(enhancement): Apply modin repartitioning where required only

awswrangler/distributed/ray/_register.py

@@ -3,16 +3,16 @@
 from awswrangler._data_types import pyarrow_types_from_pandas
 from awswrangler._distributed import MemoryFormatEnum, engine, memory_format
 from awswrangler._utils import is_pandas_frame, table_refs_to_df
-from awswrangler.distributed.ray._core import ray_remote
+from awswrangler.distributed.ray import ray_remote
 from awswrangler.lakeformation._read import _get_work_unit_results
 from awswrangler.s3._delete import _delete_objects
 from awswrangler.s3._read_parquet import _read_parquet, _read_parquet_metadata_file
 from awswrangler.s3._read_text import _read_text
 from awswrangler.s3._select import _select_object_content, _select_query
 from awswrangler.s3._wait import _wait_object_batch
 from awswrangler.s3._write_dataset import _to_buckets, _to_partitions
-from awswrangler.s3._write_parquet import _to_parquet, to_parquet
-from awswrangler.s3._write_text import _to_text, to_csv, to_json
+from awswrangler.s3._write_parquet import _to_parquet
+from awswrangler.s3._write_text import _to_text
 
 
 def register_ray() -> None:
@@ -28,7 +28,6 @@ def register_ray() -> None:
         engine.register_func(func, ray_remote(func))
 
     if memory_format.get() == MemoryFormatEnum.MODIN:
-        from awswrangler.distributed.ray.modin._core import modin_repartition
         from awswrangler.distributed.ray.modin._data_types import pyarrow_types_from_pandas_distributed
         from awswrangler.distributed.ray.modin._utils import _arrow_refs_to_df, _is_pandas_or_modin_frame
         from awswrangler.distributed.ray.modin.s3._read_parquet import _read_parquet_distributed
@@ -48,9 +47,6 @@ def register_ray() -> None:
             _to_parquet: _to_parquet_distributed,
             _to_partitions: _to_partitions_distributed,
             _to_text: _to_text_distributed,
-            to_csv: modin_repartition(to_csv),
-            to_json: modin_repartition(to_json),
-            to_parquet: modin_repartition(to_parquet),
             table_refs_to_df: _arrow_refs_to_df,
             is_pandas_frame: _is_pandas_or_modin_frame,
         }.items():

awswrangler/distributed/ray/modin/_data_types.py

@@ -6,7 +6,7 @@
 import ray
 
 from awswrangler._data_types import pyarrow_types_from_pandas
-from awswrangler.distributed.ray._core import ray_get, ray_remote
+from awswrangler.distributed.ray import ray_get, ray_remote
 
 
 def pyarrow_types_from_pandas_distributed(

awswrangler/distributed/ray/modin/s3/_write_dataset.py

@@ -9,6 +9,7 @@
 
 from awswrangler._distributed import engine
 from awswrangler.distributed.ray import ray_get, ray_remote
+from awswrangler.distributed.ray.modin import modin_repartition
 from awswrangler.s3._write_concurrent import _WriteProxy
 from awswrangler.s3._write_dataset import _delete_objects, _get_bucketing_series, _to_partitions
 
@@ -22,6 +23,7 @@ def _retrieve_paths(values: Union[str, List[Any]]) -> Iterator[str]:
     yield values
 
 
+@modin_repartition
 def _to_buckets_distributed(  # pylint: disable=unused-argument
     df: pd.DataFrame,
     func: Callable[..., List[str]],
@@ -109,6 +111,7 @@ def _write_partitions_distributed(
     return prefix, df_group.name, paths
 
 
+@modin_repartition
 def _to_partitions_distributed(  # pylint: disable=unused-argument
     df: pd.DataFrame,
     func: Callable[..., List[str]],

tests/load/test_s3.py

@@ -1,6 +1,11 @@
+import math
+
+import modin.config as cfg
 import modin.pandas as pd
+import numpy as np
 import pytest
 import ray
+from modin.distributed.dataframe.pandas import unwrap_partitions
 
 import awswrangler as wr
 
@@ -192,3 +197,27 @@ def test_wait_object_not_exists(path: str, benchmark_time: int) -> None:
         wr.s3.wait_objects_not_exist(file_paths, parallelism=16)
 
     assert timer.elapsed_time < benchmark_time
+
+
+@pytest.mark.parametrize("size", [(5000, 5000), (1, 5000), (5000, 1), (1, 1)])
+def test_wide_df(size, path) -> None:
+    df = pd.DataFrame(np.random.randint(0, 100, size=size))
+    df.columns = df.columns.map(str)
+
+    num_cols = size[0]
+    df["int"] = np.random.choice(["1", "2", None], num_cols)
+    df["decimal"] = np.random.choice(["1.0", "2.0", None], num_cols)
+    df["date"] = np.random.choice(["2020-01-01", "2020-01-02", None], num_cols)
+    df["par0"] = np.random.choice(["A", "B"], num_cols)
+
+    partitions_shape = np.array(unwrap_partitions(df)).shape
+    assert partitions_shape[1] == min(math.ceil(len(df.columns) / cfg.MinPartitionSize.get()), cfg.NPartitions.get())
+
+    dtype = {
+        "int": "tinyint",
+        "decimal": "double",
+        "date": "date",
+    }
+
+    result = wr.s3.to_csv(df=df, path=path, dataset=True, dtype=dtype, partition_cols=["par0"])
+    assert len(result["paths"]) == partitions_shape[0] * len(df["par0"].unique())

tests/unit/test_s3_text.py

@@ -163,11 +163,10 @@ def test_csv_dataset_header_modes(path, mode, glue_database, glue_table):
             index=False,
             header=True,
         )
-    dfs_conc = pd.concat(dfs)
     df_res = wr.s3.read_csv(path=path0)
 
     if mode == "append":
-        assert len(df_res) == len(dfs_conc)
+        assert len(df_res) == sum([len(df) for df in dfs])
     else:
         assert df_res.equals(dfs[-1])