Reading and Writing Single Files#

The functions read_table() and write_table() read and write the pyarrow.Table object, respectively.

Let’s look at a simple table:

>>> import numpy as np
>>> import pandas as pd
>>> import pyarrow as pa
>>> df = pd.DataFrame({'one': [-1, np.nan, 2.5],
...                    'two': ['foo', 'bar', 'baz'],
...                    'three': [True, False, True]},
...                    index=list('abc'))
>>> table = pa.Table.from_pandas(df)

We write this to Parquet format with write_table:

>>> import pyarrow.parquet as pq
>>> pq.write_table(table, 'example.parquet')

This creates a single Parquet file. In practice, a Parquet dataset may consist of many files in many directories. We can read a single file back with read_table:

>>> table2 = pq.read_table('example.parquet')
>>> table2.to_pandas()
   one  two  three
a -1.0  foo   True
b  NaN  bar  False
c  2.5  baz   True

You can pass a subset of columns to read, which can be much faster than reading the whole file (due to the columnar layout):

>>> pq.read_table('example.parquet', columns=['one', 'three'])
pyarrow.Table
one: double
three: bool
----
one: [[-1,null,2.5]]
three: [[true,false,true]]

When reading a subset of columns from a file that used a Pandas dataframe as the source, we use read_pandas to maintain any additional index column data:

>>> pq.read_pandas('example.parquet', columns=['two']).to_pandas()
   two
a  foo
b  bar
c  baz

We do not need to use a string to specify the origin of the file. It can be any of:

A file path as a string
A NativeFile from PyArrow
A Python file object

In general, a Python file object will have the worst read performance, while a string file path or an instance of NativeFile (especially memory maps) will perform the best.

Reading Parquet and Memory Mapping#

Because Parquet data needs to be decoded from the Parquet format and compression, it can’t be directly mapped from disk. Thus the memory_map option might perform better on some systems but won’t help much with resident memory consumption.

>>> pq_array = pa.parquet.read_table(path, memory_map=True)
>>> print("RSS: {}MB".format(pa.total_allocated_bytes() >> 20))
RSS: 4299MB

>>> pq_array = pa.parquet.read_table(path, memory_map=False)
>>> print("RSS: {}MB".format(pa.total_allocated_bytes() >> 20))
RSS: 4299MB

If you need to deal with Parquet data bigger than memory, the Tabular Datasets and partitioning is probably what you are looking for.

Parquet file writing options#

write_table() has a number of options to control various settings when writing a Parquet file.

version, the Parquet format version to use. '1.0' ensures compatibility with older readers, while '2.4' and greater values enable more Parquet types and encodings.
data_page_size, to control the approximate size of encoded data pages within a column chunk. This currently defaults to 1MB.
max_rows_per_page, to cap the number of rows per data page within a column chunk (default 20000). Smaller values reduce memory usage during reads at the cost of more page metadata.
flavor, to set compatibility options particular to a Parquet consumer like 'spark' for Apache Spark.
store_decimal_as_integer, to store decimals with precision 1–18 as int32 or int64 instead of fixed_len_byte_array. This produces more compact files but may not be supported by all readers.
write_time_adjusted_to_utc, to mark TIME columns as adjusted to UTC (isAdjustedToUTC=True). When False (the default), the time is treated as local/unknown timezone.
write_page_index, to write statistics to the page index instead of writing it to each data page header. Note that PyArrow does not yet use the page index on the read side.
write_page_checksum, to write a page checksum. Use with page_checksum_verification=True on read to detect data corruption.
sorting_columns, to record the sort order of the data in each row group’s metadata. The writer does not sort the data nor does it verify that the data is sorted. Readers can use this metadata to optimize queries.

Sort order is expressed as a sequence of SortingColumn objects.

See the write_table() docstring for more details.

There are some additional data type handling-specific options described below.

Omitting the DataFrame index#

When using pa.Table.from_pandas to convert to an Arrow table, by default one or more special columns are added to keep track of the index (row labels). Storing the index takes extra space, so if your index is not valuable, you may choose to omit it by passing preserve_index=False

>>> df = pd.DataFrame({'one': [-1, np.nan, 2.5],
...                    'two': ['foo', 'bar', 'baz'],
...                    'three': [True, False, True]},
...                    index=list('abc'))
>>> table = pa.Table.from_pandas(df, preserve_index=False)

Then we have:

>>> pq.write_table(table, 'example_noindex.parquet')
>>> t = pq.read_table('example_noindex.parquet')
>>> t.to_pandas()
   one  two  three
0 -1.0  foo   True
1  NaN  bar  False
2  2.5  baz   True

Here you see the index did not survive the round trip.

Finer-grained Reading and Writing#

read_table uses the ParquetFile class, which has other features:

>>> parquet_file = pq.ParquetFile('example.parquet')
>>> parquet_file.metadata
<pyarrow._parquet.FileMetaData object at ...>
  created_by: parquet-cpp-arrow version ...
  num_columns: 4
  num_rows: 3
  num_row_groups: 1
  format_version: 2.6
  serialized_size: ...
>>> parquet_file.schema
<pyarrow._parquet.ParquetSchema object at ...>
required group field_id=-1 schema {
  optional double field_id=-1 one;
  optional binary field_id=-1 two (String);
  optional boolean field_id=-1 three;
  optional binary field_id=-1 __index_level_0__ (String);
}

As you can learn more in the Apache Parquet format, a Parquet file consists of multiple row groups. read_table will read all of the row groups and concatenate them into a single table. You can read individual row groups with read_row_group:

>>> parquet_file.num_row_groups
1
>>> parquet_file.read_row_group(0)
pyarrow.Table
one: double
two: large_string
three: bool
__index_level_0__: large_string
----
one: [[-1,null,2.5]]
two: [["foo","bar","baz"]]
three: [[true,false,true]]
__index_level_0__: [["a","b","c"]]

We can similarly write a Parquet file with multiple row groups by using ParquetWriter:

>>> with pq.ParquetWriter('example2.parquet', table.schema) as writer:
...     for i in range(3):
...         writer.write_table(table)
>>> pf2 = pq.ParquetFile('example2.parquet')
>>> pf2.num_row_groups
3

For memory-efficient reads of large files, iter_batches() streams the file as a sequence of RecordBatch objects rather than loading the entire file into a single table:

>>> parquet_file = pq.ParquetFile('example.parquet')
>>> for batch in parquet_file.iter_batches(batch_size=2):
...     print(batch.num_rows)
2
1

The batch_size parameter controls the maximum number of rows per batch. The row_groups and columns parameters allow limiting which row groups and columns are read.

Inspecting the Parquet File Metadata#

The FileMetaData of a Parquet file can be accessed through ParquetFile as shown above:

>>> parquet_file = pq.ParquetFile('example.parquet')
>>> metadata = parquet_file.metadata
>>> metadata
<pyarrow._parquet.FileMetaData object at ...>
  created_by: parquet-cpp-arrow version ...
  num_columns: 4
  num_rows: 3
  num_row_groups: 1
  format_version: 2.6
  serialized_size: ...

or can also be read directly using read_metadata():

>>> metadata = pq.read_metadata('example.parquet')
>>> metadata
<pyarrow._parquet.FileMetaData object at ...>
  created_by: parquet-cpp-arrow version ...
  num_columns: 4
  num_rows: 3
  num_row_groups: 1
  format_version: 2.6
  serialized_size: ...

The returned FileMetaData object allows to inspect the Parquet file metadata, such as the row groups and column chunk metadata and statistics:

>>> metadata.row_group(0)
<pyarrow._parquet.RowGroupMetaData object at ...>
  num_columns: 4
  num_rows: 3
  total_byte_size: 290
  sorting_columns: ()
>>> metadata.row_group(0).column(0)
<pyarrow._parquet.ColumnChunkMetaData object at ...>
  file_offset: 0
  file_path:...
  physical_type: DOUBLE
  num_values: 3
  path_in_schema: one
  is_stats_set: True
  statistics:
    <pyarrow._parquet.Statistics object at ...>
      has_min_max: True
      min: -1.0
      max: 2.5
      null_count: 1
      distinct_count: None
      num_values: 2
      physical_type: DOUBLE
      logical_type: None
      converted_type (legacy): NONE
  geo_statistics:
    None
  compression: SNAPPY
  encodings: ('PLAIN', 'RLE', 'RLE_DICTIONARY')
  has_dictionary_page: True
  dictionary_page_offset: 4
  data_page_offset: 36
  total_compressed_size: 106
  total_uncompressed_size: 102
  bloom_filter_offset: None
  bloom_filter_length: None

Filtering / Predicate Pushdown#

The filters parameter of read_table() pushes predicates into the reader, skipping entire row groups based on their column statistics (min/max values stored in the row group metadata). For large files with many row groups this can dramatically reduce the data read.

Predicates can be expressed as a Expression:

>>> import pyarrow.compute as pc
>>> pq.read_table('example.parquet', filters=pc.field('three') == True)
pyarrow.Table
one: double
two: large_string
three: bool
__index_level_0__: large_string
----
one: [[-1,2.5]]
two: [["foo","baz"]]
three: [[true,true]]
__index_level_0__: [["a","c"]]

Or in disjunctive normal form (DNF) as a list of (column, operator, value) tuples:

>>> pq.read_table('example.parquet', filters=[('three', '==', True)])
pyarrow.Table
one: double
two: large_string
three: bool
__index_level_0__: large_string
----
one: [[-1,2.5]]
two: [["foo","baz"]]
three: [[true,true]]
__index_level_0__: [["a","c"]]

Supported operators are = / ==, !=, <, >, <=, >=, in, and not in. Multiple predicates in the inner list are combined as AND; the outer list combines predicate groups as OR:

>>> filters = [
...     [('one', '>', 0), ('three', '==', True)],
...     [('two', 'in', ['foo', 'bar'])],
... ]
>>> pq.read_table('example.parquet', filters=filters)
pyarrow.Table
one: double
two: large_string
three: bool
__index_level_0__: large_string
----
one: [[-1,null,2.5]]
two: [["foo","bar","baz"]]
three: [[true,false,true]]
__index_level_0__: [["a","b","c"]]

Row group skipping based on column statistics is automatic when filters are applied.

Writing Bloom Filters#

Bloom filters are a probabilistic data structure stored per column per row group that can answer, for a given value, either that value is “definitely not present” or “probably present”. This makes them useful for readers that support Bloom filter-based row group skipping.

The probability of false positives is configurable.

Note

PyArrow reader does not currently use Bloom filters during filtering on read.

Bloom filters are enabled per column via the bloom_filter_options parameter:

>>> pq.write_table(table, 'bloom.parquet',
...                bloom_filter_options={'two': True})

Passing True uses the defaults: ndv=1048576 (number of distinct values) and fpp=0.05 (false-positive probability).

Both can also be configured explicitly via a dictionary:

>>> pq.write_table(table, 'bloom.parquet',
...                bloom_filter_options={
...                    'two': {'ndv': 100, 'fpp': 0.01},
...                })

Recommended value for ndv is the number of rows. Lower values of fpp reduce false positives but require more space (space grows roughly proportional to log(1/FPP)). Recommended values are 0.1, 0.05, or 0.01.

Whether a Bloom filter was written can be confirmed with observing column chunk metadata, see: Inspecting the Parquet File Metadata.

Multithreaded Reads#

Each of the reading functions by default use multi-threading for reading columns in parallel. Depending on the speed of IO and how expensive it is to decode the columns in a particular file (particularly with GZIP compression), this can yield significantly higher data throughput.

This can be disabled by specifying use_threads=False.

Note

The number of threads to use concurrently is automatically inferred by Arrow and can be inspected using the cpu_count() function.

Compression, Encoding, and File Compatibility#

The most commonly used Parquet implementations use dictionary encoding when writing files; if the dictionaries grow too large, then they “fall back” to plain encoding. Whether dictionary encoding is used can be toggled using the use_dictionary option:

>>> pq.write_table(table, 'example.parquet', use_dictionary=False)

The data pages within a column in a row group can be compressed after the encoding passes (dictionary, RLE encoding). In PyArrow we use Snappy compression by default, but Brotli, Gzip, ZSTD, LZ4, and uncompressed are also supported:

>>> pq.write_table(table, 'example.parquet', compression='snappy')
>>> pq.write_table(table, 'example.parquet', compression='gzip')
>>> pq.write_table(table, 'example.parquet', compression='brotli')
>>> pq.write_table(table, 'example.parquet', compression='zstd')
>>> pq.write_table(table, 'example.parquet', compression='lz4')
>>> pq.write_table(table, 'example.parquet', compression='none')

Snappy generally results in better performance, while Gzip may yield smaller files.

'lz4_raw' is also accepted as an alias for 'lz4'. Both use the LZ4_RAW codec as defined in the Parquet specification.

These settings can also be set on a per-column basis:

>>> pq.write_table(table, 'example.parquet', compression={'one': 'snappy', 'two': 'gzip'},
...                use_dictionary=['one', 'two'])

Using with Spark#

Spark places some constraints on the types of Parquet files it will read. The option flavor='spark' will set these options automatically and also sanitize field characters unsupported by Spark SQL.

Reading from cloud storage#

In addition to local files, pyarrow supports other filesystems, such as cloud filesystems, through the filesystem keyword:

>>> from pyarrow import fs

>>> s3  = fs.S3FileSystem(region="us-east-2")
>>> table = pq.read_table("bucket/object/key/prefix", filesystem=s3)

Currently, HDFS and Amazon S3-compatible storage are supported. See the Filesystem Interface docs for more details. For those built-in filesystems, the filesystem can also be inferred from the file path, if specified as a URI:

>>> table = pq.read_table("s3://bucket/object/key/prefix")

Other filesystems can still be supported if there is an fsspec-compatible implementation available. See Using fsspec-compatible filesystems with Arrow for more details. One example is Azure Blob storage, which can be interfaced through the adlfs package.

>>> from adlfs import AzureBlobFileSystem
>>> abfs = AzureBlobFileSystem(account_name="XXXX", account_key="XXXX", container_name="XXXX")
>>> table = pq.read_table("file.parquet", filesystem=abfs)

Content-Defined Chunking#

Note

This feature is experimental and may change in future releases.

PyArrow introduces an experimental feature for optimizing Parquet files for content addressable storage (CAS) systems using content-defined chunking (CDC). This feature enables efficient deduplication of data across files, improving network transfers and storage efficiency.

When enabled, data pages are written according to content-defined chunk boundaries, determined by a rolling hash algorithm that identifies chunk boundaries based on the actual content of the data. When data in a column is modified (e.g., inserted, deleted, or updated), this approach minimizes the number of changed data pages.

The feature can be enabled by setting the use_content_defined_chunking parameter in the Parquet writer. It accepts either a boolean or a dictionary for configuration:

True: Uses the default configuration with:
- Minimum chunk size: 256 KiB
- Maximum chunk size: 1024 KiB
- Normalization level: 0
dict: Allows customization of the chunking parameters:
- min_chunk_size: Minimum chunk size in bytes (default: 256 KiB).
- max_chunk_size: Maximum chunk size in bytes (default: 1024 KiB).
- norm_level: Normalization level to adjust chunk size distribution (default: 0).

Note that the chunk size is calculated on the logical values before applying any encoding or compression. The actual size of the data pages may vary based on the encoding and compression used.

Note

To make the most of this feature, you should ensure that Parquet write options remain consistent across writes and files. Using different write options (like compression, encoding, or row group size) for different files may prevent proper deduplication and lead to suboptimal storage efficiency.

>>> table = pa.Table.from_pandas(df)

>>> # Enable content-defined chunking with default settings
>>> pq.write_table(table, 'example.parquet', use_content_defined_chunking=True)

>>> # Enable content-defined chunking with custom settings
>>> pq.write_table(
...     table,
...     'example_custom.parquet',
...     use_content_defined_chunking={
...         'min_chunk_size': 128 * 1024,  # 128 KiB
...         'max_chunk_size': 512 * 1024,  # 512 KiB
...     }
... )