deltacat 1.1.36__py3-none-any.whl → 2.0__py3-none-any.whl

deltacat/utils/filesystem.py

@@ -0,0 +1,320 @@
+from __future__ import annotations
+
+import re
+from typing import Optional, Tuple, Union, List
+
+import sys
+import urllib
+import pathlib
+
+import pyarrow
+import pyarrow as pa
+from pyarrow.fs import (
+    _resolve_filesystem_and_path,
+    FileSelector,
+    FileInfo,
+    FileType,
+    FileSystem,
+    FSSpecHandler,
+    PyFileSystem,
+)
+
+_LOCAL_SCHEME = "local"
+
+
+def resolve_paths_and_filesystem(
+    paths: Union[str, List[str]],
+    filesystem: pyarrow.fs.FileSystem = None,
+) -> Tuple[List[str], pyarrow.fs.FileSystem]:
+    """
+    Resolves and normalizes all provided paths, infers a filesystem from the
+    paths or validates the provided filesystem against the paths and ensures
+    that all paths use the same filesystem.
+
+    Args:
+        paths: A single file/directory path or a list of file/directory paths.
+            A list of paths can contain both files and directories.
+        filesystem: The filesystem implementation that should be used for
+            reading these files. If None, a filesystem will be inferred. If not
+            None, the provided filesystem will still be validated against all
+            filesystems inferred from the provided paths to ensure
+            compatibility.
+    """
+    if isinstance(paths, str):
+        paths = [paths]
+    if isinstance(paths, pathlib.Path):
+        paths = [str(paths)]
+    elif not isinstance(paths, list) or any(not isinstance(p, str) for p in paths):
+        raise ValueError(
+            "Expected `paths` to be a `str`, `pathlib.Path`, or `list[str]`, but got "
+            f"`{paths}`."
+        )
+    elif len(paths) == 0:
+        raise ValueError("Must provide at least one path.")
+
+    need_unwrap_path_protocol = True
+    if filesystem and not isinstance(filesystem, FileSystem):
+        err_msg = (
+            f"The filesystem passed must either conform to "
+            f"pyarrow.fs.FileSystem, or "
+            f"fsspec.spec.AbstractFileSystem. The provided "
+            f"filesystem was: {filesystem}"
+        )
+        try:
+            import fsspec
+            from fsspec.implementations.http import HTTPFileSystem
+        except ModuleNotFoundError:
+            # If filesystem is not a pyarrow filesystem and fsspec isn't
+            # installed, then filesystem is neither a pyarrow filesystem nor
+            # an fsspec filesystem, so we raise a TypeError.
+            raise TypeError(err_msg) from None
+        if not isinstance(filesystem, fsspec.spec.AbstractFileSystem):
+            raise TypeError(err_msg) from None
+        if isinstance(filesystem, HTTPFileSystem):
+            # If filesystem is fsspec HTTPFileSystem, the protocol/scheme of paths
+            # should not be unwrapped/removed, because HTTPFileSystem expects full file
+            # paths including protocol/scheme. This is different behavior compared to
+            # file systems implementation in pyarrow.fs.FileSystem.
+            need_unwrap_path_protocol = False
+
+        filesystem = PyFileSystem(FSSpecHandler(filesystem))
+
+    resolved_paths = []
+    for path in paths:
+        path = _resolve_custom_scheme(path)
+        try:
+            resolved_filesystem, resolved_path = _resolve_filesystem_and_path(
+                path, filesystem
+            )
+        except pa.lib.ArrowInvalid as e:
+            if "Cannot parse URI" in str(e):
+                resolved_filesystem, resolved_path = _resolve_filesystem_and_path(
+                    _encode_url(path), filesystem
+                )
+                resolved_path = _decode_url(resolved_path)
+            elif "Unrecognized filesystem type in URI" in str(e):
+                scheme = urllib.parse.urlparse(path, allow_fragments=False).scheme
+                if scheme in ["http", "https"]:
+                    # If scheme of path is HTTP and filesystem is not resolved,
+                    # try to use fsspec HTTPFileSystem. This expects fsspec is
+                    # installed.
+                    try:
+                        from fsspec.implementations.http import HTTPFileSystem
+                    except ModuleNotFoundError:
+                        raise ImportError(
+                            "Please install fsspec to read files from HTTP."
+                        ) from None
+
+                    resolved_filesystem = PyFileSystem(FSSpecHandler(HTTPFileSystem()))
+                    resolved_path = path
+                    need_unwrap_path_protocol = False
+                else:
+                    raise
+            else:
+                raise
+        if filesystem is None:
+            filesystem = resolved_filesystem
+        elif need_unwrap_path_protocol:
+            resolved_path = _unwrap_protocol(resolved_path)
+        resolved_path = filesystem.normalize_path(resolved_path)
+        resolved_paths.append(resolved_path)
+
+    return resolved_paths, filesystem
+
+
+def resolve_path_and_filesystem(
+    path: str,
+    filesystem: Optional[pyarrow.fs.FileSystem] = None,
+) -> Tuple[str, pyarrow.fs.FileSystem]:
+    """
+    Resolves and normalizes the provided path, infers a filesystem from the
+    path or validates the provided filesystem against the path.
+
+    Args:
+        path: A single file/directory path.
+        filesystem: The filesystem implementation that should be used for
+            reading these files. If None, a filesystem will be inferred. If not
+            None, the provided filesystem will still be validated against all
+            filesystems inferred from the provided paths to ensure
+            compatibility.
+    """
+    paths, filesystem = resolve_paths_and_filesystem(
+        paths=path,
+        filesystem=filesystem,
+    )
+    assert len(paths) == 1, len(paths)
+    return paths[0], filesystem
+
+
+def list_directory(
+    path: str,
+    filesystem: pyarrow.fs.FileSystem,
+    exclude_prefixes: Optional[List[str]] = None,
+    ignore_missing_path: bool = False,
+    recursive: bool = False,
+) -> List[Tuple[str, int]]:
+    """
+    Expand the provided directory path to a list of file paths.
+
+    Args:
+        path: The directory path to expand.
+        filesystem: The filesystem implementation that should be used for
+            reading these files.
+        exclude_prefixes: The file relative path prefixes that should be
+            excluded from the returned file set. Default excluded prefixes are
+            "." and "_".
+        recursive: Whether to expand subdirectories or not.
+
+    Returns:
+        An iterator of (file_path, file_size) tuples.
+    """
+    if exclude_prefixes is None:
+        exclude_prefixes = [".", "_"]
+
+    selector = FileSelector(
+        base_dir=path,
+        recursive=recursive,
+        allow_not_found=ignore_missing_path,
+    )
+    try:
+        files = filesystem.get_file_info(selector)
+    except OSError as e:
+        if isinstance(e, FileNotFoundError):
+            files = []
+        else:
+            _handle_read_os_error(e, path)
+    base_path = selector.base_dir
+    out = []
+    for file_ in files:
+        file_path = file_.path
+        if not file_path.startswith(base_path):
+            continue
+        relative = file_path[len(base_path) :]
+        if any(relative.startswith(prefix) for prefix in exclude_prefixes):
+            continue
+        out.append((file_path, file_.size))
+    # We sort the paths to guarantee a stable order.
+    return sorted(out)
+
+
+def get_file_info(
+    path: str,
+    filesystem: pyarrow.fs.FileSystem,
+    ignore_missing_path: bool = False,
+) -> FileInfo:
+    """Get the file info for the provided path."""
+    try:
+        file_info = filesystem.get_file_info(path)
+    except OSError as e:
+        _handle_read_os_error(e, path)
+    if file_info.type == FileType.NotFound and not ignore_missing_path:
+        raise FileNotFoundError(path)
+
+    return file_info
+
+
+def _handle_read_os_error(
+    error: OSError,
+    paths: Union[str, List[str]],
+) -> str:
+    # NOTE: this is not comprehensive yet, and should be extended as more errors arise.
+    # NOTE: The latter patterns are raised in Arrow 10+, while the former is raised in
+    # Arrow < 10.
+    aws_error_pattern = (
+        r"^(?:(.*)AWS Error \[code \d+\]: No response body\.(.*))|"
+        r"(?:(.*)AWS Error UNKNOWN \(HTTP status 400\) during HeadObject operation: "
+        r"No response body\.(.*))|"
+        r"(?:(.*)AWS Error ACCESS_DENIED during HeadObject operation: No response "
+        r"body\.(.*))$"
+    )
+    if re.match(aws_error_pattern, str(error)):
+        # Specially handle AWS error when reading files, to give a clearer error
+        # message to avoid confusing users. The real issue is most likely that the AWS
+        # S3 file credentials have not been properly configured yet.
+        if isinstance(paths, str):
+            # Quote to highlight single file path in error message for better
+            # readability. List of file paths will be shown up as ['foo', 'boo'],
+            # so only quote single file path here.
+            paths = f'"{paths}"'
+        raise OSError(
+            (
+                f"Failing to read AWS S3 file(s): {paths}. "
+                "Please check that file exists and has properly configured access. "
+                "You can also run AWS CLI command to get more detailed error message "
+                "(e.g., aws s3 ls <file-name>). "
+                "See https://awscli.amazonaws.com/v2/documentation/api/latest/reference/s3/index.html "  # noqa
+                "for more information."
+            )
+        )
+    else:
+        raise error
+
+
+def _is_local_windows_path(path: str) -> bool:
+    """Determines if path is a Windows file-system location."""
+    if sys.platform != "win32":
+        return False
+
+    if len(path) >= 1 and path[0] == "\\":
+        return True
+    if (
+        len(path) >= 3
+        and path[1] == ":"
+        and (path[2] == "/" or path[2] == "\\")
+        and path[0].isalpha()
+    ):
+        return True
+    return False
+
+
+def _unwrap_protocol(path):
+    """
+    Slice off any protocol prefixes on path.
+    """
+    if sys.platform == "win32" and _is_local_windows_path(path):
+        # Represent as posix path such that downstream functions properly handle it.
+        # This is executed when 'file://' is NOT included in the path.
+        return pathlib.Path(path).as_posix()
+
+    parsed = urllib.parse.urlparse(path, allow_fragments=False)  # support '#' in path
+    query = "?" + parsed.query if parsed.query else ""  # support '?' in path
+    netloc = parsed.netloc
+    if parsed.scheme == "s3" and "@" in parsed.netloc:
+        # If the path contains an @, it is assumed to be an anonymous
+        # credentialed path, and we need to strip off the credentials.
+        netloc = parsed.netloc.split("@")[-1]
+
+    parsed_path = parsed.path
+    # urlparse prepends the path with a '/'. This does not work on Windows
+    # so if this is the case strip the leading slash.
+    if (
+        sys.platform == "win32"
+        and not netloc
+        and len(parsed_path) >= 3
+        and parsed_path[0] == "/"  # The problematic leading slash
+        and parsed_path[1].isalpha()  # Ensure it is a drive letter.
+        and parsed_path[2:4] in (":", ":/")
+    ):
+        parsed_path = parsed_path[1:]
+
+    return netloc + parsed_path + query
+
+
+def _encode_url(path):
+    return urllib.parse.quote(path, safe="/:")
+
+
+def _decode_url(path):
+    return urllib.parse.unquote(path)
+
+
+def _resolve_custom_scheme(path: str) -> str:
+    """Returns the resolved path if the given path follows a Ray-specific custom
+    scheme. Othewise, returns the path unchanged.
+
+    The supported custom schemes are: "local", "example".
+    """
+    parsed_uri = urllib.parse.urlparse(path)
+    if parsed_uri.scheme == _LOCAL_SCHEME:
+        path = parsed_uri.netloc + parsed_uri.path
+    return path

deltacat/utils/metafile_locator.py

@@ -0,0 +1,73 @@
+import posixpath
+import pyarrow.fs
+
+from deltacat.storage.model.partition import PartitionLocator
+from deltacat.utils.filesystem import resolve_path_and_filesystem
+
+"""
+Helper functions to work with deltacat metadata paths.
+TODO: Replace with direct calls to Deltacat storage interface.
+"""
+
+
+def _find_first_child_with_rev(
+    parent_path: str, filesystem: pyarrow.fs.FileSystem
+) -> str:
+    """
+    Walks the filesystem to find the first child directory with a `rev/` folder.
+
+    This is a temporary solution to locate the first Namespace and Table directories.
+    The Deltacat Storage interface will provide a more robust way to locate these directories.
+
+    param: parent_path: The parent directory to search for a child with a `rev/` folder.
+    param: filesystem: The filesystem to search for the child directory.
+    returns: The name of the first child directory with a `rev/` folder.
+    """
+    children = filesystem.get_file_info(
+        pyarrow.fs.FileSelector(parent_path, allow_not_found=True)
+    )
+    for child in children:
+        if child.type == pyarrow.fs.FileType.Directory:
+            rev_path = posixpath.join(child.path, "rev")
+            if filesystem.get_file_info(rev_path).type == pyarrow.fs.FileType.Directory:
+                return child.base_name
+    raise ValueError(f"No directory with 'rev/' found under {parent_path}")
+
+
+def _find_table_path(root_path: str, filesystem: pyarrow.fs.FileSystem):
+    """
+    Finds a path with structure: root/namespace_id/table_id
+    Uses _find_first_child_with_rev to determine the namespace and table ids.
+
+    param: root_path: The root directory to search for the namespace and table directories.
+    param: filesystem: The filesystem to search for the namespace and table directories.
+    returns: The path to the table directory.
+    raises: ValueError if the namespace or table directories are not found.
+    """
+    try:
+        # Find Namespace (first directory under root with rev/)
+        namespace_id = _find_first_child_with_rev(root_path, filesystem)
+        namespace_path = posixpath.join(root_path, namespace_id)
+
+        # Find Table (first directory under namespace with rev/)
+        table_id = _find_first_child_with_rev(namespace_path, filesystem)
+        return posixpath.join(namespace_path, table_id)
+
+    except ValueError as e:
+        raise ValueError(f"Failed to locate Namespace or Table: {e}") from e
+
+
+def _find_partition_path(root_path: str, locator: PartitionLocator) -> str:
+    """
+    Finds the path to the partition directory for the specified locator.
+
+    param: root_uri: The root URI of the dataset.
+    param: locator: The DeltaLocator for the delta.
+    returns: The path to the delta directory.
+    """
+    root_path, filesystem = resolve_path_and_filesystem(root_path)
+    return posixpath.join(
+        _find_table_path(root_path, filesystem),
+        locator.table_version,
+        locator.stream_id,
+    )

deltacat/utils/pyarrow.py

@@ -1,7 +1,6 @@
 # Allow classes to use self-referencing Type hints in Python 3.7.
 from __future__ import annotations
 
-import copy
 import bz2
 import gzip
 import io
@@ -47,19 +46,6 @@ logger = logs.configure_deltacat_logger(logging.getLogger(__name__))
 
 RAISE_ON_EMPTY_CSV_KWARG = "raise_on_empty_csv"
 READER_TYPE_KWARG = "reader_type"
-OVERRIDE_CONTENT_ENCODING_FOR_PARQUET_KWARG = "override_content_encoding_for_parquet"
-
-"""
-By default, round decimal values using half_to_even round mode when
-rescaling a decimal to the given scale and precision in the schema would cause
-data loss. Setting any non null value of this argument will result
-in an error instead.
-"""
-RAISE_ON_DECIMAL_OVERFLOW = "raise_on_decimal_overflow"
-# Note the maximum from https://arrow.apache.org/docs/python/generated/pyarrow.Decimal256Type.html#pyarrow.Decimal256Type
-DECIMAL256_DEFAULT_SCALE = 38
-DECIMAL256_MAX_PRECISION = 76
-MAX_INT_BYTES = 2147483646
 
 
 def _filter_schema_for_columns(schema: pa.Schema, columns: List[str]) -> pa.Schema:
@@ -78,164 +64,45 @@ def _filter_schema_for_columns(schema: pa.Schema, columns: List[str]) -> pa.Sche
     return target_schema
 
 
-def _extract_arrow_schema_from_read_csv_kwargs(kwargs: Dict[str, Any]) -> pa.Schema:
-    schema = None
-    if (
-        "convert_options" in kwargs
-        and kwargs["convert_options"].column_types is not None
-    ):
-        schema = kwargs["convert_options"].column_types
-        if not isinstance(schema, pa.Schema):
-            schema = pa.schema(schema)
-        if kwargs["convert_options"].include_columns:
-            schema = _filter_schema_for_columns(
-                schema, kwargs["convert_options"].include_columns
-            )
-        elif (
-            kwargs.get("read_options") is not None
-            and kwargs["read_options"].column_names
-        ):
-            schema = _filter_schema_for_columns(
-                schema, kwargs["read_options"].column_names
-            )
-    else:
-        logger.debug(
-            "Schema not specified in the kwargs."
-            " Hence, schema could not be inferred from the empty CSV."
-        )
-
-    return schema
-
-
-def _new_schema_with_replaced_fields(
-    schema: pa.Schema, field_to_replace: Callable[[pa.Field], Optional[pa.Field]]
-) -> pa.Schema:
-    if schema is None:
-        return None
-
-    new_schema_fields = []
-    for field in schema:
-        new_field = field_to_replace(field)
-        if new_field is not None:
-            new_schema_fields.append(new_field)
-        else:
-            new_schema_fields.append(field)
-
-    return pa.schema(new_schema_fields, metadata=schema.metadata)
-
-
-def _read_csv_rounding_decimal_columns_to_fit_scale(
-    schema: pa.Schema, reader_args: List[Any], reader_kwargs: Dict[str, Any]
-) -> pa.Table:
-    # Note: We read decimals as strings first because CSV
-    # conversion to decimal256 isn't implemented as of pyarrow==12.0.1
-    new_schema = _new_schema_with_replaced_fields(
-        schema,
-        lambda fld: (
-            pa.field(fld.name, pa.string(), metadata=fld.metadata)
-            if pa.types.is_decimal128(fld.type) or pa.types.is_decimal256(fld.type)
-            else None
-        ),
-    )
-    new_kwargs = sanitize_kwargs_by_supported_kwargs(
-        ["read_options", "parse_options", "convert_options", "memory_pool"],
-        reader_kwargs,
-    )
-    # Creating a shallow copy for efficiency
-    new_convert_options = copy.copy(new_kwargs["convert_options"])
-    new_convert_options.column_types = new_schema
-    new_reader_kwargs = {**new_kwargs, "convert_options": new_convert_options}
-    arrow_table = pacsv.read_csv(*reader_args, **new_reader_kwargs)
-
-    for column_index, field in enumerate(schema):
-        if pa.types.is_decimal128(field.type) or pa.types.is_decimal256(field.type):
-            column_array = arrow_table[field.name]
-            # We always cast to decimal256 to accomodate fixed scale of 38
-            cast_to_type = pa.decimal256(
-                DECIMAL256_MAX_PRECISION, DECIMAL256_DEFAULT_SCALE
-            )
-            casted_decimal_array = pc.cast(column_array, cast_to_type)
-            # Note that scale can be negative
-            rounded_column_array = pc.round(
-                casted_decimal_array, ndigits=field.type.scale
-            )
-            final_decimal_array = pc.cast(rounded_column_array, field.type)
-            arrow_table = arrow_table.set_column(
-                column_index,
-                field,
-                final_decimal_array,
-            )
-            logger.debug(
-                f"Rounded decimal column: {field.name} to {field.type.scale} scale and"
-                f" {field.type.precision} precision"
-            )
-
-    return arrow_table
-
-
-def pyarrow_read_csv_default(*args, **kwargs):
-    new_kwargs = sanitize_kwargs_by_supported_kwargs(
-        ["read_options", "parse_options", "convert_options", "memory_pool"], kwargs
-    )
-
+def pyarrow_read_csv(*args, **kwargs) -> pa.Table:
     try:
+        new_kwargs = sanitize_kwargs_by_supported_kwargs(
+            ["read_options", "parse_options", "convert_options", "memory_pool"], kwargs
+        )
         return pacsv.read_csv(*args, **new_kwargs)
     except pa.lib.ArrowInvalid as e:
-        error_str = e.__str__()
-        schema = _extract_arrow_schema_from_read_csv_kwargs(kwargs)
-
-        if error_str == "Empty CSV file" and not kwargs.get(RAISE_ON_EMPTY_CSV_KWARG):
-            logger.debug(f"Read CSV empty schema being used: {schema}")
-            return pa.Table.from_pylist([], schema=schema)
-        if not kwargs.get(RAISE_ON_DECIMAL_OVERFLOW):
-            # Note, this logic requires expensive casting. To prevent downgrading performance
-            # for happy path reads, we are handling this case in response to an error.
-            logger.warning(
-                "Rescaling Decimal to the given scale in the schema. "
-                f"Original error: {error_str}"
-            )
-
-            if schema is not None and "convert_options" in kwargs:
-                if (
-                    "Rescaling Decimal" in error_str
-                    and "value would cause data loss" in error_str
+        if e.__str__() == "Empty CSV file" and not kwargs.get(RAISE_ON_EMPTY_CSV_KWARG):
+            schema = None
+            if (
+                "convert_options" in kwargs
+                and kwargs["convert_options"].column_types is not None
+            ):
+                schema = kwargs["convert_options"].column_types
+                if not isinstance(schema, pa.Schema):
+                    schema = pa.schema(schema)
+                if kwargs["convert_options"].include_columns:
+                    schema = _filter_schema_for_columns(
+                        schema, kwargs["convert_options"].include_columns
+                    )
+                elif (
+                    kwargs.get("read_options") is not None
+                    and kwargs["read_options"].column_names
                 ):
-                    logger.debug(f"Checking if the file: {args[0]}...")
-                    # Since we are re-reading the file, we have to seek to beginning
-                    if isinstance(args[0], io.IOBase) and args[0].seekable():
-                        logger.debug(f"Seeking to the beginning of the file {args[0]}")
-                        args[0].seek(0)
-                    return _read_csv_rounding_decimal_columns_to_fit_scale(
-                        schema=schema, reader_args=args, reader_kwargs=kwargs
+                    schema = _filter_schema_for_columns(
+                        schema, kwargs["read_options"].column_names
                     )
+
             else:
                 logger.debug(
-                    "Schema is None when trying to adjust decimal values. "
-                    "Hence, bubbling up exception..."
+                    "Schema not specified in the kwargs."
+                    " Hence, schema could not be inferred from the empty CSV."
                 )
 
+            logger.debug(f"Read CSV empty schema being used: {schema}")
+            return pa.Table.from_pylist([], schema=schema)
         raise e
 
 
-def pyarrow_read_csv(*args, **kwargs) -> pa.Table:
-    schema = _extract_arrow_schema_from_read_csv_kwargs(kwargs)
-
-    # CSV conversion to decimal256 isn't supported as of pyarrow=12.0.1
-    # Below ensures decimal256 is casted properly.
-    schema_includes_decimal256 = (
-        (True if any([pa.types.is_decimal256(x.type) for x in schema]) else False)
-        if schema is not None
-        else None
-    )
-    if schema_includes_decimal256 and not kwargs.get(RAISE_ON_DECIMAL_OVERFLOW):
-        # falling back to expensive method of reading CSV
-        return _read_csv_rounding_decimal_columns_to_fit_scale(
-            schema, reader_args=args, reader_kwargs=kwargs
-        )
-    else:
-        return pyarrow_read_csv_default(*args, **kwargs)
-
-
 CONTENT_TYPE_TO_PA_READ_FUNC: Dict[str, Callable] = {
     ContentType.UNESCAPED_TSV.value: pyarrow_read_csv,
     ContentType.TSV.value: pyarrow_read_csv,
@@ -544,15 +411,6 @@ def s3_file_to_table(
     if pa_read_func_kwargs_provider is not None:
         kwargs = pa_read_func_kwargs_provider(content_type, kwargs)
 
-    if OVERRIDE_CONTENT_ENCODING_FOR_PARQUET_KWARG in kwargs:
-        new_content_encoding = kwargs.pop(OVERRIDE_CONTENT_ENCODING_FOR_PARQUET_KWARG)
-        if content_type == ContentType.PARQUET.value:
-            logger.debug(
-                f"Overriding {s3_url} content encoding from {content_encoding} "
-                f"to {new_content_encoding}"
-            )
-            content_encoding = new_content_encoding
-
     if (
         content_type == ContentType.PARQUET.value
         and content_encoding == ContentEncoding.IDENTITY.value
@@ -582,8 +440,8 @@ def s3_file_to_table(
                 **s3_client_kwargs,
             )
 
-    if READER_TYPE_KWARG in kwargs:
-        kwargs.pop(READER_TYPE_KWARG)
+        if READER_TYPE_KWARG in kwargs:
+            kwargs.pop(READER_TYPE_KWARG)
 
     filesystem = io
     if s3_url.startswith("s3://"):
@@ -617,18 +475,7 @@ def s3_file_to_parquet(
         f"Reading {s3_url} to PyArrow ParquetFile. "
         f"Content type: {content_type}. Encoding: {content_encoding}"
     )
-    kwargs = {}
-    if pa_read_func_kwargs_provider:
-        kwargs = pa_read_func_kwargs_provider(content_type, kwargs)
 
-    if OVERRIDE_CONTENT_ENCODING_FOR_PARQUET_KWARG in kwargs:
-        new_content_encoding = kwargs.pop(OVERRIDE_CONTENT_ENCODING_FOR_PARQUET_KWARG)
-        if content_type == ContentType.PARQUET.value:
-            logger.debug(
-                f"Overriding {s3_url} content encoding from {content_encoding} "
-                f"to {new_content_encoding}"
-            )
-            content_encoding = new_content_encoding
     if (
         content_type != ContentType.PARQUET.value
         or content_encoding != ContentEncoding.IDENTITY
@@ -641,10 +488,15 @@ def s3_file_to_parquet(
     if s3_client_kwargs is None:
         s3_client_kwargs = {}
 
+    kwargs = {}
+
     if s3_url.startswith("s3://"):
         s3_file_system = create_s3_file_system(s3_client_kwargs)
         kwargs["filesystem"] = s3_file_system
 
+    if pa_read_func_kwargs_provider:
+        kwargs = pa_read_func_kwargs_provider(content_type, kwargs)
+
     logger.debug(f"Pre-sanitize kwargs for {s3_url}: {kwargs}")
 
     kwargs = sanitize_kwargs_to_callable(ParquetFile.__init__, kwargs)
@@ -931,6 +783,7 @@ def sliced_string_cast(array: pa.ChunkedArray) -> pa.ChunkedArray:
     TODO: deprecate this function when pyarrow performs proper ChunkedArray -> ChunkedArray casting
     """
     dtype = array.type
+    MAX_BYTES = 2147483646
     max_str_len = None
     if pa.types.is_integer(dtype):
         max_str_len = _int_max_string_len()
@@ -942,7 +795,7 @@ def sliced_string_cast(array: pa.ChunkedArray) -> pa.ChunkedArray:
         max_str_len = _max_decimal256_string_len()
 
     if max_str_len is not None:
-        max_elems_per_chunk = MAX_INT_BYTES // (2 * max_str_len)  # safety factor of 2
+        max_elems_per_chunk = MAX_BYTES // (2 * max_str_len)  # safety factor of 2
         all_chunks = []
         for chunk in array.chunks:
             if len(chunk) < max_elems_per_chunk: