read

The read method will read a single Parquet file, metadata file, or directory with a recognized layout into an in-memory table.

Syntax

read(
    path: str,
    col_instructions: str = None,
    is_legacy_parquet: bool = False,
    is_refreshing: bool = False,
    file_layout: Optional[ParquetFileLayout] = None,
    table_definition: Union[Dict[str, DType], List[Column], None] = None,
    special_instructions: Optional[s3.S3Instructions] = None
) -> Table

Parameters

Parameter	Type	Description
path	str	The file to load into a table. The file should exist and end with the `.parquet` extension.
col_instructions optional	list[ColumnInstruction]	One or more optional `ColumnInstruction` objects that provide instructions for how to read particular columns in the file.
is_legacy_parquet optional	bool	Whether or not the Parquet data is legacy.
is_refreshing optional	bool	Whether or not the Parquet data represents a refreshing source.
file_layout optional	Optional[ParquetFileLayout]	The Parquet file layout. The default is `None`, which infers the layout.
table_definition optional	Union[Dict[str, DType], List[Column], None]	The table definition. The default is `None`, which infers the table definition from the Parquet file(s). If set, `file_layout` must also be set.
special_instructions optional	Optional[s3.S3Instructions]	Special instructions for reading Parquet files. Mostly used when reading from nonlocal filesystems, such as AWS S3 buckets. Default is `None`.

Returns

A new in-memory table from a Parquet file, metadata file, or directory with a recognized layout.

Examples

Note

All but the final example in this document use data mounted in /data in Deephaven. For more information on the relation between this location in Deephaven and on your local file system, see Docker data volumes.

Single Parquet file

In this example, read is used to load the file /data/examples/Taxi/parquet/taxi.parquet into a Deephaven table.

from deephaven.parquet import read

source = read("/data/examples/Taxi/parquet/taxi.parquet")

Compression codec

In this example, read is used to load the file /data/output_GZIP.parquet, with GZIP compression, into a Deephaven table.

Caution

This file needs to exist for this example to work. To generate this file, see write.

from deephaven.parquet import read, write
from deephaven import new_table
from deephaven.column import string_col, int_col

source = new_table(
    [
        string_col("X", ["A", "B", "B", "C", "B", "A", "B", "B", "C"]),
        int_col("Y", [2, 4, 2, 1, 2, 3, 4, 2, 3]),
        int_col("Z", [55, 76, 20, 4, 230, 50, 73, 137, 214]),
    ]
)

write(source, "/data/output_GZIP.parquet", compression_codec_name="GZIP")

source = read("/data/output_GZIP.parquet")

Partitioned datasets

_metadata and/or _common_metadata files are occasionally present in partitioned datasets. These files can be used to load Parquet data sets more quickly. These files are specific to only certain frameworks and are not required to read the data into a Deephaven table.

_common_metadata: File containing schema information needed to load the whole dataset faster.
_metadata: File containing (1) complete relative pathnames to individual data files, and (2) column statistics, such as min, max, etc., for the individual data files.

Warning

For a directory of Parquet files, all sub-directories are also searched. Only files with a .parquet extension or _common_metadata and _metadata files should be located in these directories. All files ending with .parquet need the same schema.

Note

The following examples use data in Deephaven's example repository. Follow the instructions in Launch Deephaven from pre-built images to download and manage the example data.

In this example, read is used to load the directory /data/examples/Pems/parquet/pems into a Deephaven table.

from deephaven.parquet import read

source = read("/data/examples/Pems/parquet/pems")

Refreshing tables

The following example set demonstrates how to read refreshing Parquet files into Deephaven.

First, we create a Parquet table with write.

from deephaven import new_table
from deephaven.column import int_col, double_col, string_col
from deephaven import parquet
from deephaven.parquet import read, write
import os, shutil

# Create new tables
grades1 = new_table(
    [
        string_col("Name", ["Ashton", "Jeffrey", "Samantha", "Zachary"]),
        int_col("Test1", [92, 78, 87, 74]),
        int_col("Test2", [94, 88, 81, 70]),
        int_col("Average", [93, 83, 84, 72]),
        double_col("GPA", [3.9, 2.9, 3.0, 1.8]),
    ]
)
grades2 = new_table(
    [
        string_col("Name", ["Josh", "Martin", "Mariah", "Rick"]),
        int_col("Test1", [67, 92, 87, 54]),
        int_col("Test2", [97, 99, 92, 63]),
        int_col("Average", [82, 96, 93, 59]),
        double_col("GPA", [4.0, 3.2, 3.6, 2.7]),
    ]
)

# Write both tables to parquet files
write(grades1, "/data/grades/part1.parquet")
write(grades2, "/data/grades/part2.parquet")

## Read the tables. is_refreshing must be True, or the result will be static
refreshing_result = read(path="/data/grades/", is_refreshing=True)

static_result = read(path="/data/grades/", is_refreshing=False)

Next, we list our current partitions, and then create a new Deephaven table by using the read method.

# List the files in /tmp/grades
print(sorted(os.listdir("/data/grades")))

# Create table from parquet files
grades = parquet.read("/data/grades")

Finally, we create a third partition that is a copy of the first.

# Make a 3rd partition by copying the first
shutil.copyfile("/data/grades/part1.parquet", "/data/grades/part3.parquet")

Read from a nonlocal filesystem

Deephaven currently supports reading Parquet files from your local filesystem and AWS S3. The following code block uses special instructions to read a public Parquet dataset from an S3 bucket.

from deephaven import parquet
from deephaven.experimental import s3
from datetime import timedelta

drivestats = parquet.read(
    "s3://drivestats-parquet/drivestats/year=2023/month=02/2023-02-1.parquet",
    special_instructions=s3.S3Instructions(
        region_name="us-west-004",
        endpoint_override="https://s3.us-west-004.backblazeb2.com",
        anonymous_access=True,
        read_ahead_count=8,
        fragment_size=65536,
        read_timeout=timedelta(seconds=10),
    ),
)

When reading from AWS S3, you must always specify instructions for doing so via S3Instructions. The following input parameters can be used to construct these special instructions:

region_name: This mandatory parameter defines the region name of the AWS S3 bucket where the Parquet data exists.
max_concurrent_requests: The maximum number of concurrent requests to make to S3. The default is 50.
read_ahead_count: The number of fragments that are asynchronously read ahead of the current fragment as the current fragment is being read. The default is 1.
fragment_size: The maximum size of each fragment to read in bytes. The default is 5 MB.
max_cache_size: The maximum number of fragments to cache in memory while reading. The default is 32.
connection_timeout: The amount of time to wait for a successful S3 connection before timing out. The default is 2 seconds.
read_timeout: The amount of time it takes to time out while reading a fragment. The default is 2 seconds.
access_key_id: The access key for reading files. If set, secret_access_key must also be set.
secret_access_key: The secret access key for reading files.
anonymous_access: A boolean indicating to use anonymous credentials. The default is False.
endpoint_override: The endpoint to connect to. The default is None.

Additionally, the S3.maxFragmentSize configuration property can be set upon server startup. It sets the buffer size when reading Parquet from S3. The default is 5 MB. The buffer size should be set based on the largest expected fragment.