beam-postgres

beam-postgres

Light IO transforms for Postgres read/write in Apache Beam pipelines.

Goal

The project aims to provide highly performant and customizable transforms and is not intended to support many different SQL database engines.

Features

• ReadAllFromPostgres, ReadFromPostgres`` and WriteToPostgres` transforms
• Records can be mapped to tuples, dictionaries or dataclasses
• Reads and writes are in configurable batches

Usage

Printing data from the database table:

import apache_beam as beam
from psycopg.rows import dict_row

from beam_postgres.io import ReadAllFromPostgres

with beam.Pipeline() as p:
    data = p | "Reading example records from database" >> ReadAllFromPostgres(
        "host=localhost dbname=examples user=postgres password=postgres",
        "select id, data from source",
        dict_row,
    )
    data | "Writing to stdout" >> beam.Map(print)

Writing data to the database table:

from dataclasses import dataclass

import apache_beam as beam
from apache_beam.options.pipeline_options import PipelineOptions

from beam_postgres.io import WriteToPostgres


@dataclass
class Example:
    data: str


with beam.Pipeline(options=PipelineOptions()) as p:
    data = p | "Reading example records" >> beam.Create(
        [
            Example("example1"),
            Example("example2"),
        ]
    )
    data | "Writing example records to database" >> WriteToPostgres(
        "host=localhost dbname=examples user=postgres password=postgres",
        "insert into sink (data) values (%(data)s)",
    )

See here for more examples.

Reading in batches

There may be situations when you have so much data that it will not fit into the memory - then you want to read your table data in batches. You can see an example code here (the code reads records in a batches of 1).