Initial read_gbq support

.github/workflows/pre-commit.yml

@@ -0,0 +1,16 @@
+name: Linting
+
+on:
+  push:
+    branches: main
+  pull_request:
+    branches: main
+
+jobs:
+  checks:
+    name: "pre-commit hooks"
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v2
+      - uses: actions/setup-python@v2
+      - uses: pre-commit/[email protected]

.github/workflows/tests.yml

@@ -0,0 +1,54 @@
+name: Tests
+
+on: push
+
+# When this workflow is queued, automatically cancel any previous running
+# or pending jobs from the same branch
+concurrency:
+  group: ${{ github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  test:
+    runs-on: ${{ matrix.os }}
+    defaults:
+      run:
+        shell: bash -l {0}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: ["windows-latest", "ubuntu-latest", "macos-latest"]
+        python-version: ["3.7", "3.8", "3.9"]
+
+    steps:
+      - name: Checkout source
+        uses: actions/checkout@v2
+        with:
+          fetch-depth: 0  # Needed by codecov.io
+
+      - name: Setup Conda Environment
+        uses: conda-incubator/setup-miniconda@v2
+        with:
+          miniforge-variant: Mambaforge
+          miniforge-version: latest
+          use-mamba: true
+          channel-priority: strict
+          python-version: ${{ matrix.python-version }}
+          environment-file: ci/environment-${{ matrix.python-version }}.yaml
+          activate-environment: test-environment
+          auto-activate-base: false
+
+      - name: Install dask-bigquery
+        run: python -m pip install --no-deps -e .
+
+      - name: Set up Cloud SDK
+        uses: google-github-actions/setup-gcloud@master
+        with:
+          project_id: ${{ secrets.GCP_PROJECT_ID }}
+          service_account_key: ${{ secrets.GCP_SA_KEY }}
+          export_default_credentials: true
+
+      - name: Run tests
+        env:
+          DASK_BIGQUERY_PROJECT_ID: "${{ secrets.GCP_PROJECT_ID }}"
+        run: pytest -v dask_bigquery

.pre-commit-config.yaml

@@ -0,0 +1,17 @@
+repos:
+-   repo: https://github.com/psf/black
+    rev: 20.8b1
+    hooks:
+    - id: black
+      language_version: python3
+      exclude: versioneer.py
+-   repo: https://gitlab.com/pycqa/flake8
+    rev: 3.8.3
+    hooks:
+    - id: flake8
+      language_version: python3
+-   repo: https://github.com/pycqa/isort
+    rev: 5.8.0
+    hooks:
+    - id: isort
+      language_version: python3

ci/environment-3.7.yaml

@@ -0,0 +1,14 @@
+name: test-environment
+channels:
+  - conda-forge
+dependencies:
+  - python=3.7
+  - dask
+  - distributed
+  - pandas
+  - pyarrow
+  - pytest
+  - grpcio
+  - pandas-gbq
+  - google-cloud-bigquery
+  - google-cloud-bigquery-storage

ci/environment-3.8.yaml

@@ -0,0 +1,14 @@
+name: test-environment
+channels:
+  - conda-forge
+dependencies:
+  - python=3.8
+  - dask
+  - distributed
+  - pandas
+  - pyarrow
+  - pytest
+  - grpcio
+  - pandas-gbq
+  - google-cloud-bigquery
+  - google-cloud-bigquery-storage

ci/environment-3.9.yaml

@@ -0,0 +1,14 @@
+name: test-environment
+channels:
+  - conda-forge
+dependencies:
+  - python=3.9
+  - dask
+  - distributed
+  - pandas
+  - pyarrow
+  - pytest
+  - grpcio
+  - pandas-gbq
+  - google-cloud-bigquery
+  - google-cloud-bigquery-storage

dask_bigquery/__init__.py

@@ -0,0 +1 @@
+from .core import read_gbq

dask_bigquery/core.py

@@ -0,0 +1,155 @@
+from __future__ import annotations
+
+from contextlib import contextmanager
+from functools import partial
+
+import pandas as pd
+import pyarrow
+from dask.base import tokenize
+from dask.dataframe.core import new_dd_object
+from dask.highlevelgraph import HighLevelGraph
+from dask.layers import DataFrameIOLayer
+from google.cloud import bigquery, bigquery_storage
+
+
+@contextmanager
+def bigquery_client(project_id=None):
+    """This context manager is a temporary solution until there is an
+    upstream solution to handle this.
+    See  googleapis/google-cloud-python#9457
+    and googleapis/gapic-generator-python#575 for reference.
+    """
+    with bigquery.Client(project_id) as bq_client:
+        bq_storage_client = bigquery_storage.BigQueryReadClient(
+            credentials=bq_client._credentials
+        )
+        yield bq_client, bq_storage_client
+        bq_storage_client.transport.grpc_channel.close()
+
+
+def _stream_to_dfs(bqs_client, stream_name, schema, read_kwargs):
+    """Given a Storage API client and a stream name, yield all dataframes."""
+    return [
+        pyarrow.ipc.read_record_batch(
+            pyarrow.py_buffer(message.arrow_record_batch.serialized_record_batch),
+            schema,
+        ).to_pandas()
+        for message in bqs_client.read_rows(name=stream_name, offset=0, **read_kwargs)
+    ]
+
+
+def bigquery_read(
+    make_create_read_session_request: callable,
+    project_id: str,
+    read_kwargs: dict,
+    stream_name: str,
+) -> pd.DataFrame:
+    """Read a single batch of rows via BQ Storage API, in Arrow binary format.
+
+    Parameters
+    ----------
+    create_read_session_request: callable
+      kwargs to pass to `bqs_client.create_read_session` as `request`
+    project_id: str
+      Name of the BigQuery project.
+    read_kwargs: dict
+      kwargs to pass to read_rows()
+    stream_name: str
+      BigQuery Storage API Stream "name"
+      NOTE: Please set if reading from Storage API without any `row_restriction`.
+            https://cloud.google.com/bigquery/docs/reference/storage/rpc/google.cloud.bigquery.storage.v1beta1#stream
+    """
+    with bigquery_client(project_id) as (bq_client, bqs_client):
+        session = bqs_client.create_read_session(make_create_read_session_request())
+        schema = pyarrow.ipc.read_schema(
+            pyarrow.py_buffer(session.arrow_schema.serialized_schema)
+        )
+        shards = _stream_to_dfs(bqs_client, stream_name, schema, read_kwargs)
+        # NOTE: BQ Storage API can return empty streams
+        if len(shards) == 0:
+            shards = [schema.empty_table().to_pandas()]
+
+    return pd.concat(shards)
+
+
+def read_gbq(
+    project_id: str,
+    dataset_id: str,
+    table_id: str,
+    row_filter="",
+    read_kwargs: dict = None,
+):
+    """Read table as dask dataframe using BigQuery Storage API via Arrow format.
+    Partitions will be approximately balanced according to BigQuery stream allocation logic.
+
+    Parameters
+    ----------
+    project_id: str
+      Name of the BigQuery project.
+    dataset_id: str
+      BigQuery dataset within project
+    table_id: str
+      BigQuery table within dataset
+    row_filter: str
+      SQL text filtering statement to pass to `row_restriction`
+    read_kwargs: dict
+      kwargs to pass to read_rows()
+
+    Returns
+    -------
+        Dask DataFrame
+    """
+    read_kwargs = read_kwargs or {}
+    with bigquery_client(project_id) as (bq_client, bqs_client):
+        table_ref = bq_client.get_table(f"{dataset_id}.{table_id}")
+        if table_ref.table_type == "VIEW":
+            raise TypeError("Table type VIEW not supported")
+
+        def make_create_read_session_request(row_filter=""):
+            return bigquery_storage.types.CreateReadSessionRequest(
+                max_stream_count=100,  # 0 -> use as many streams as BQ Storage will provide
+                parent=f"projects/{project_id}",
+                read_session=bigquery_storage.types.ReadSession(
+                    data_format=bigquery_storage.types.DataFormat.ARROW,
+                    read_options=bigquery_storage.types.ReadSession.TableReadOptions(
+                        row_restriction=row_filter,
+                    ),
+                    table=table_ref.to_bqstorage(),
+                ),
+            )
+
+        # Create a read session in order to detect the schema.
+        # Read sessions are light weight and will be auto-deleted after 24 hours.
+        session = bqs_client.create_read_session(
+            make_create_read_session_request(row_filter=row_filter)
+        )
+        schema = pyarrow.ipc.read_schema(
+            pyarrow.py_buffer(session.arrow_schema.serialized_schema)
+        )
+        meta = schema.empty_table().to_pandas()
+
+        label = "read-gbq-"
+        output_name = label + tokenize(
+            project_id,
+            dataset_id,
+            table_id,
+            row_filter,
+            read_kwargs,
+        )
+
+        layer = DataFrameIOLayer(
+            output_name,
+            meta.columns,
+            [stream.name for stream in session.streams],
+            partial(
+                bigquery_read,
+                make_create_read_session_request,
+                project_id,
+                read_kwargs,
+            ),
+            label=label,
+        )
+        divisions = tuple([None] * (len(session.streams) + 1))
+
+        graph = HighLevelGraph({output_name: layer}, {output_name: set()})
+        return new_dd_object(graph, output_name, meta, divisions)

dask_bigquery/tests/test_core.py

@@ -0,0 +1,89 @@
+import os
+import random
+import uuid
+
+import pandas as pd
+import pytest
+from dask.dataframe.utils import assert_eq
+from distributed.utils_test import cluster_fixture  # noqa: F401
+from distributed.utils_test import client, loop  # noqa: F401
+from google.cloud import bigquery
+
+from dask_bigquery import read_gbq
+
+
+@pytest.fixture
+def df():
+    records = [
+        {
+            "name": random.choice(["fred", "wilma", "barney", "betty"]),
+            "number": random.randint(0, 100),
+            "idx": i,
+        }
+        for i in range(10)
+    ]
+
+    yield pd.DataFrame(records)
+
+
+@pytest.fixture
+def dataset(df):
+    "Push some data to BigQuery using pandas gbq"
+    project_id = os.environ.get("DASK_BIGQUERY_PROJECT_ID", "dask-bigquery")
+    dataset_id = uuid.uuid4().hex
+    table_id = "table_test"
+    # push data to gbq
+    pd.DataFrame.to_gbq(
+        df,
+        destination_table=f"{dataset_id}.{table_id}",
+        project_id=project_id,
+        chunksize=5,
+        if_exists="append",
+    )
+    yield (project_id, dataset_id, table_id)
+
+    with bigquery.Client() as bq_client:
+        bq_client.delete_dataset(
+            dataset=f"{project_id}.{dataset_id}",
+            delete_contents=True,
+        )
+
+
+# test simple read
+def test_read_gbq(df, dataset, client):
+    """Test simple read of data pushed to BigQuery using pandas-gbq"""
+    project_id, dataset_id, table_id = dataset
+    ddf = read_gbq(project_id=project_id, dataset_id=dataset_id, table_id=table_id)
+
+    assert list(ddf.columns) == ["name", "number", "idx"]
+    assert ddf.npartitions == 2
+    assert assert_eq(ddf.set_index("idx"), df.set_index("idx"))
+
+
+def test_read_row_filter(df, dataset, client):
+    "Test read data with a row restriction providing `row_filter`"
+    project_id, dataset_id, table_id = dataset
+    ddf = read_gbq(
+        project_id=project_id,
+        dataset_id=dataset_id,
+        table_id=table_id,
+        row_filter="idx < 5",
+    )
+
+    assert list(ddf.columns) == ["name", "number", "idx"]
+    assert ddf.npartitions == 2
+    assert assert_eq(ddf.set_index("idx").loc[:4], df.set_index("idx").loc[:4])
+
+
+def test_read_kwargs(df, dataset, client):
+    "Test read data with a `read_kwargs`"
+    project_id, dataset_id, table_id = dataset
+    ddf = read_gbq(
+        project_id=project_id,
+        dataset_id=dataset_id,
+        table_id=table_id,
+        read_kwargs={"timeout": 1e-12},
+    )
+
+    with pytest.raises(Exception, match="504 Deadline Exceeded"):
+        ddf.compute()

requirements.txt

@@ -0,0 +1,7 @@
+dask
+distributed
+google-cloud-bigquery >= 2.11.0
+google-cloud-bigquery-storage
+pandas
+pandas-gbq
+pyarrow