[BEAM-10111] Create methods in fileio to read from / write to archive files

• Type: Improvement
• Status: Open
• Priority: P3
• Resolution: Unresolved
• Component/s: io-py-files
• Assignee: Unassigned
• Reporter: Ashwin Ramaswami
• Created: 27/May/20 16:06
• Updated: 04/Jun/22 16:24

Description

Discussion here: https://lists.apache.org/thread.html/r784701bda9edf9a52d5ee593f44a8870aab96b6df1dc8eedd2c8a249%40%3Cdev.beam.apache.org%3E

It would be good to be able to read from / write to archive files (.zip, .tar) using fileio. The difference between this proposal and what we already have with CompressionTypes is that this would allow converting one file -> multiple files and vice versa. Here's how it might look like:

Reading all contents from archive files:

    files = (
        p
        | fileio.MatchFiles('hdfs://path/to/*.zip')
        | fileio.ExtractMatches()
        | fileio.MatchAll()
        | fileio.ReadMatches()
        | beam.Map(lambda x: (x.metadata.path, x.metadata._parent_archive_paths, x.read_utf8()))
    )

Nested archive example: (look for all inside of .tar inside of .zip)

    files = (
        p
        | fileio.MatchFiles('hdfs://path/to/*.zip')
        | fileio.ExtractMatches()
        | fileio.MatchAll('*.tar')
        | fileio.Extract()
        | fileio.MatchAll() # gets all entries
        | fileio.ReadMatches()
        | beam.Map(lambda x: (x.metadata.path, x.read_utf8()))
    )

Note that in this case, this would involve modifying MatchAll() to take an argument, which would filter the files in the pcollection in the earlier stage of the pipeline.

Reading from archive files and explicitly specifying the archive type (when it can't be inferred by the file extension):

    files = (
        p
        | fileio.MatchFiles('hdfs://path/to/archive')
        | fileio.ExtractMatches(archivesystem=ArchiveSystem.TAR)
        | fileio.MatchAll(archive_path='*.txt')
        | fileio.ReadMatches()
        | beam.Map(lambda x: (x.metadata.path, x.read_utf8()))
    )

`ArchiveSystem` would be a generic class, just like `FileSystem`, which would allow for different implementations of methods such as `list()` and `extract()`. It would be implemented for .zip, .tar, etc.

*Writing multiple files to an archive file:*

    files = (
        p
        | fileio.MatchFiles('hdfs://path/to/files/*.txt')
        | fileio.CompressMatches(archivesystem=ArchiveSystem.ZIP)
        | fileio.WriteToArchive("output.zip")
    )

*Writing to a .tar.gz file:*

    files = (
        p
        | fileio.MatchFiles('hdfs://path/to/files/*.txt')
        | fileio.CompressMatches(archivesystem=ArchiveSystem.TAR)
        | fileio.WriteToArchive("output.tar.gz")
    )

Issue Links

• links to: GitHub Pull Request #11840