Skip to content

pipe_file fails to invalidate directory cache #903

Closed
@TomAugspurger

Description

@TomAugspurger

The sequence of calls ls -> pipe_file -> ls causes the newly written file to not be found in the second ls. Here's a failing test (using Zarr's test setup, which I think was taken from s3fs).

async def test_invalidate_cache(s3: s3fs.S3FileSystem) -> None:
    await s3._call_s3("put_object", Bucket=test_bucket_name, Key="a/b.txt", Body=b"abc")
    before = await s3._ls(f"{test_bucket_name}/a/")
    assert sorted(before) == ["test/a/b.txt"]
    await s3._pipe_file(f"{test_bucket_name}/a/c.txt", data=b"abc")
    after = await s3._ls(f"{test_bucket_name}/a/")
    assert sorted(after) == ["test/a/b.txt", "test/a/c.txt"]

That fails with

    async def test_invalidate_cache(s3: s3fs.S3FileSystem) -> None:
        await s3._call_s3("put_object", Bucket=test_bucket_name, Key="a/b.txt", Body=b"abc")
        before = await s3._ls(f"{test_bucket_name}/a/")
        assert sorted(before) == ["test/a/b.txt"]
        await s3._pipe_file(f"{test_bucket_name}/a/c.txt", data=b"abc")
        after = await s3._ls(f"{test_bucket_name}/a/")
>       assert sorted(after) == ["test/a/b.txt", "test/a/c.txt"]
E       AssertionError: assert ['test/a/b.txt'] == ['test/a/b.tx...test/a/c.txt']
E         
E         Right contains one more item: 'test/a/c.txt'
E         
E         Full diff:
E           [
E               'test/a/b.txt',
E         -     'test/a/c.txt',
E           ]

tests/test_store/test_remote.py:275: AssertionError

I think the issue is inpipe_file, which has an early return at

s3fs/s3fs/core.py

Lines 1141 to 1143 in dd75a1a

return await self._call_s3(
"put_object", Bucket=bucket, Key=key, Body=data, **kwargs
)
. The self.invalidate_cache at the end of the function isn't run.

xref pydata/xarray#9515

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions