Class FilesystemResolver

petastorm/fs_utils.py:41–176

Resolves a dataset URL, makes a connection via pyarrow, and provides a filesystem object.

class FilesystemResolver(object):
    """Resolves a dataset URL, makes a connection via pyarrow, and provides a filesystem object."""

    def __init__(self, dataset_url, hadoop_configuration=None, connector=HdfsConnector,
                 hdfs_driver='libhdfs3', user=None, storage_options=None):
        """
        Given a dataset URL and an optional hadoop configuration, parse and interpret the URL to
        instantiate a pyarrow filesystem.

        Interpretation of the URL ``scheme://hostname:port/path`` occurs in the following order:

        1. If no ``scheme``, no longer supported, so raise an exception!
        2. If ``scheme`` is ``file``, use local filesystem path.
        3. If ``scheme`` is ``hdfs``:
           a. Try the ``hostname`` as a namespace and attempt to connect to a name node.
              1. If that doesn't work, try connecting directly to namenode ``hostname:port``.
           b. If no host, connect to the default name node.
        5. If ``scheme`` is ``s3``, use s3fs. The user must manually install s3fs before using s3
        6. If ``scheme`` is ``gs``or ``gcs``, use gcsfs. The user must manually install gcsfs before using GCS
        7. Fail otherwise.

        :param dataset_url: The hdfs URL or absolute path to the dataset
        :param hadoop_configuration: an optional hadoop configuration
        :param connector: the HDFS connector object to use (ONLY override for testing purposes)
        :param hdfs_driver: A string denoting the hdfs driver to use (if using a dataset on hdfs). Current choices are
        libhdfs (java through JNI) or libhdfs3 (C++)
        :param user: String denoting username when connecting to HDFS. None implies login user.
        :param storage_options: Dict of kwargs forwarded to ``fsspec`` to initialize the filesystem.
        """
        # Cache both the original URL and the resolved, urlparsed dataset_url
        self._dataset_url = dataset_url
        self._parsed_dataset_url = None
        # Cache the instantiated filesystem object
        self._filesystem = None

        if isinstance(self._dataset_url, six.string_types):
            self._parsed_dataset_url = urlparse(self._dataset_url)
        else:
            self._parsed_dataset_url = self._dataset_url

        if not self._parsed_dataset_url.scheme:
            # Case 1
            raise ValueError('ERROR! A scheme-less dataset url ({}) is no longer supported. '
                             'Please prepend "file://" for local filesystem.'.format(self._parsed_dataset_url.scheme))

        elif self._parsed_dataset_url.scheme == 'file':
            # Case 2: definitely local
            self._filesystem = pyarrow.localfs
            self._filesystem_factory = lambda: pyarrow.localfs

        elif self._parsed_dataset_url.scheme == 'hdfs':

            if hdfs_driver == 'libhdfs3':
                # libhdfs3 does not do any namenode resolution itself so we do it manually. This is not necessary
                # if using libhdfs

                # Obtain singleton and force hadoop config evaluation
                namenode_resolver = HdfsNamenodeResolver(hadoop_configuration)