Support for PdReader kwargs (AmpX-AI#8)

Co-authored-by: vojta tuma <vtuma@amp.energy>

fsql/deser.py

@@ -17,14 +17,18 @@
 opt for the lazy approach (such as in Dask), and don't materialize inside neither the `read_single` nor
 `concat` methods.
 
+Existing readers such as PandasReader allow customisation via passing through any kwargs to the underlying
+pandas read method.
+
 The user should *not* bake in any specific business logic in here -- a more prefered approach is to
-return an object such as data frame as early as possible, and apply any transformations later on.
+return an object such as (lazy) data frame as early as possible, and apply any transformations later on.
 """
 from __future__ import annotations
 
 import json
 import logging
 from abc import ABC, abstractmethod
+from collections import defaultdict
 from collections.abc import Iterable
 from concurrent.futures import ThreadPoolExecutor
 from enum import Enum, auto, unique
@@ -87,31 +91,51 @@ def read_and_concat(
 
 
 class PandasReader(DataReader):
+    """Wraps various pandas read methods (parquet, json, csv, excel) into a single interface.
+    Behaviour can be customised via passing any kwargs to the constructor.
+    """
+
+    def __init__(self, input_format=InputFormat.AUTO, **pdread_kwargs):
+        super().__init__(input_format=input_format)
+        self.pdread_user_kwargs = pdread_kwargs
+        self.pdread_default_kwargs = defaultdict(dict)
+        self.pdread_default_kwargs[InputFormat.PARQUET] = {
+            "engine": "fastparquet",
+        }
+        self.pdread_default_kwargs[InputFormat.JSON] = {
+            "lines": "true",
+        }
+        self.pdread_default_kwargs[InputFormat.XLSX] = {
+            "engine": "openpyxl",
+        }
+
     def read_single(self, partition: Partition, fs: AbstractFileSystem) -> pd.DataFrame:
         logger.debug(f"read dataframe for partition {partition}")
         input_format = self.detect_format(partition.url)
-        # TODO allow for user spec of engine and other params, essentially any quark
+        logger.debug(f"format detected for partition {input_format} <- {partition}")
         if input_format is InputFormat.PARQUET:
-            reader = lambda fd: pd.read_parquet(fd, engine="fastparquet")  # noqa: E731
+            reader = pd.read_parquet
         elif input_format is InputFormat.JSON:
-            reader = lambda fd: pd.read_json(fd, lines=True)  # noqa: E731
+            reader = pd.read_json
         elif input_format is InputFormat.CSV:
             reader = pd.read_csv
         elif input_format is InputFormat.XLSX:
-            reader = lambda fd: pd.read_excel(fd, engine="openpyxl")  # noqa: E731
+            reader = pd.read_excel
         elif input_format is InputFormat.AUTO:
             raise ValueError(f"partition had format detected as auto -> invalid state. Partition: {partition}")
         else:
             assert_exhaustive_enum(input_format)
 
+        pdread_kwargs = {**self.pdread_default_kwargs[input_format], **self.pdread_user_kwargs}
+        logger.debug(f"reader kwargs {pdread_kwargs} for partition {partition}")
         try:
             with fs.open(partition.url, "rb") as fd:
-                df = reader(fd)
+                df = reader(fd, **pdread_kwargs)
         except FileNotFoundError as e:
             logger.warning(f"file {partition} reading exception {type(e)}, attempting cache invalidation and reread")
             fs.invalidate_cache()
             with fs.open(partition.url, "rb") as fd:
-                df = reader(fd)
+                df = reader(fd, **pdread_kwargs)
 
         for key, value in partition.columns.items():
             df[key] = value

tests/test_pandasreader.py

@@ -0,0 +1,37 @@
+import pandas as pd
+import pytest
+from pandas.testing import assert_frame_equal
+
+from fsql.api import read_partitioned_table
+from fsql.deser import InputFormat, PandasReader
+from fsql.query import Q_TRUE
+
+df1 = pd.DataFrame(data={"c1": [0, 1], "c2": ["hello", "world"]})
+
+
+def test_input_format_override(tmp_path):
+    """Test that explicitly setting format overrides suffix."""
+
+    case1_path = tmp_path / "table1"
+    case1_path.mkdir(parents=True)
+    df1.to_csv(case1_path / "f1.json", index=False)  # confuse the default by bad suffix
+
+    with pytest.raises(ValueError, match="Expected object or value"):
+        # this test condition is quite brittle! A better match would be desired
+        failure_result = read_partitioned_table(f"file://{case1_path}/", Q_TRUE)
+
+    reader = PandasReader(input_format=InputFormat.CSV)
+    succ_result = read_partitioned_table(f"file://{case1_path}/", Q_TRUE, data_reader=reader)
+    assert_frame_equal(df1, succ_result)
+
+
+def test_parquet_kwargs(tmp_path):
+    """Test that a kwarg (`columns`) gets passed through and obeyed."""
+
+    case1_path = tmp_path / "table1"
+    case1_path.mkdir(parents=True)
+    df1.to_parquet(case1_path / "f1.parquet", index=False)
+
+    reader = PandasReader(columns=["c2"])
+    result = read_partitioned_table(f"file://{case1_path}/", Q_TRUE, data_reader=reader)
+    assert_frame_equal(df1[["c2"]], result)