def __init__(self,
pyarrow_filesystem,
dataset_path,
schema_fields=None,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
reader_pool=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
cache=None,
worker_class=None):
"""Initializes a reader object.
:param pyarrow_filesystem: An instance of ``pyarrow.FileSystem`` that will be used. If not specified,
then a default one will be selected based on the url (only for ``hdfs://`` or ``file://``; for
``s3://`` support, use ``make_reader``). The default hdfs driver is ``libhdfs3``. If you want
to to use ``libhdfs``, use
``pyarrow_filesystem=pyarrow.hdfs.connect('hdfs:///some/path', driver='libhdfs')``.
:param dataset_path: filepath to a parquet directory on the specified filesystem.
e.g. ``'/user/yevgeni/parquet8'``, or ``'/tmp/mydataset'``.
:param schema_fields: Either list of unischema fields to subset, or ``None`` to read all fields.
OR an NGram object, then it will return an NGram of the specified properties.
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of predicate object to filter rows to be returned by reader.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param reader_pool: parallelization pool. ``ThreadPool(10)`` (10 threads) is used by default.
This pool is a custom implementation used to parallelize reading data from the dataset.
Any object from workers_pool package can be used
(e.g. :class:`petastorm.workers_pool.process_pool.ProcessPool`).
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number used. Each reader instance should
pass in a unique shard number in the range ``[0, shard_count)``.
``shard_count`` must be supplied as well. Defaults to None
:param shard_count: An int denoting the number of shard partitions there are. Defaults to None
:param cache: An object conforming to :class:`.CacheBase` interface. Before loading row groups from a parquet
file the Reader will attempt to load these values from cache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shards are used).
By default, use the :class:`.NullCache` implementation.
:param worker_class: This is the class that will be instantiated on a different thread/process. It's
responsibility is to load and filter the data.
"""
# 1. Open the parquet storage (dataset)
# 2. Get a list of all groups
# 3. Filter rowgroups
# a. predicates
# b. row-group selector (our indexing mechanism)
# c. partition: used to get a subset of data for distributed training
# 4. Create a rowgroup ventilator object
# 5. Start workers pool
if not (isinstance(schema_fields, collections.Iterable)
or isinstance(schema_fields, NGram) or schema_fields is None):
raise ValueError(
"""Fields must be either None, an iterable collection of Unischema fields or an NGram
object.""")
self.ngram = schema_fields if isinstance(schema_fields,
NGram) else None
# By default, use original method of working with list of dictionaries and not arrow tables
worker_class = worker_class or PyDictReaderWorker
self._results_queue_reader = worker_class.new_results_queue_reader()
if self.ngram and not self.ngram.timestamp_overlap and shuffle_row_drop_partitions > 1:
raise NotImplementedError(
'Using timestamp_overlap=False is not implemented with'
' shuffle_options.shuffle_row_drop_partitions > 1')
cache = cache or NullCache()
self._workers_pool = reader_pool or ThreadPool(10)
# 1. Resolve dataset path (hdfs://, file://) and open the parquet storage (dataset)
self.dataset = pq.ParquetDataset(dataset_path,
filesystem=pyarrow_filesystem,
validate_schema=False)
stored_schema = infer_or_load_unischema(self.dataset)
# Make a schema view (a view is a Unischema containing only a subset of fields
# Will raise an exception if invalid schema fields are in schema_fields
fields = schema_fields if isinstance(schema_fields,
collections.Iterable) else None
self.schema = stored_schema.create_schema_view(
fields) if fields else stored_schema
# 2. Get a list of all row groups
row_groups = dataset_metadata.load_row_groups(self.dataset)
# 3. Filter rowgroups
filtered_row_group_indexes, worker_predicate = self._filter_row_groups(
self.dataset, row_groups, predicate, rowgroup_selector, cur_shard,
shard_count)
# 4. Create a rowgroup ventilator object
normalized_shuffle_row_drop_partitions = \
self._normalize_shuffle_options(shuffle_row_drop_partitions, self.dataset)
ventilator = self._create_ventilator(
filtered_row_group_indexes, shuffle_row_groups,
normalized_shuffle_row_drop_partitions, num_epochs,
worker_predicate,
self._workers_pool.workers_count + _VENTILATE_EXTRA_ROWGROUPS)
# 5. Start workers pool
self._workers_pool.start(worker_class,
(pyarrow_filesystem, dataset_path,
self.schema, self.ngram, row_groups, cache),
ventilator=ventilator)
logger.debug('Workers pool started')
self.last_row_consumed = False
def make_reader(dataset_url,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
pyarrow_serialize=False,
results_queue_size=50,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
cache_type='null',
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3',
reader_engine='reader_v1',
reader_engine_params=None):
"""
Creates an instance of Reader for reading Petastorm datasets. A Petastorm dataset is a dataset generated using
:func:`~petastorm.etl.dataset_metadata.materialize_dataset` context manager as explained
`here <https://petastorm.readthedocs.io/en/latest/readme_include.html#generating-a-dataset>`_.
See :func:`~petastorm.make_batch_reader` to read from a Parquet store that was not generated using
:func:`~petastorm.etl.dataset_metadata.materialize_dataset`.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'file:///tmp/mydataset'``
or ``'s3://bucket/mydataset'``.
:param schema_fields: Can be: a list of unischema fields and/or regex pattern strings; ``None`` to read all fields;
an NGram object, then it will return an NGram of the specified fields.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param pyarrow_serialize: Whether to use pyarrow for serialization. Currently only applicable to process pool.
Defaults to False.
:param results_queue_size: Size of the results queue to store prefetched rows. Currently only applicable to
thread reader pool type.
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a single row and must return a boolean value indicating whether to include it in the results.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 reader_engine: Multiple engine implementations exist ('reader_v1' and 'experimental_reader_v2'). 'reader_v1'
(the default value) selects a stable reader implementation.
:param reader_engine_params: For advanced usage: a dictionary with arguments passed directly to a reader
implementation constructor chosen by ``reader_engine`` argument. You should not use this parameter, unless you
fine-tuning of a reader.
:return: A :class:`Reader` object
"""
if dataset_url is None or not isinstance(dataset_url, six.string_types):
raise ValueError("""dataset_url must be a string""")
dataset_url = dataset_url[:-1] if dataset_url[-1] == '/' else dataset_url
logger.debug('dataset_url: %s', dataset_url)
resolver = FilesystemResolver(dataset_url, hdfs_driver=hdfs_driver)
filesystem = resolver.filesystem()
dataset_path = resolver.get_dataset_path()
if cache_type is None or cache_type == 'null':
cache = NullCache()
elif cache_type == 'local-disk':
cache = LocalDiskCache(cache_location, cache_size_limit,
cache_row_size_estimate, **cache_extra_settings
or {})
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
# Fail if this is a non-petastorm dataset. Typically, a Parquet store will have hundred thousands rows in a single
# rowgroup. Using PyDictReaderWorker or ReaderV2 implementation is very inefficient as it processes data on a
# row by row basis. ArrowReaderWorker (used by make_batch_reader) is much more efficient in these cases.
try:
dataset_metadata.get_schema_from_dataset_url(dataset_url)
except PetastormMetadataError:
raise RuntimeError(
'Currently make_reader supports reading only Petastorm datasets. '
'To read from a non-Petastorm Parquet store use make_batch_reader')
if reader_engine == 'reader_v1':
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count, results_queue_size)
elif reader_pool_type == 'process':
if pyarrow_serialize:
serializer = PyArrowSerializer()
else:
serializer = PickleSerializer()
reader_pool = ProcessPool(workers_count, serializer)
elif reader_pool_type == 'dummy':
reader_pool = DummyPool()
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
# Create a dictionary with all ReaderV2 parameters, so we can merge with reader_engine_params if specified
kwargs = {
'schema_fields': schema_fields,
'reader_pool': reader_pool,
'shuffle_row_groups': shuffle_row_groups,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
'predicate': predicate,
'rowgroup_selector': rowgroup_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'cache': cache,
}
if reader_engine_params:
kwargs.update(reader_engine_params)
try:
return Reader(filesystem,
dataset_path,
worker_class=PyDictReaderWorker,
**kwargs)
except PetastormMetadataError as e:
logger.error('Unexpected exception: %s', str(e))
raise RuntimeError(
'make_reader has failed. If you were trying to open a Parquet store that was not '
'created using Petastorm materialize_dataset and it contains only scalar columns, '
'you may use make_batch_reader to read it.\n'
'Inner exception: %s', str(e))
elif reader_engine == 'experimental_reader_v2':
if reader_pool_type == 'thread':
decoder_pool = ThreadPoolExecutor(workers_count)
elif reader_pool_type == 'process':
decoder_pool = ProcessPoolExecutor(workers_count)
elif reader_pool_type == 'dummy':
decoder_pool = SameThreadExecutor()
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
# TODO(yevgeni): once ReaderV2 is ready to be out of experimental status, we should extend
# the make_reader interfaces to take shuffling buffer parameters explicitly
shuffling_queue = RandomShufflingBuffer(
1000, 800) if shuffle_row_groups else NoopShufflingBuffer()
# Create a dictionary with all ReaderV2 parameters, so we can merge with reader_engine_params if specified
kwargs = {
'schema_fields': schema_fields,
'predicate': predicate,
'rowgroup_selector': rowgroup_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'cache': cache,
'decoder_pool': decoder_pool,
'shuffling_queue': shuffling_queue,
'shuffle_row_groups': shuffle_row_groups,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
}
if reader_engine_params:
kwargs.update(reader_engine_params)
return ReaderV2(dataset_url, **kwargs)
else:
raise ValueError(
'Unexpected value of reader_engine argument \'%s\'. '
'Supported reader_engine values are \'reader_v1\' and \'experimental_reader_v2\'',
reader_engine)
def test_null_cache(self):
"""Testing trivial NullCache: should trigger value generating function on each run"""
cache = NullCache()
self.assertEqual(42, cache.get('some_key', lambda: 42))
def make_batch_reader(dataset_url,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
cache_type='null',
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3'):
"""
Creates an instance of Reader for reading batches out of a non-Petastorm Parquet store.
Currently, only stores having native scalar parquet data types are supported.
Use :func:`~petastorm.make_reader` to read Petastorm Parquet stores generated with
:func:`~petastorm.etl.dataset_metadata.materialize_dataset`.
NOTE: only scalar columns are currently supported.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'file:///tmp/mydataset'``
or ``'s3://bucket/mydataset'``.
:param schema_fields: A list of regex pattern strings. Only columns matching at least one of the
patterns in the list will be loaded.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a pandas DataFrame object and must return a pandas Series with boolean values of matching
dimensions.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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++)
:return: A :class:`Reader` object
"""
if dataset_url is None or not isinstance(dataset_url, six.string_types):
raise ValueError("""dataset_url must be a string""")
dataset_url = dataset_url[:-1] if dataset_url[-1] == '/' else dataset_url
logger.debug('dataset_url: %s', dataset_url)
resolver = FilesystemResolver(dataset_url, hdfs_driver=hdfs_driver)
filesystem = resolver.filesystem()
dataset_path = resolver.parsed_dataset_url().path
if cache_type is None or cache_type == 'null':
cache = NullCache()
elif cache_type == 'local-disk':
cache = LocalDiskArrowTableCache(cache_location, cache_size_limit,
cache_row_size_estimate,
**cache_extra_settings or {})
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count)
elif reader_pool_type == 'process':
serializer = ArrowTableSerializer()
reader_pool = ProcessPool(workers_count, serializer)
elif reader_pool_type == 'dummy':
reader_pool = DummyPool()
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
return Reader(filesystem,
dataset_path,
schema_fields=schema_fields,
worker_class=ArrowReaderWorker,
reader_pool=reader_pool,
shuffle_row_groups=shuffle_row_groups,
shuffle_row_drop_partitions=shuffle_row_drop_partitions,
predicate=predicate,
rowgroup_selector=rowgroup_selector,
num_epochs=num_epochs,
cur_shard=cur_shard,
shard_count=shard_count,
cache=cache)
def make_batch_reader(dataset_url_or_urls,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
shard_seed=None,
cache_type='null',
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3',
transform_spec=None,
filters=None,
storage_options=None,
zmq_copy_buffers=True,
filesystem=None):
"""
Creates an instance of Reader for reading batches out of a non-Petastorm Parquet store.
Currently, only stores having native scalar parquet data types are supported.
Use :func:`~petastorm.make_reader` to read Petastorm Parquet stores generated with
:func:`~petastorm.etl.dataset_metadata.materialize_dataset`.
NOTE: only scalar columns or array type (of primitive type element) columns are currently supported.
NOTE: If without `schema_fields` specified, the reader schema will be inferred from parquet dataset. then the
reader schema fields order will preserve parqeut dataset fields order (partition column come first), but if
setting `transform_spec` and specified `TransformSpec.selected_fields`, then the reader schema fields order
will be the order of 'selected_fields'.
:param dataset_url_or_urls: a url to a parquet directory or a url list (with the same scheme) to parquet files.
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'file:///tmp/mydataset'``,
or ``'s3://bucket/mydataset'``, or ``'gs://bucket/mydataset'``,
or ``[file:///tmp/mydataset/00000.parquet, file:///tmp/mydataset/00001.parquet]``.
:param schema_fields: A list of regex pattern strings. Only columns matching at least one of the
patterns in the list will be loaded.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a pandas DataFrame object and must return a pandas Series with boolean values of matching
dimensions.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param shard_seed: Random seed to shuffle row groups for data sharding. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 transform_spec: An instance of :class:`~petastorm.transform.TransformSpec` object defining how a record
is transformed after it is loaded and decoded. The transformation occurs on a worker thread/process (depends
on the ``reader_pool_type`` value).
:param filters: (List[Tuple] or List[List[Tuple]]): Standard PyArrow filters.
These will be applied when loading the parquet file with PyArrow. More information
here: https://arrow.apache.org/docs/python/generated/pyarrow.parquet.ParquetDataset.html
:param storage_options: Dict of kwargs forwarded to ``fsspec`` to initialize the filesystem.
:param zmq_copy_buffers: A bool indicating whether to use 0mq copy buffers with ProcessPool.
:param filesystem: An instance of ``pyarrow.FileSystem`` to use. Will ignore storage_options and
other filesystem configs if it's provided.
:return: A :class:`Reader` object
"""
dataset_url_or_urls = normalize_dataset_url_or_urls(dataset_url_or_urls)
filesystem, dataset_path_or_paths = get_filesystem_and_path_or_paths(
dataset_url_or_urls,
hdfs_driver,
storage_options=storage_options,
filesystem=filesystem)
try:
dataset_metadata.get_schema_from_dataset_url(
dataset_url_or_urls,
hdfs_driver=hdfs_driver,
storage_options=storage_options,
filesystem=filesystem)
warnings.warn(
'Please use make_reader (instead of \'make_batch_dataset\' function to read this dataset. '
'You may get unexpected results. '
'Currently make_batch_reader supports reading only Parquet stores that contain '
'standard Parquet data types and do not require petastorm decoding.'
)
except PetastormMetadataError:
pass
if cache_type is None or cache_type == NULL_CACHE:
cache = NullCache()
elif cache_type == LOCAL_DISK_CACHE:
cache = LocalDiskArrowTableCache(cache_location, cache_size_limit,
cache_row_size_estimate,
**cache_extra_settings or {})
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count)
elif reader_pool_type == 'process':
serializer = ArrowTableSerializer()
reader_pool = ProcessPool(workers_count,
serializer,
zmq_copy_buffers=zmq_copy_buffers)
elif reader_pool_type == 'dummy':
reader_pool = DummyPool()
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
return Reader(filesystem,
dataset_path_or_paths,
schema_fields=schema_fields,
worker_class=ArrowReaderWorker,
reader_pool=reader_pool,
shuffle_row_groups=shuffle_row_groups,
shuffle_row_drop_partitions=shuffle_row_drop_partitions,
predicate=predicate,
rowgroup_selector=rowgroup_selector,
num_epochs=num_epochs,
cur_shard=cur_shard,
shard_count=shard_count,
shard_seed=shard_seed,
cache=cache,
transform_spec=transform_spec,
is_batched_reader=True,
filters=filters)
def make_reader(dataset_url,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
pyarrow_serialize=False,
results_queue_size=50,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
shard_seed=None,
cache_type=NULL_CACHE,
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3',
transform_spec=None,
filters=None,
storage_options=None,
zmq_copy_buffers=True,
filesystem=None):
"""
Creates an instance of Reader for reading Petastorm datasets. A Petastorm dataset is a dataset generated using
:func:`~petastorm.etl.dataset_metadata.materialize_dataset` context manager as explained
`here <https://petastorm.readthedocs.io/en/latest/readme_include.html#generating-a-dataset>`_.
See :func:`~petastorm.make_batch_reader` to read from a Parquet store that was not generated using
:func:`~petastorm.etl.dataset_metadata.materialize_dataset`.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'file:///tmp/mydataset'``,
or ``'s3://bucket/mydataset'``, or ``'gs://bucket/mydataset'``.
:param schema_fields: Can be: a list of unischema fields and/or regex pattern strings; ``None`` to read all fields;
an NGram object, then it will return an NGram of the specified fields.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param pyarrow_serialize: THE ARGUMENT IS DEPRECATED AND WILL BE REMOVED IN FUTURE VERSIONS.
:param results_queue_size: Size of the results queue to store prefetched row-groups. Currently only applicable to
thread reader pool type.
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a single row and must return a boolean value indicating whether to include it in the results.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param shard_seed: Random seed to shuffle row groups for data sharding. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 transform_spec: An instance of :class:`~petastorm.transform.TransformSpec` object defining how a record
is transformed after it is loaded and decoded. The transformation occurs on a worker thread/process (depends
on the ``reader_pool_type`` value).
:param filters: (List[Tuple] or List[List[Tuple]]): Standard PyArrow filters.
These will be applied when loading the parquet file with PyArrow. More information
here: https://arrow.apache.org/docs/python/generated/pyarrow.parquet.ParquetDataset.html
:param storage_options: Dict of kwargs forwarded to ``fsspec`` to initialize the filesystem.
:param zmq_copy_buffers: A bool indicating whether to use 0mq copy buffers with ProcessPool.
:param filesystem: An instance of ``pyarrow.FileSystem`` to use. Will ignore storage_options and
other filesystem configs if it's provided.
:return: A :class:`Reader` object
"""
dataset_url = normalize_dir_url(dataset_url)
filesystem, dataset_path = get_filesystem_and_path_or_paths(
dataset_url,
hdfs_driver,
storage_options=storage_options,
filesystem=filesystem)
if cache_type is None or cache_type == NULL_CACHE:
cache = NullCache()
elif cache_type == LOCAL_DISK_CACHE:
cache = LocalDiskCache(cache_location, cache_size_limit,
cache_row_size_estimate, **cache_extra_settings
or {})
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
try:
dataset_metadata.get_schema_from_dataset_url(
dataset_url,
hdfs_driver=hdfs_driver,
storage_options=storage_options,
filesystem=filesystem)
except PetastormMetadataError:
warnings.warn(
'Currently make_reader supports reading only Petastorm datasets. '
'To read from a non-Petastorm Parquet store use make_batch_reader')
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count, results_queue_size)
elif reader_pool_type == 'process':
if pyarrow_serialize:
warnings.warn(
"pyarrow_serializer was deprecated and will be removed in future versions. "
"The argument no longer has any effect.")
serializer = PickleSerializer()
reader_pool = ProcessPool(workers_count,
serializer,
zmq_copy_buffers=zmq_copy_buffers)
elif reader_pool_type == 'dummy':
reader_pool = DummyPool()
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
kwargs = {
'schema_fields': schema_fields,
'reader_pool': reader_pool,
'shuffle_row_groups': shuffle_row_groups,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
'predicate': predicate,
'rowgroup_selector': rowgroup_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'shard_seed': shard_seed,
'cache': cache,
'transform_spec': transform_spec,
'filters': filters
}
try:
return Reader(filesystem,
dataset_path,
worker_class=PyDictReaderWorker,
is_batched_reader=False,
**kwargs)
except PetastormMetadataError as e:
logger.error('Unexpected exception: %s', str(e))
raise RuntimeError(
'make_reader has failed. If you were trying to open a Parquet store that was not '
'created using Petastorm materialize_dataset and it contains only scalar columns, '
'you may use make_batch_reader to read it.\n'
'Inner exception: %s', str(e))
def make_reader(dataset_url,
schema_fields=None,
reader_pool_type='thread', workers_count=10, pyarrow_serialize=False,
shuffle_row_groups=True, shuffle_row_drop_partitions=1,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
cur_shard=None, shard_count=None,
cache_type='null', cache_location=None, cache_size_limit=None,
cache_row_size_estimate=None, cache_extra_settings=None,
hdfs_driver='libhdfs3',
infer_schema=False,
reader_engine='reader_v1', reader_engine_params=None):
"""
Factory convenience method for :class:`Reader`.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'file:///tmp/mydataset'``
or ``'s3://bucket/mydataset'``.
:param schema_fields: Either list of unischema fields to subset, or ``None`` to read all fields.
OR an NGram object, then it will return an NGram of the specified properties.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param pyarrow_serialize: Whether to use pyarrow for serialization. Currently only applicable to process pool.
Defaults to False.
:param shuffle_row_groups: Whether to shuffle row groups (the order in which full row groups are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a row group into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each row group read will drop half of the rows within every row group and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 infer_schema: Whether to infer the unischema object from the parquet schema.
Only works for schemas containing certain scalar type. This option allows getting around explicitly
generating petastorm metadata using :func:`petastorm.etl.dataset_metadata.materialize_dataset` or
petastorm-generate-metadata.py
:param reader_engine: Multiple engine implementations exist ('reader_v1' and 'experimental_reader_v2'). 'reader_v1'
(the default value) selects a stable reader implementation.
:param reader_engine_params: For advanced usage: a dictionary with arguments passed directly to a reader
implementation constructor chosen by ``reader_engine`` argument. You should not use this parameter, unless you
fine-tuning of a reader.
:return: A :class:`Reader` object
"""
if dataset_url is None or not isinstance(dataset_url, six.string_types):
raise ValueError("""dataset_url must be a string""")
dataset_url = dataset_url[:-1] if dataset_url[-1] == '/' else dataset_url
logger.debug('dataset_url: %s', dataset_url)
resolver = FilesystemResolver(dataset_url, hdfs_driver=hdfs_driver)
filesystem = resolver.filesystem()
dataset_path = resolver.get_dataset_path()
if cache_type is None or cache_type == 'null':
cache = NullCache()
elif cache_type == 'local-disk':
cache = LocalDiskCache(cache_location, cache_size_limit, cache_row_size_estimate, **cache_extra_settings or {})
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
if reader_engine == 'reader_v1':
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count)
elif reader_pool_type == 'process':
reader_pool = ProcessPool(workers_count, pyarrow_serialize=pyarrow_serialize)
elif reader_pool_type == 'dummy':
reader_pool = DummyPool()
else:
raise ValueError('Unknown reader_pool_type: {}'.format(reader_pool_type))
# Create a dictionary with all ReaderV2 parameters, so we can merge with reader_engine_params if specified
kwargs = {
'schema_fields': schema_fields,
'reader_pool': reader_pool,
'shuffle_row_groups': shuffle_row_groups,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
'predicate': predicate,
'rowgroup_selector': rowgroup_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'cache': cache,
'infer_schema': infer_schema,
}
if reader_engine_params:
kwargs.update(reader_engine_params)
return Reader(filesystem, dataset_path, **kwargs)
elif reader_engine == 'experimental_reader_v2':
if reader_pool_type == 'thread':
decoder_pool = ThreadPoolExecutor(workers_count)
elif reader_pool_type == 'process':
decoder_pool = ProcessPoolExecutor(workers_count)
elif reader_pool_type == 'dummy':
decoder_pool = SameThreadExecutor()
else:
raise ValueError('Unknown reader_pool_type: {}'.format(reader_pool_type))
# TODO(yevgeni): once ReaderV2 is ready to be out of experimental status, we should extend
# the make_reader interfaces to take shuffling buffer parameters explicitly
shuffling_queue = RandomShufflingBuffer(1000, 800) if shuffle_row_groups else NoopShufflingBuffer()
# Create a dictionary with all ReaderV2 parameters, so we can merge with reader_engine_params if specified
kwargs = {
'schema_fields': schema_fields,
'predicate': predicate,
'rowgroup_selector': rowgroup_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'cache': cache,
'decoder_pool': decoder_pool,
'shuffling_queue': shuffling_queue,
'shuffle_row_groups': shuffle_row_groups,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
'infer_schema': infer_schema,
}
if reader_engine_params:
kwargs.update(reader_engine_params)
return ReaderV2(dataset_url, **kwargs)
else:
raise ValueError('Unexpected value of reader_engine argument \'%s\'. '
'Supported reader_engine values are \'reader_v1\' and \'experimental_reader_v2\'',
reader_engine)
def __init__(self,
dataset_url,
schema_fields=None,
shuffle=None,
predicate=None,
rowgroup_selector=None,
reader_pool=None,
num_epochs=1,
sequence=None,
training_partition=None,
num_training_partitions=None,
read_timeout_s=None,
cache=None,
shuffle_options=None):
"""Initializes a reader object.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/parquet8'``, or ``'/tmp/mydataset'``.
:param schema_fields: Either list of unischema fields to subset, or ``None`` to read all fields.
OR an NGram object, then it will return an NGram of the specified properties.
:param predicate: instance of predicate object to filter rows to be returned by reader.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param reader_pool: parallelization pool. ``ThreadPool(10)`` (10 threads) is used by default.
This pool is a custom implementation used to parallelize reading data from the dataset.
Any object from workers_pool package can be used
(e.g. :class:`petastorm.workers_pool.process_pool.ProcessPool`).
:param num_epochs: An epoch is a single pass over all samples in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param training_partition: An int denoting the partition number used for multi node training. Each node should
pass in a unique partition number in the range ``[0, num_training_partitions)``.
``num_training_partitions`` must be supplied as well.
:param num_training_partitions: An int denoting the number of training partitions (how many nodes are performing
the multi node training).
:param read_timeout_s: A numeric with the amount of time in seconds you would like to give a read before it
times out and raises an EmptyResultError. Pass in None for an infinite timeout.
:param cache: An object conforming to :class:`.CacheBase` interface. Before loading row groups from a parquet
file the Reader will attempt to load these values from cache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if num_training_partitions is used).
By default, use the :class:`.NullCache` implementation.
:param shuffle_options: ShuffleOptions object to describe how to shuffle dataset (supercedes shuffle parameter)
defaults to shuffling row groups but not to drop rows based on partitions.
:param sequence: *DEPRECATED* To use sequence/ngram, please supply the argument in
``schema_fields`` instead.
:param shuffle: *DEPRECATED* Boolean whether to shuffle the row group order.
Use ``shuffle_row_groups`` in :class:`.ShuffleOptions` instead.
"""
# 1. Resolve dataset path (hdfs://, file://) and open the parquet storage (dataset)
# 2. Get a list of all groups
# 3. Filter rowgroups
# a. predicates
# b. row-group selector (our indexing mechanism)
# c. partition: used to get a subset of data for distributed training
# 4. Create a rowgroup ventilator object
# 5. Start workers pool
if dataset_url is None or not isinstance(dataset_url,
six.string_types):
raise ValueError("""dataset_url must be a string""")
if not (isinstance(schema_fields, collections.Iterable)
or isinstance(schema_fields, NGram) or schema_fields is None):
raise ValueError(
"""Fields must be either None, an iterable collection of Unischema fields or an NGram
object.""")
if sequence is not None:
raise ValueError(
"""'sequence' argument of Reader object is deprecated. Please pass an NGram instance to
'schema_fields' argument instead.""")
self.ngram = schema_fields if isinstance(schema_fields,
NGram) else None
if self.ngram and not self.ngram.timestamp_overlap and shuffle_options.shuffle_row_drop_partitions > 1:
raise NotImplementedError(
'Using timestamp_overlap=False is not implemented with'
' shuffle_options.shuffle_row_drop_partitions > 1')
cache = cache or NullCache()
dataset_url = dataset_url[:-1] if dataset_url[
-1] == '/' else dataset_url
self._workers_pool = reader_pool or ThreadPool(10)
# 1. Resolve dataset path (hdfs://, file://) and open the parquet storage (dataset)
logger.debug('dataset_url: %s', dataset_url)
resolver = FilesystemResolver(dataset_url)
self.dataset = pq.ParquetDataset(resolver.parsed_dataset_url().path,
filesystem=resolver.filesystem(),
validate_schema=False)
# Get a unischema stored in the dataset metadata.
stored_schema = dataset_metadata.get_schema(self.dataset)
# Make a schema view (a view is a Unischema containing only a subset of fields
# Will raise an exception if invalid schema fields are in schema_fields
fields = schema_fields if isinstance(schema_fields,
collections.Iterable) else None
self.schema = stored_schema.create_schema_view(
fields) if fields else stored_schema
# 2. Get a list of all groups
row_groups = dataset_metadata.load_row_groups(self.dataset)
# 3. Filter rowgroups
filtered_row_group_indexes, worker_predicate = self._filter_row_groups(
self.dataset, row_groups, predicate, rowgroup_selector,
training_partition, num_training_partitions)
# 4. Create a rowgroup ventilator object
if shuffle_options is None:
if shuffle is None:
shuffle = True
else:
logger.warning(
'shuffle option is deprecated. Please use shuffle_options instead'
)
shuffle_options = ShuffleOptions(shuffle)
self._normalize_shuffle_options(shuffle_options, self.dataset)
ventilator = self._create_ventilator(filtered_row_group_indexes,
shuffle_options, num_epochs,
worker_predicate)
# 5. Start workers pool
self._workers_pool.start(ReaderWorker,
(dataset_url, self.schema, self.ngram,
row_groups, cache, worker_predicate),
ventilator=ventilator)
self._read_timeout_s = read_timeout_s
def make_carbon_reader(dataset_url,
key=None,
secret=None,
endpoint=None,
proxy=None,
proxy_port=None,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
results_queue_size=100,
shuffle_blocklets=True,
shuffle_row_drop_partitions=1,
predicate=None,
blocklet_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
cache_type='null',
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3',
reader_engine='reader_v1',
reader_engine_params=None,
transform_spec=None):
"""
Creates an instance of Reader for reading Pycarbon datasets. A Pycarbon dataset is a dataset generated using
:func:`~pycarbon.etl.carbon_dataset_metadata.materialize_dataset_carbon` context manager as explained
See :func:`~pycarbon.make_batch_carbon_reader` to read from a Carbon store that was not generated using
:func:`~pycarbon.etl.carbon_dataset_metadata.materialize_dataset_carbon`.
:param dataset_url: an filepath or a url to a carbon directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/carbon8'``, or ``'file:///tmp/mydataset'``
or ``'s3://bucket/mydataset'``.
:param key: access key
:param secret: secret key
:param endpoint: endpoint_url
:param proxy: proxy
:param proxy_port: proxy_port
:param schema_fields: Can be: a list of unischema fields and/or regex pattern strings; ``None`` to read all fields;
an NGram object, then it will return an NGram of the specified fields.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
TODO: process support
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param results_queue_size: Size of the results queue to store prefetched rows. Currently only applicable to
thread reader pool type.
:param shuffle_blocklets: Whether to shuffle blocklets (the order in which full blocklets are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a blocklet into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each blocklet read will drop half of the rows within every blocklet and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a single row and must return a boolean value indicating whether to include it in the results.
:param blocklet_selector: instance of blocklet selector object to select blocklet to be read
TODO: blocklet_selector
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
TODO: cur_shard & shard_count
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 reader_engine: Multiple engine implementations exist ('reader_v1' and 'experimental_reader_v2'). 'reader_v1'
(the default value) selects a stable reader implementation.
TODO: experimental_reader_v2 for carbon
:param reader_engine_params: For advanced usage: a dictionary with arguments passed directly to a reader
implementation constructor chosen by ``reader_engine`` argument. You should not use this parameter, unless you
fine-tuning of a reader.
:param transform_spec: An instance of :class:`~petastorm.transform.TransformSpec` object defining how a record
is transformed after it is loaded and decoded. The transformation occurs on a worker thread/process (depends
on the ``reader_pool_type`` value).
:return: A :class:`Reader` object
"""
if dataset_url is None or not isinstance(dataset_url, six.string_types):
raise ValueError("""dataset_url must be a string""")
dataset_url = dataset_url[:-1] if dataset_url[-1] == '/' else dataset_url
logger.debug('dataset_url: %s', dataset_url)
resolver = CarbonFilesystemResolver(dataset_url,
key=key,
secret=secret,
endpoint=endpoint,
proxy=proxy,
proxy_port=proxy_port,
hdfs_driver=hdfs_driver)
filesystem = resolver.filesystem()
if cache_type is None or cache_type == 'null':
cache = NullCache()
elif cache_type == 'local-disk':
cache = LocalDiskCache(cache_location, cache_size_limit,
cache_row_size_estimate, **cache_extra_settings
or {})
elif cache_type == 'memory-cache':
cache = LocalMemoryCache(cache_size_limit)
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
# Fail if this is a non-pycarbon dataset. Typically, a Carbon store will have hundred thousands rows in a single
# blocklet. Using PyDictCarbonReaderWorker or ReaderV2 implementation is very inefficient as it processes data on a
# row by row basis. ArrowCarbonReaderWorker (used by make_batch_carbon_reader) is much more efficient in these cases.
try:
carbon_dataset_metadata.get_schema_from_dataset_url_carbon(
dataset_url,
key=key,
secret=secret,
endpoint=endpoint,
proxy=proxy,
proxy_port=proxy_port,
filesystem=filesystem)
except PycarbonMetadataError:
raise RuntimeError(
'Currently make_carbon_reader supports reading only Pycarbon datasets(has unischema). '
'To read from a non-Pycarbon Carbon store use make_batch_carbon_reader'
)
if reader_engine == 'reader_v1':
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count, results_queue_size)
elif reader_pool_type == 'process':
raise NotImplementedError(
'not support process reader_pool_type now.')
elif reader_pool_type == 'dummy':
raise NotImplementedError(
'not support dummy reader_pool_type now.')
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
# Create a dictionary with all ReaderV2 parameters, so we can merge with reader_engine_params if specified
kwargs = {
'key': key,
'secret': secret,
'endpoint': endpoint,
'proxy': proxy,
'proxy_port': proxy_port,
'schema_fields': schema_fields,
'reader_pool': reader_pool,
'shuffle_blocklets': shuffle_blocklets,
'shuffle_row_drop_partitions': shuffle_row_drop_partitions,
'predicate': predicate,
'blocklet_selector': blocklet_selector,
'num_epochs': num_epochs,
'cur_shard': cur_shard,
'shard_count': shard_count,
'cache': cache,
'transform_spec': transform_spec,
}
if reader_engine_params:
kwargs.update(reader_engine_params)
try:
return CarbonDataReader(filesystem,
dataset_url,
worker_class=PyDictCarbonReaderWorker,
**kwargs)
except PycarbonMetadataError as e:
logger.error('Unexpected exception: %s', str(e))
raise RuntimeError(
'make_carbon_reader has failed. If you were trying to open a Carbon store that was not '
'created using Pycarbon materialize_dataset_carbon and it contains only scalar columns, '
'you may use make_batch_reader to read it.\n'
'Inner exception: %s', str(e))
elif reader_engine == 'experimental_reader_v2':
raise NotImplementedError(
'not support experimental_reader_v2 reader engine now.')
else:
raise ValueError(
'Unexpected value of reader_engine argument \'%s\'. '
'Supported reader_engine values are \'reader_v1\' and \'experimental_reader_v2\'',
reader_engine)
def make_batch_carbon_reader(dataset_url,
key=None,
secret=None,
endpoint=None,
proxy=None,
proxy_port=None,
schema_fields=None,
reader_pool_type='thread',
workers_count=10,
results_queue_size=100,
shuffle_blocklets=True,
shuffle_row_drop_partitions=1,
predicate=None,
blocklet_selector=None,
num_epochs=1,
cur_shard=None,
shard_count=None,
cache_type='null',
cache_location=None,
cache_size_limit=None,
cache_row_size_estimate=None,
cache_extra_settings=None,
hdfs_driver='libhdfs3',
transform_spec=None):
"""
Creates an instance of Reader for reading batches out of a non-Pycarbon Carbon store.
Currently, only stores having native scalar carbon data types are supported.
Use :func:`~pycarbon.make_carbon_reader` to read Pycarbon Carbon stores generated with
:func:`~pycarbon.etl.carbon_dataset_metadata.materialize_dataset_carbon`.
NOTE: only scalar columns are currently supported.
:param dataset_url: an filepath or a url to a carbon directory,
e.g. ``'hdfs://some_hdfs_cluster/user/yevgeni/carbon8'``, or ``'file:///tmp/mydataset'``
or ``'s3://bucket/mydataset'``.
:param key: access key
:param secret: secret key
:param endpoint: endpoint_url
:param proxy: proxy
:param proxy_port: proxy_port
:param schema_fields: A list of regex pattern strings. Only columns matching at least one of the
patterns in the list will be loaded.
:param reader_pool_type: A string denoting the reader pool type. Should be one of ['thread', 'process', 'dummy']
denoting a thread pool, process pool, or running everything in the master thread. Defaults to 'thread'
:param workers_count: An int for the number of workers to use in the reader pool. This only is used for the
thread or process pool. Defaults to 10
:param results_queue_size: Size of the results queue to store prefetched rows. Currently only applicable to
thread reader pool type.
:param shuffle_blocklets: Whether to shuffle blocklets (the order in which full blocklets are read)
:param shuffle_row_drop_partitions: This is is a positive integer which determines how many partitions to
break up a blocklet into for increased shuffling in exchange for worse performance (extra reads).
For example if you specify 2 each blocklet read will drop half of the rows within every blocklet and
read the remaining rows in separate reads. It is recommended to keep this number below the regular row
group size in order to not waste reads which drop all rows.
:param predicate: instance of :class:`.PredicateBase` object to filter rows to be returned by reader. The predicate
will be passed a pandas DataFrame object and must return a pandas Series with boolean values of matching
dimensions.
:param blocklet_selector: instance of blocklet selector object to select blocklets to be read
:param num_epochs: An epoch is a single pass over all rows in the dataset. Setting ``num_epochs`` to
``None`` will result in an infinite number of epochs.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count). shard_count must be supplied as well.
Defaults to None
:param shard_count: An int denoting the number of shards to break this dataset into. Defaults to None
:param cache_type: A string denoting the cache type, if desired. Options are [None, 'null', 'local-disk'] to
either have a null/noop cache or a cache implemented using diskcache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if shard_count is used). By default will be a null cache.
:param cache_location: A string denoting the location or path of the cache.
:param cache_size_limit: An int specifying the size limit of the cache in bytes
:param cache_row_size_estimate: An int specifying the estimated size of a row in the dataset
:param cache_extra_settings: A dictionary of extra settings to pass to the cache implementation,
: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 transform_spec: An instance of :class:`~petastorm.transform.TransformSpec` object defining how a record
is transformed after it is loaded and decoded. The transformation occurs on a worker thread/process (depends
on the ``reader_pool_type`` value).
:return: A :class:`Reader` object
"""
if dataset_url is None or not isinstance(dataset_url, six.string_types):
raise ValueError("""dataset_url must be a string""")
dataset_url = dataset_url[:-1] if dataset_url[-1] == '/' else dataset_url
logger.debug('dataset_url: %s', dataset_url)
resolver = CarbonFilesystemResolver(dataset_url,
key=key,
secret=secret,
endpoint=endpoint,
proxy=proxy,
proxy_port=proxy_port,
hdfs_driver=hdfs_driver)
filesystem = resolver.filesystem()
try:
carbon_dataset_metadata.get_schema_from_dataset_url_carbon(
dataset_url,
key=key,
secret=secret,
endpoint=endpoint,
proxy=proxy,
proxy_port=proxy_port,
filesystem=filesystem)
warnings.warn(
'Please use make_carbon_reader (instead of \'make_batch_carbon_reader\' function '
'to read this dataset as it contains unischema file.')
except PycarbonMetadataError:
pass
if cache_type is None or cache_type == 'null':
cache = NullCache()
elif cache_type == 'local-disk':
cache = LocalDiskCache(cache_location, cache_size_limit,
cache_row_size_estimate, **cache_extra_settings
or {})
elif cache_type == 'memory-cache':
cache = LocalMemoryCache(cache_size_limit)
else:
raise ValueError('Unknown cache_type: {}'.format(cache_type))
if reader_pool_type == 'thread':
reader_pool = ThreadPool(workers_count, results_queue_size)
elif reader_pool_type == 'process':
raise NotImplementedError('not support process reader_pool_type now.')
elif reader_pool_type == 'dummy':
raise NotImplementedError('not support dummy reader_pool_type now.')
else:
raise ValueError(
'Unknown reader_pool_type: {}'.format(reader_pool_type))
return CarbonDataReader(
filesystem,
dataset_url,
key=key,
secret=secret,
endpoint=endpoint,
proxy=proxy,
proxy_port=proxy_port,
schema_fields=schema_fields,
worker_class=ArrowCarbonReaderWorker,
reader_pool=reader_pool,
shuffle_blocklets=shuffle_blocklets,
shuffle_row_drop_partitions=shuffle_row_drop_partitions,
predicate=predicate,
blocklet_selector=blocklet_selector,
num_epochs=num_epochs,
cur_shard=cur_shard,
shard_count=shard_count,
cache=cache,
transform_spec=transform_spec)
def __init__(self,
dataset_url,
schema_fields=None,
predicate=None,
rowgroup_selector=None,
num_epochs=1,
sequence=None,
cur_shard=None,
shard_count=None,
read_timeout_s=None,
cache=None,
loader_pool=None,
decoder_pool=None,
shuffling_queue=None,
shuffle_row_groups=True,
shuffle_row_drop_partitions=1,
pyarrow_filesystem=None,
hdfs_driver='libhdfs3'):
"""Initializes a reader object.
:param dataset_url: an filepath or a url to a parquet directory,
e.g. 'hdfs://some_hdfs_cluster/user/yevgeni/parquet8', or '/tmp/mydataset'
or ``'s3://bucket/mydataset'``.
:param schema_fields:
Either list of unischema fields to subset, or None to read all fields.
OR an NGram object, then it will return an NGram of the specified properties.
:param predicate: instance of predicate object to filter rows to be returned by reader.
:param rowgroup_selector: instance of row group selector object to select row groups to be read
:param reader_pool: parallelization pool. ThreadPool(10) (10 threads) is used by default.
This pool is a custom implementation used to parallelize reading data from the dataset.
Any object from workers_pool package can be used (e.g. ProcessPool)
:param num_epochs: An epoch is a single pass over all samples in the dataset. Setting num_epochs to 'None' will
result in an infinite number of epochs.
:param sequence: This is deprecated. To use sequence/ngram, please supply the argument in schema_fields instead.
:param cur_shard: An int denoting the current shard number. Each node reading a shard should
pass in a unique shard number in the range [0, shard_count).
shard count must be supplied as well.
:param shard_count An int denoting the number of shards to break this dataset into.
:param read_timeout_s: A numeric with the amount of time in seconds you would like to give a read before it
times out and raises an EmptyResultError. Pass in None for an infinite timeout
:param cache: An object conforming to `cache.CacheBase` interface. Before loading row groups from a parquet file
the Reader will attempt to load these values from cache. Caching is useful when communication
to the main data store is either slow or expensive and the local machine has large enough storage
to store entire dataset (or a partition of a dataset if num_training_partitions is used).
:param decoder_pool: An instance of a concurrent.futures pool executor used for decoding. If None,
a default ThreadPoolExecutor(5) will be used.
:param loader_pool: An instance of a concurrent.futures pool executor used for decoding. If None,
a default ThreadPoolExecutor(5) will be used.
: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++)
By default, `NullCache` implementation
"""
# 1. Resolve dataset path (hdfs://, file://) and open the parquet storage (dataset)
# 2. Get a list of all groups
# 3. Filter rowgroups
# a. predicates
# b. row-group selector (our indexing mechanism)
# c. partition: used to get a subset of data for distributed training
# 4. Launch a new thread running `worker_loop` function.
if dataset_url is None or not isinstance(dataset_url,
six.string_types):
raise ValueError("""dataset_url must be a string""")
if not (isinstance(schema_fields, collections.Iterable)
or isinstance(schema_fields, NGram) or schema_fields is None):
raise ValueError(
"""Fields must be either None, an iterable collection of Unischema fields or an NGram
object.""")
if sequence is not None:
raise ValueError(
"""'sequence' argument of Reader object is deprecated. Please pass an NGram instance to
'schema_fields' argument instead.""")
# Can not rely on a check in epochs.py since it runs on a separate thread. Inform user earlier about invalid
# argument value.
if num_epochs is not None and (not isinstance(num_epochs, int)
or num_epochs < 1):
raise ValueError('iterations must be positive integer or None')
self.ngram = schema_fields if isinstance(schema_fields,
NGram) else None
if self.ngram and not self.ngram.timestamp_overlap and shuffle_row_drop_partitions > 1:
raise NotImplementedError(
'Using timestamp_overlap=False is not implemented with'
' shuffle_options.shuffle_row_drop_partitions > 1')
cache = cache or NullCache()
dataset_url = dataset_url[:-1] if dataset_url[
-1] == '/' else dataset_url
# 1. Resolve dataset path (hdfs://, file://) and open the parquet storage (dataset)
logger.debug('dataset_url: %s', dataset_url)
if pyarrow_filesystem is not None:
filesystem = pyarrow_filesystem
dataset_path = urlparse(dataset_url).path
else:
resolver = FilesystemResolver(dataset_url)
filesystem = resolver.filesystem()
dataset_path = resolver.get_dataset_path()
self._dataset = pq.ParquetDataset(dataset_path,
filesystem=filesystem,
validate_schema=False)
shuffle_row_drop_partitions = self._normalize_shuffle_options(
shuffle_row_drop_partitions, self._dataset)
stored_schema = infer_or_load_unischema(self._dataset)
# Make a schema view (a view is a Unischema containing only a subset of fields
# Will raise an exception if invalid schema fields are in schema_fields
fields = schema_fields if isinstance(schema_fields,
collections.Iterable) else None
self.schema = stored_schema.create_schema_view(
fields) if fields else stored_schema
# 2. Get a list of all groups
row_groups = dataset_metadata.load_row_groups(self._dataset)
# 3. Filter rowgroups
filtered_row_groups, worker_predicate = self._filter_row_groups(
self._dataset, row_groups, predicate, rowgroup_selector, cur_shard,
shard_count)
epoch_items = self._apply_row_drop_partition(
filtered_row_groups, shuffle_row_drop_partitions)
# 4. Launch a new thread running `worker_loop` function.
def epochs_iterator():
return epoch_generator(epoch_items, num_epochs, shuffle_row_groups)
self._results_queue = Queue(_OUTPUT_QUEUE_SIZE)
loader = RowGroupLoader(dataset_url,
self.schema,
self.ngram,
cache,
worker_predicate,
hdfs_driver=hdfs_driver)
decoder = RowDecoder(self.schema, self.ngram)
self._loader_pool = loader_pool or ThreadPoolExecutor(5)
self._decoder_pool = decoder_pool or ThreadPoolExecutor(5)
self._stop_flow_manager_event = threading.Event()
self._diags = Counter()
if not shuffling_queue:
shuffling_queue = NoopShufflingBuffer()
self._flow_manager_thread = threading.Thread(
target=worker_loop,
args=(epochs_iterator, self._loader_pool, loader,
self._decoder_pool, decoder, shuffling_queue,
self._results_queue, self._stop_flow_manager_event,
self._diags))
self._flow_manager_thread.daemon = True
self._flow_manager_thread.start()
self._read_timeout_s = read_timeout_s
self.batched_output = False
