samredai commented on a change in pull request #4081: URL: https://github.com/apache/iceberg/pull/4081#discussion_r806126150
########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,215 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesytem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +from typing import Union +from urllib.parse import ParseResult, urlparse + +from pyarrow import NativeFile +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + + +class PyArrowInputFile(InputFile): + """An InputFile implementation that uses a pyarrow filesystem to generate pyarrow.lib.NativeFile instances for reading + + Args: + location(str): A URI or a path to a local file + + Attributes: + location(str): The URI or path to a local file for a PyArrowInputFile instance + parsed_location(urllib.parse.ParseResult): The parsed location with attributes `scheme`, `netloc`, `path`, `params`, + `query`, and `fragment` + exists(bool): Whether the file exists or not + + Examples: + >>> from iceberg.io.pyarrow import PyArrowInputFile + >>> input_file = PyArrowInputFile("s3://foo/bar.txt") + >>> file_content = input_file.open().read() + """ + + def __init__(self, location: str): + parsed_location = urlparse(location) # Create a ParseResult from the uri + + if parsed_location.scheme and parsed_location.scheme not in ( + "file", + "mock", + "s3fs", + "hdfs", + "viewfs", + ): # Validate that a uri is provided with a scheme of `file` + raise ValueError("PyArrowInputFile location must have a scheme of `file`, `mock`, `s3fs`, `hdfs`, or `viewfs`") Review comment: Based on the response on dev@arrow, it looks like caching here is reasonable. I updated the PR to do this and it actually streamlined a few things (removed the need for `parsed_location` to be a class attribute and made some patching in tests a bit easier where you can just override self._filesystem now). The approach I took was adding a module level `_FILESYSTEM_INSTANCES` dictionary to use as a cache and a module level function `get_filesystem`. The function takes a location, pulls out the scheme, and then checks `_FILESYSTEM_INSTANCES` to if a filesystem already exists for that scheme. If the key doesn't exist, it instantiates a new filesystem using `FileSystem.from_uri(location)` and adds that filesystem instance to `_FILESYSTEM_INSTANCES`. ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,215 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesytem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +from typing import Union +from urllib.parse import ParseResult, urlparse + +from pyarrow import NativeFile +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + + +class PyArrowInputFile(InputFile): + """An InputFile implementation that uses a pyarrow filesystem to generate pyarrow.lib.NativeFile instances for reading + + Args: + location(str): A URI or a path to a local file + + Attributes: + location(str): The URI or path to a local file for a PyArrowInputFile instance + parsed_location(urllib.parse.ParseResult): The parsed location with attributes `scheme`, `netloc`, `path`, `params`, + `query`, and `fragment` + exists(bool): Whether the file exists or not + + Examples: + >>> from iceberg.io.pyarrow import PyArrowInputFile + >>> input_file = PyArrowInputFile("s3://foo/bar.txt") + >>> file_content = input_file.open().read() + """ + + def __init__(self, location: str): + parsed_location = urlparse(location) # Create a ParseResult from the uri + + if parsed_location.scheme and parsed_location.scheme not in ( + "file", + "mock", + "s3fs", + "hdfs", + "viewfs", + ): # Validate that a uri is provided with a scheme of `file` + raise ValueError("PyArrowInputFile location must have a scheme of `file`, `mock`, `s3fs`, `hdfs`, or `viewfs`") Review comment: Based on the response on dev@arrow, it looks like caching here is reasonable. I updated the PR to do this and it actually streamlined a few things (removed the need for `parsed_location` to be a class attribute and made some patching in tests a bit easier where you can just override self._filesystem now). The approach I took was adding a module level `_FILESYSTEM_INSTANCES` dictionary to use as a cache and a module level function `get_filesystem`. The function takes a location, pulls out the scheme, and then checks `_FILESYSTEM_INSTANCES` to determine if a cached filesystem already exists for that scheme. If the key doesn't exist, it instantiates a new filesystem using `FileSystem.from_uri(location)` and adds that filesystem instance to `_FILESYSTEM_INSTANCES`. ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,215 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesytem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +from typing import Union +from urllib.parse import ParseResult, urlparse + +from pyarrow import NativeFile +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + + +class PyArrowInputFile(InputFile): + """An InputFile implementation that uses a pyarrow filesystem to generate pyarrow.lib.NativeFile instances for reading + + Args: + location(str): A URI or a path to a local file + + Attributes: + location(str): The URI or path to a local file for a PyArrowInputFile instance + parsed_location(urllib.parse.ParseResult): The parsed location with attributes `scheme`, `netloc`, `path`, `params`, + `query`, and `fragment` + exists(bool): Whether the file exists or not + + Examples: + >>> from iceberg.io.pyarrow import PyArrowInputFile + >>> input_file = PyArrowInputFile("s3://foo/bar.txt") + >>> file_content = input_file.open().read() + """ + + def __init__(self, location: str): + parsed_location = urlparse(location) # Create a ParseResult from the uri + + if parsed_location.scheme and parsed_location.scheme not in ( + "file", + "mock", + "s3fs", + "hdfs", + "viewfs", + ): # Validate that a uri is provided with a scheme of `file` + raise ValueError("PyArrowInputFile location must have a scheme of `file`, `mock`, `s3fs`, `hdfs`, or `viewfs`") Review comment: Based on the response on dev@arrow, it looks like caching here is reasonable. I updated the PR to do this and it actually streamlined a few things (removed the need for `parsed_location` to be a class attribute and made some patching in tests a bit easier where you can just override self._filesystem now). The approach I took was adding a module level `_FILESYSTEM_INSTANCES` dictionary to use as a cache and a module level function `get_filesystem`. The function takes a location, pulls out the scheme, and then checks `_FILESYSTEM_INSTANCES` to determine if a cached filesystem already exists for that scheme. If the key doesn't exist, it instantiates a new filesystem using `FileSystem.from_uri(location)` and adds that filesystem instance to `_FILESYSTEM_INSTANCES`. The constructor for `PyArrowFile` then just does `self._filesystem = get_filesystem(location)`. ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,215 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesytem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +from typing import Union +from urllib.parse import ParseResult, urlparse + +from pyarrow import NativeFile +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + + +class PyArrowInputFile(InputFile): + """An InputFile implementation that uses a pyarrow filesystem to generate pyarrow.lib.NativeFile instances for reading + + Args: + location(str): A URI or a path to a local file + + Attributes: + location(str): The URI or path to a local file for a PyArrowInputFile instance + parsed_location(urllib.parse.ParseResult): The parsed location with attributes `scheme`, `netloc`, `path`, `params`, + `query`, and `fragment` + exists(bool): Whether the file exists or not + + Examples: + >>> from iceberg.io.pyarrow import PyArrowInputFile + >>> input_file = PyArrowInputFile("s3://foo/bar.txt") + >>> file_content = input_file.open().read() + """ + + def __init__(self, location: str): + parsed_location = urlparse(location) # Create a ParseResult from the uri + + if parsed_location.scheme and parsed_location.scheme not in ( + "file", + "mock", + "s3fs", + "hdfs", + "viewfs", + ): # Validate that a uri is provided with a scheme of `file` + raise ValueError("PyArrowInputFile location must have a scheme of `file`, `mock`, `s3fs`, `hdfs`, or `viewfs`") Review comment: Based on the response on dev@arrow, it looks like caching here is reasonable. I updated the PR to do this and it actually streamlined a few things (removed the need for `parsed_location` to be a class attribute and made some patching in tests a bit easier where you can just override self._filesystem now). The approach I took was adding a module level `_FILESYSTEM_INSTANCES` dictionary to use as a cache and a module level function `get_filesystem`. The function takes a location, pulls out the scheme, and then checks `_FILESYSTEM_INSTANCES` to determine if a cached filesystem already exists for that scheme. If the key doesn't exist, it instantiates a new filesystem using `FileSystem.from_uri(location)` and adds that filesystem instance to `_FILESYSTEM_INSTANCES`. The constructor for `PyArrowFile` then just does `self._filesystem = get_filesystem(location)` and uses that attribute throughout the other methods. ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,178 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesystem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +import os +from typing import Union +from urllib.parse import urlparse + +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + +_FILESYSTEM_INSTANCES: dict = {} + + +def get_filesystem(location: str): + """Retrieve a pyarrow.fs.FileSystem instance + + This method checks _FILESYSTEM_INSTANCES for an existing filesystem based on the location's + scheme, i.e. s3, hdfs, viewfs. If an existing filesystem has not been cached, it instantiates a new + filesystem using `pyarrow.fs.FileSystem.from_uri(location)`, caches the returned filesystem, and + also returns that filesystem. If a path with no scheme is provided, it's assumed to be a path to + a local file. + + Args: + location(str): The location of the file + + Returns: + pyarrow.fs.FileSystem: An implementation of the FileSystem base class inferred from the location + + Raises: + ArrowInvalid: A suitable FileSystem implementation cannot be found based on the location provided + """ + parsed_location = urlparse(location) # Create a ParseResult from the uri + if not parsed_location.scheme: # If no scheme, assume the path is to a local file + if _FILESYSTEM_INSTANCES.get("local"): + filesystem = _FILESYSTEM_INSTANCES["local"] + else: + filesystem, _ = FileSystem.from_uri(os.path.abspath(location)) + _FILESYSTEM_INSTANCES["local"] = filesystem + elif _FILESYSTEM_INSTANCES.get(parsed_location.scheme): # Check for a cached filesystem Review comment: Hmm, I see, I just read Weston's response. I'll revert this commit and just do `from_uri` in the `PyArrowFile` constructor. Maybe we can then just wait to optimize that and think through some approaches at that time. How does that sound? ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,178 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesystem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +import os +from typing import Union +from urllib.parse import urlparse + +from pyarrow.fs import FileSystem, FileType + +from iceberg.io.base import FileIO, InputFile, InputStream, OutputFile, OutputStream + +_FILESYSTEM_INSTANCES: dict = {} + + +def get_filesystem(location: str): + """Retrieve a pyarrow.fs.FileSystem instance + + This method checks _FILESYSTEM_INSTANCES for an existing filesystem based on the location's + scheme, i.e. s3, hdfs, viewfs. If an existing filesystem has not been cached, it instantiates a new + filesystem using `pyarrow.fs.FileSystem.from_uri(location)`, caches the returned filesystem, and + also returns that filesystem. If a path with no scheme is provided, it's assumed to be a path to + a local file. + + Args: + location(str): The location of the file + + Returns: + pyarrow.fs.FileSystem: An implementation of the FileSystem base class inferred from the location + + Raises: + ArrowInvalid: A suitable FileSystem implementation cannot be found based on the location provided + """ + parsed_location = urlparse(location) # Create a ParseResult from the uri + if not parsed_location.scheme: # If no scheme, assume the path is to a local file + if _FILESYSTEM_INSTANCES.get("local"): + filesystem = _FILESYSTEM_INSTANCES["local"] + else: + filesystem, _ = FileSystem.from_uri(os.path.abspath(location)) + _FILESYSTEM_INSTANCES["local"] = filesystem + elif _FILESYSTEM_INSTANCES.get(parsed_location.scheme): # Check for a cached filesystem Review comment: I just realized urlparse actually parses the bucket name as `netloc` which makes sense since the bucket name is the authority in the URI. For example: ```py from urllib.parse import urlparse urlparse("s3://foobucket/test.txt") # ParseResult(scheme='s3', netloc='foobucket', path='/test.txt', params='', query='', fragment='') ``` So we could cache for S3 only based on bucket like you suggested by using a tuple as the key. Something like this: ```py parsed_location = urlparse(location) if parsed_location.scheme == "s3": s3_fs_lookup = (parsed_location.scheme, parsed_location.netloc) if _FILESYSTEM_INSTANCES.get(s3_fs_lookup): filesystem = _FILESYSTEM_INSTANCES[s3_fs_lookup] else: filesystem, _ = FileSystem.from_uri(location) _FILESYSTEM_INSTANCES[s3_fs_lookup] = filesystem return filesystem filesystem, _ = FileSystem.from_uri(location) return filesystem ``` ########## File path: python/src/iceberg/io/pyarrow.py ########## @@ -0,0 +1,178 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. +"""FileIO implementation for reading and writing table files that uses pyarrow.fs + +This file contains a FileIO implementation that relies on the filesystem interface provided +by PyArrow. It relies on PyArrow's `from_uri` method that infers the correct filesystem +type to use. Theoretically, this allows the supported storage types to grow naturally +with the pyarrow library. +""" + +import os +from typing import Union +from urllib.parse import urlparse + +from pyarrow.fs import FileSystem, FileType Review comment: Yes I would say it's a matter of not importing this module. If we import from this file anywhere else in the library then I think we'd have to do a graceful fail there (but I'm expecting that we'll just use the FileIO abc's throughout the library). -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
