s3fs 2024.3.0 fails reading glob patterns through pandas #862

calloc · 2024-03-17T09:09:59Z

Intention is to read all CSV files under an s3 directory using Pandas via s3fs.
We specify path as s3://some-workspace/tmp/jr_1d02d47beccb8f7d2c2e8429bbaa3e143f90fcb7837373fb565abdade4a8c845_audit/*.csv

Reverting back to old version s3fs==2024.2.0 works. Note pandas==2.1.4.

[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venv-callsn9i4utg/script.py", line 48, in generate_partition_sql
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     df = pd.read_csv(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -          ^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/parsers/readers.py", line 948, in read_csv
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     return _read(filepath_or_buffer, kwds)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/parsers/readers.py", line 611, in _read
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     parser = TextFileReader(filepath_or_buffer, **kwds)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/parsers/readers.py", line 1448, in __init__
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     self._engine = self._make_engine(f, self.engine)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -                    ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/parsers/readers.py", line 1705, in _make_engine
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     self.handles = get_handle(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -                    ^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/common.py", line 718, in get_handle
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     ioargs = _get_filepath_or_buffer(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -              ^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/home/***/.local/lib/python3.11/site-packages/pandas/io/common.py", line 420, in _get_filepath_or_buffer
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     ).open()
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -       ^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/core.py", line 135, in open
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     return self.__enter__()
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -            ^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/core.py", line 103, in __enter__
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     f = self.fs.open(self.path, mode=mode)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/spec.py", line 1293, in open
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     f = self._open(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -         ^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/s3fs/core.py", line 685, in _open
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     return S3File(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -            ^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/s3fs/core.py", line 2179, in __init__
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     super().__init__(
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/spec.py", line 1651, in __init__
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     self.size = self.details["size"]
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -                 ^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/spec.py", line 1664, in details
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     self._details = self.fs.info(self.path)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -                     ^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/asyn.py", line 118, in wrapper
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     return sync(self.loop, func, *args, **kwargs)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/asyn.py", line 103, in sync
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     raise return_result
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/fsspec/asyn.py", line 56, in _runner
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     result[0] = await coro
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -                 ^^^^^^^^^^
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -   File "/tmp/venvogubdfca/lib/python3.11/site-packages/s3fs/core.py", line 1418, in _info
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO -     raise FileNotFoundError(path)
[2024-03-17, 02:39:41 +07] {process_utils.py:190} INFO - FileNotFoundError: some-workspace/tmp/jr_1d02d47beccb8f7d2c2e8429bbaa3e143f90fcb7837373fb565abdade4a8c845_audit/*.csv
[2024-03-17, 02:39:41 +07] {taskinstance.py:2698} ERROR - Task failed with exception
Traceback (most recent call last):
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/models/taskinstance.py", line 433, in _execute_task
    result = execute_callable(context=context, **execute_callable_kwargs)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/decorators/base.py", line 241, in execute
    return_value = super().execute(context)
                   ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 400, in execute
    return super().execute(context=serializable_context)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 199, in execute
    return_value = self.execute_callable()
                   ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 716, in execute_callable
    result = self._execute_python_callable_in_subprocess(python_path)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 471, in _execute_python_callable_in_subprocess
    raise AirflowException(error_msg) from None
airflow.exceptions.AirflowException: Process returned non-zero exit status 1.
some-workspace/tmp/jr_1d02d47beccb8f7d2c2e8429bbaa3e143f90fcb7837373fb565abdade4a8c845_audit/*.csv

The text was updated successfully, but these errors were encountered:

martindurant · 2024-03-17T13:55:42Z

This is a consequence of fsspec/filesystem_spec#1536 (cc @Skylion007 ), and the logic is: pandas only reads single files, but glob can return many. So with the previous version, we would simply pick the first match, but now we assume single paths are not to be expanded, as that would cause ambiguity. The glob() methods still functions as before, and open_files will expand (e.g., as used by dask's read_csv ).

martindurant · 2024-03-17T14:02:47Z

@Skylion007 , do you think we should fall back to the old behaviour in the case that no file is found with the exact given path, and it contains special glob characters?

Skylion007 · 2024-03-17T19:00:08Z

The old behavior can be manually specified by passing expand=True if that is actually the behavior you desire.

martindurant · 2024-03-17T19:14:03Z

The old behavior can be manually specified by passing expand=True if that is actually the behavior you desire.

> pd.read_csv("gcs://mdtemp/*.csv", storage_options=dict(expand=True))
TypeError: fsspec.core.open_files() got multiple values for keyword argument 'expand'

Skylion007 · 2024-03-17T19:32:29Z

Ugh... maybe allowing that arg to be overridden better would be a nicer place to start?

martindurant · 2024-03-17T19:37:53Z

I don't think most users will want to add an extra kwarg to their calls (right, @calloc ?)

Skylion007 · 2024-03-17T19:38:50Z

I am bit confused about the old behavior, wouldn't the previous glob pattern only return the first element if it went through fsspec.open(). fsspec.open shouldn't return a list.

martindurant · 2024-03-17T19:39:34Z

fsspec.open returned fsspec.open_files(..)[0]

martindurant · 2024-03-18T14:50:02Z

The following fixes passing expand=False

--- a/fsspec/core.py
+++ b/fsspec/core.py
@@ -456,6 +456,8 @@ def open(
     - For implementations in separate packages see
       https://filesystem-spec.readthedocs.io/en/latest/api.html#other-known-implementations
     """
+    kw = {"expand": False}
+    kw.update(kwargs)
     out = open_files(
         urlpath=[urlpath],
         mode=mode,
@@ -464,8 +466,7 @@ def open(
         errors=errors,
         protocol=protocol,
         newline=newline,
-        expand=False,
-        **kwargs,
+        **kw,
     )
     if not out:

but simply falling back if open(expand=False) fails does not work, because in the current implementation, open() always succeeds whether or not the target file exists

In [2]: fsspec.open("s3://completely/unknown")
Out[2]: <OpenFile 'completely/unknown'>

and only fails when this OpenFile is actually used. OTOH, we don't really want to check for every concrete filepath passed whether it exists or not, and assume the caller have reasonable parameters. So in the case that the path can be interpreted in multiple ways (i.e., has globp characters), we need special handling.

Thoughts, @Skylion007 ?

martindurant · 2024-03-19T13:29:52Z

I fixed expand= in fsspec/filesystem_spec#1551 , which made release 2024.3.1, but I am still considering the right default value, depending on how many people were relying on the previous (accidental) behaviour.

Skylion007 · 2024-03-19T14:17:29Z

@martindurant This does feel like a trap since it pandas csv will only read 1 glob file in this case, but i suppose supporting this use case is better than not doing so. i was thinking that we may want to refactor it a bit so expand could only be overriden in fsspec.open() if we want to keep the old erroring behavior moving forward (by using kwargs.setdefault or something similar),

martindurant · 2024-03-19T14:20:37Z

Following my PR, changing the default is easy, although I suppose it could be a config option rather than/as well as a kwarg.

martindurant mentioned this issue Mar 19, 2024

FileNotFoundError since 2024.3.1 fsspec/gcsfs#616

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

s3fs 2024.3.0 fails reading glob patterns through pandas #862

s3fs 2024.3.0 fails reading glob patterns through pandas #862

calloc commented Mar 17, 2024 •

edited

martindurant commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

martindurant commented Mar 18, 2024

martindurant commented Mar 19, 2024

Skylion007 commented Mar 19, 2024

martindurant commented Mar 19, 2024

s3fs 2024.3.0 fails reading glob patterns through pandas #862

s3fs 2024.3.0 fails reading glob patterns through pandas #862

Comments

calloc commented Mar 17, 2024 • edited

martindurant commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

Skylion007 commented Mar 17, 2024

martindurant commented Mar 17, 2024

martindurant commented Mar 18, 2024

martindurant commented Mar 19, 2024

Skylion007 commented Mar 19, 2024

martindurant commented Mar 19, 2024

calloc commented Mar 17, 2024 •

edited