding a lot of data it's ok.

For temperature we simply declare it as float, since we can't use temperature values to pre-filter data in the API.

Now we define our __init__ method, which initializes the adapter with the location and API key:

def __init__(self, location: str, api_key: str):
    self.location = location
    self.api_key = api_key

Finally, we define a method to download data from the API:

def get_data(self, bounds: Dict[str, Filter], order: List[Tuple[str, RequestedOrder]]) -> Iterator[Row]:
    ts_range: Range = bounds["ts"]
    today = date.today()
    start = ts_range.start.date() if ts_range.start else today - timedelta(days=7)
    end = ts_range.end.date() if ts_range.end else today

    while start <= end:
        url = (
            f"https://api.weatherapi.com/v1/history.json?key={self.api_key}"
            f"&q={self.location}&dt={start}"
        )
        response = requests.get(url)
        if response.ok:
            payload = response.json()
            hourly_data = payload["forecast"]["forecastday"][0]["hour"]
            for record in hourly_data:
                dt = dateutil.parser.parse(record["time"])
                yield {
                    "rowid": int(dt.timestamp()),
                    "ts": dt.isoformat(),
                    "temperature": record["temp_c"],
                }

        start += timedelta(days=1)

The important thing to know here is that since we defined ts as being filtered through a Range, a corresponding range will be passed to the get_data method specifying how ts should be filtered. The range has optional start and end values, which when not present are defaulted to 7 days ago and today, respectively.

Note also that the method yields rows as dictionaries. In addition to values for ts and temperature it also returns a row ID. This should be a unique value for each row.

We also need to define some dispatching methods, so our adapter can be found:

@staticmethod
def supports(uri: str) -> bool:
    """https://api.weatherapi.com/v1/history.json?key=XXX&q=94158"""
    parsed = urllib.parse.urlparse(uri)
    query_string = urllib.parse.parse_qs(parsed.query)
    return (
        parsed.netloc == "api.weatherapi.com"
        and parsed.path == "/v1/history.json"
        and "key" in query_string
        and "q" in query_string
    )

@staticmethod
def parse_uri(uri: str) -> Tuple[str, str]:
    parsed = urllib.parse.urlparse(uri)
    query_string = urllib.parse.parse_qs(parsed.query)
    location = query_string["q"][0]
    api_key = query_string["key"][0]

    return (location, api_key)

Now we can use our class to query the API using Sqlite:

from shillelagh.backends.apsw.db import connect

connection = connect(":memory:")
cursor = connection.cursor()

api_key = "XXX"
query = f"""
    SELECT *
    FROM "https://api.weatherapi.com/v1/history.json?key={api_key}&q=94923" AS bodega_bay
    WHERE ts >= '2020-01-01T12:00:00'
"""
for row in cursor.execute(query):
    print(row)

GitHub

https://github.com/betodealmeida/shillelagh