Skip to main content
Version: Python

How to write and read multiple Parquet files

This guide will show you how to read a directory of similar Parquet files into a Deephaven table, supplying just the directory path, using the read method.

Directory structure​

It is common to use Parquet for large-scale data. In Deephaven, when we load a Parquet file into a table we do not load the whole file into RAM. This means that files much larger than the available RAM can be loaded as tables. However, you might already have a directory of Parquet files that you wish to read. These Parquet files may also be partitioned into directories and sub-directories.

To read an entire directory of Parquet files (and potentially sub-directories), every Parquet file in the directory needs to have the same schema.

Create files​

To see this in practice, you first need multiple Parquet files in your directory. Start by creating the grades1 and grades2 tables, containing student names, test scores, and GPAs.

from deephaven import new_table
from deephaven.column import int_col, double_col, string_col
grades1 = new_table([
    string_col("Name", ["Ashley", "Jeff", "Rita", "Zach"]),
    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", ["Jose", "Martha", "Mary", "Richard"]),
    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])
])

Now, use the write method to export each table as a Parquet file. write takes the following arguments:

  1. The table to be written. In this case, grades1 and grades2.
  2. The Parquet file to write to. In this case, /data/grades/part1.parquet and /data/grades/part2.parquet.
  3. (Optional) parquetInstructions for writing files using compression codecs. Accepted values are:
    • LZ4: Compression codec loosely based on the LZ4 compression algorithm, but with an additional undocumented framing scheme. The framing is part of the original Hadoop compression library and was historically copied first in parquet-mr, then emulated with mixed results by parquet-cpp.
    • LZO: Compression codec based on or interoperable with the LZO compression library.
    • GZIP: Compression codec based on the GZIP format (not the closely-related "zlib" or "deflate" formats) defined by RFC 1952.
    • ZSTD: Compression codec with the highest compression ratio based on the Zstandard format defined by RFC 8478.
from deephaven.parquet import write

write(grades1, "/data/grades/part1.parquet")
write(grades2, "/data/grades/part2.parquet")
note

Deephaven writes files to locations relative to the base of its Docker container. See Docker data volumes to learn more about the relation between locations in the container and the local file system.

Read directory​

Now, use the read method to import the entire Parquet directory as one table. read takes the following arguments:

  1. The Parquet directory to read. In this case, /data/grades/.
  2. (Optional )parquetInstructions for codecs when the file type can not be successfully inferred. Accepted values are:
    • LZ4: Compression codec loosely based on the LZ4 compression algorithm, but with an additional undocumented framing scheme. The framing is part of the original Hadoop compression library and was historically copied first in parquet-mr, then emulated with mixed results by parquet-cpp.
    • LZO: Compression codec based on or interoperable with the LZO compression library.
    • GZIP: Compression codec based on the GZIP format (not the closely-related "zlib" or "deflate" formats) defined by RFC 1952.
    • ZSTD: Compression codec with the highest compression ratio based on the Zstandard format defined by RFC 8478.
    • LEGACY: Load any binary fields as strings. Helpful to load files written in older versions of Parquet that lacked a distinction between binary and string.
from deephaven.parquet import read

result = read("/data/grades")