Skip to main content

Loading Data from PostgreSQL to Databricks with dlt in Python

Need help deploying these pipelines, or figuring out how to run them in your data stack?

Join our Slack community or book a call with our support engineer Violetta.

Loading data from PostgreSQL to Databricks involves using the open-source python library dlt. PostgreSQL is a powerful, open-source object-relational database system with over 35 years of active development, known for its reliability, feature robustness, and performance. On the other hand, Databricks is a unified data analytics platform created by the original developers of Apache Spark™, designed to accelerate innovation by merging data science, engineering, and business. This documentation will guide you through the process of using dlt to transfer data seamlessly from PostgreSQL to Databricks, leveraging the strengths of both platforms. For more information on PostgreSQL, visit here.

dlt Key Features​

  • Pipeline Metadata: dlt pipelines leverage metadata to provide governance capabilities, including load IDs for tracking data loads and enabling data lineage and traceability. Read more about lineage.
  • Schema Enforcement and Curation: Ensure data consistency and quality by enforcing and curating schemas, guiding the processing and loading of data. Read more about Adjust a schema docs.
  • Schema Evolution: dlt notifies users of schema changes, allowing them to review, validate, and update downstream processes. This proactive governance promotes data consistency and control. Read more about schema changes.
  • Scaling and Finetuning: dlt offers several mechanisms and configuration options to scale up and finetune pipelines, including parallel processing and memory buffer adjustments. Read more about performance.
  • Robust Governance Support: dlt pipelines provide governance support through metadata utilization, schema enforcement, and schema change alerts, promoting data consistency, traceability, and control. Read more about governance support.

Getting started with your pipeline locally​

0. Prerequisites​

dlt requires Python 3.8 or higher. Additionally, you need to have the pip package manager installed, and we recommend using a virtual environment to manage your dependencies. You can learn more about preparing your computer for dlt in our installation reference.

1. Install dlt​

First you need to install the dlt library with the correct extras for Databricks:

pip install "dlt[databricks]"

The dlt cli has a useful command to get you started with any combination of source and destination. For this example, we want to load data from PostgreSQL to Databricks. You can run the following commands to create a starting point for loading data from PostgreSQL to Databricks:

# create a new directory
mkdir sql_database_postgres_pipeline
cd sql_database_postgres_pipeline
# initialize a new pipeline with your source and destination
dlt init sql_database databricks
# install the required dependencies
pip install -r requirements.txt

The last command will install the required dependencies for your pipeline. The dependencies are listed in the requirements.txt:


sqlalchemy>=1.4
dlt[databricks]>=0.4.7

You now have the following folder structure in your project:

sql_database_postgres_pipeline/
├── .dlt/
│   ├── config.toml          # configs for your pipeline
│   └── secrets.toml         # secrets for your pipeline
├── sql_database/                   # folder with source specific files
│   └── ...
├── sql_database_pipeline.py        # your main pipeline script
├── requirements.txt         # dependencies for your pipeline
└── .gitignore               # ignore files for git (not required)

2. Configuring your source and destination credentials​

The dlt cli will have created a .dlt directory in your project folder. This directory contains a config.toml file and a secrets.toml file that you can use to configure your pipeline. The automatically created version of these files look like this:

generated config.toml​

# put your configuration values here

[runtime]
log_level="WARNING"  # the system log level of dlt
# use the dlthub_telemetry setting to enable/disable anonymous usage data reporting, see https://dlthub.com/docs/telemetry
dlthub_telemetry = true

generated secrets.toml​

# put your secret values and credentials here. do not share this file and do not push it to github

[sources.sql_database.credentials]
drivername = "drivername" # please set me up!
database = "database" # please set me up!
password = "password" # please set me up!
username = "username" # please set me up!
host = "host" # please set me up!
port = 0 # please set me up!

[destination.databricks]
dataset_name = "dataset_name" # please set me up!

[destination.databricks.credentials]
catalog = "catalog" # please set me up!
server_hostname = "server_hostname" # please set me up!
http_path = "http_path" # please set me up!
access_token = "access_token" # please set me up!

2.1. Adjust the generated code to your usecase​

Further help setting up your source and destinations
  • Read more about setting up the PostgreSQL source in our docs.
  • Read more about setting up the Databricks destination in our docs.

The default sql_database pipeline is configured to connect to an example postgres database. The sql_database source supports all sql dialects supported by SQLAlchemy. Please refer to the PostgreSQL of the SQLAlchemy docs for additional info about SQLAlchemy and PostgresSQL.

To connect to PostgresSQL with this example pipeline, you'll need to install an PostgresSQL Python DB API driver:

pip install psycopg2

And use an PostgresSQL connection String in your code:

credentials = ConnectionStringCredentials(
    "postgresql+psycopg2://username:password@host:5432/database"
)

Or your secrets.toml:

[sources.sql_database.credentials]
drivername = "postgresql+psycopg2" 
database = "database"
password = "password"
username = "username"
host = "host"
port = 5432

3. Running your pipeline for the first time​

The dlt cli has also created a main pipeline script for you at sql_database_pipeline.py, as well as a folder sql_database that contains additional python files for your source. These files are your local copies which you can modify to fit your needs. In some cases you may find that you only need to do small changes to your pipelines or add some configurations, in other cases these files can serve as a working starting point for your code, but will need to be adjusted to do what you need them to do.

The main pipeline script will look something like this:


import sqlalchemy as sa
import humanize

import dlt
from dlt.common import pendulum
from dlt.sources.credentials import ConnectionStringCredentials

from sql_database import sql_database, sql_table, Table


def load_select_tables_from_database() -> None:
    """Use the sql_database source to reflect an entire database schema and load select tables from it.

    This example sources data from the public Rfam MySQL database.
    """
    # Create a pipeline
    pipeline = dlt.pipeline(
        pipeline_name="rfam", destination='databricks', dataset_name="rfam_data"
    )

    # Credentials for the sample database.
    # Note: It is recommended to configure credentials in `.dlt/secrets.toml` under `sources.sql_database.credentials`
    credentials = ConnectionStringCredentials(
        "mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam"
    )
    # To pass the credentials from `secrets.toml`, comment out the above credentials.
    # And the credentials will be automatically read from `secrets.toml`.

    # Configure the source to load a few select tables incrementally
    source_1 = sql_database(credentials).with_resources("family", "clan")
    # Add incremental config to the resources. "updated" is a timestamp column in these tables that gets used as a cursor
    source_1.family.apply_hints(incremental=dlt.sources.incremental("updated"))
    source_1.clan.apply_hints(incremental=dlt.sources.incremental("updated"))

    # Run the pipeline. The merge write disposition merges existing rows in the destination by primary key
    info = pipeline.run(source_1, write_disposition="merge")
    print(info)

    # Load some other tables with replace write disposition. This overwrites the existing tables in destination
    source_2 = sql_database(credentials).with_resources("features", "author")
    info = pipeline.run(source_2, write_disposition="replace")
    print(info)

    # Load a table incrementally with append write disposition
    # this is good when a table only has new rows inserted, but not updated
    source_3 = sql_database(credentials).with_resources("genome")
    source_3.genome.apply_hints(incremental=dlt.sources.incremental("created"))

    info = pipeline.run(source_3, write_disposition="append")
    print(info)


def load_entire_database() -> None:
    """Use the sql_database source to completely load all tables in a database"""
    pipeline = dlt.pipeline(
        pipeline_name="rfam", destination='databricks', dataset_name="rfam_data"
    )

    # By default the sql_database source reflects all tables in the schema
    # The database credentials are sourced from the `.dlt/secrets.toml` configuration
    source = sql_database()

    # Run the pipeline. For a large db this may take a while
    info = pipeline.run(source, write_disposition="replace")
    print(
        humanize.precisedelta(
            pipeline.last_trace.finished_at - pipeline.last_trace.started_at
        )
    )
    print(info)


def load_standalone_table_resource() -> None:
    """Load a few known tables with the standalone sql_table resource, request full schema and deferred
    table reflection"""
    pipeline = dlt.pipeline(
        pipeline_name="rfam_database",
        destination='databricks',
        dataset_name="rfam_data",
        full_refresh=True,
    )

    # Load a table incrementally starting at a given date
    # Adding incremental via argument like this makes extraction more efficient
    # as only rows newer than the start date are fetched from the table
    # we also use `detect_precision_hints` to get detailed column schema
    # and defer_table_reflect to reflect schema only during execution
    family = sql_table(
        credentials="mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam",
        table="family",
        incremental=dlt.sources.incremental(
            "updated",
        ),
        detect_precision_hints=True,
        defer_table_reflect=True,
    )
    # columns will be empty here due to defer_table_reflect set to True
    print(family.compute_table_schema())

    # Load all data from another table
    genome = sql_table(
        credentials="mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam",
        table="genome",
        detect_precision_hints=True,
        defer_table_reflect=True,
    )

    # Run the resources together
    info = pipeline.extract([family, genome], write_disposition="merge")
    print(info)
    # Show inferred columns
    print(pipeline.default_schema.to_pretty_yaml())


def select_columns() -> None:
    """Uses table adapter callback to modify list of columns to be selected"""
    pipeline = dlt.pipeline(
        pipeline_name="rfam_database",
        destination='databricks',
        dataset_name="rfam_data_cols",
        full_refresh=True,
    )

    def table_adapter(table: Table) -> None:
        print(table.name)
        if table.name == "family":
            # this is SqlAlchemy table. _columns are writable
            # let's drop updated column
            table._columns.remove(table.columns["updated"])

    family = sql_table(
        credentials="mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam",
        table="family",
        chunk_size=10,
        detect_precision_hints=True,
        table_adapter_callback=table_adapter,
    )

    # also we do not want the whole table, so we add limit to get just one chunk (10 records)
    pipeline.run(family.add_limit(1))
    # only 10 rows
    print(pipeline.last_trace.last_normalize_info)
    # no "updated" column in "family" table
    print(pipeline.default_schema.to_pretty_yaml())


def select_with_end_value_and_row_order() -> None:
    """Gets data from a table withing a specified range and sorts rows descending"""
    pipeline = dlt.pipeline(
        pipeline_name="rfam_database",
        destination='databricks',
        dataset_name="rfam_data",
        full_refresh=True,
    )

    # gets data from this range
    start_date = pendulum.now().subtract(years=1)
    end_date = pendulum.now()

    family = sql_table(
        credentials="mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam",
        table="family",
        incremental=dlt.sources.incremental(  # declares desc row order
            "updated", initial_value=start_date, end_value=end_date, row_order="desc"
        ),
        chunk_size=10,
    )
    # also we do not want the whole table, so we add limit to get just one chunk (10 records)
    pipeline.run(family.add_limit(1))
    # only 10 rows
    print(pipeline.last_trace.last_normalize_info)


def my_sql_via_pyarrow() -> None:
    """Uses pyarrow backend to load tables from mysql"""

    # uncomment line below to get load_id into your data (slows pyarrow loading down)
    # dlt.config["normalize.parquet_normalizer.add_dlt_load_id"] = True

    # Create a pipeline
    pipeline = dlt.pipeline(
        pipeline_name="rfam_cx",
        destination='databricks',
        dataset_name="rfam_data_arrow_4",
    )

    def _double_as_decimal_adapter(table: sa.Table) -> None:
        """Return double as double, not decimals"""
        for column in table.columns.values():
            if isinstance(column.type, sa.Double):
                column.type.asdecimal = False

    sql_alchemy_source = sql_database(
        "mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam?&binary_prefix=true",
        backend="pyarrow",
        table_adapter_callback=_double_as_decimal_adapter,
    ).with_resources("family", "genome")

    info = pipeline.run(sql_alchemy_source)
    print(info)


def create_unsw_flow() -> None:
    """Uploads UNSW_Flow dataset to postgres via csv stream skipping dlt normalizer.
    You need to download the dataset from https://github.com/rdpahalavan/nids-datasets
    """
    from pyarrow.parquet import ParquetFile

    # from dlt.destinations import postgres

    # use those config to get 3x speedup on parallelism
    # [sources.data_writer]
    # file_max_bytes=3000000
    # buffer_max_items=200000

    # [normalize]
    # workers=3

    data_iter = ParquetFile("UNSW-NB15/Network-Flows/UNSW_Flow.parquet").iter_batches(
        batch_size=128 * 1024
    )

    pipeline = dlt.pipeline(
        pipeline_name="unsw_upload",
        # destination=postgres("postgres://loader:loader@localhost:5432/dlt_data"),
        destination='databricks',
        progress="log",
    )
    pipeline.run(
        data_iter,
        dataset_name="speed_test",
        table_name="unsw_flow_7",
        loader_file_format="csv",
    )


def test_connectorx_speed() -> None:
    """Uses unsw_flow dataset (~2mln rows, 25+ columns) to test connectorx speed"""
    import os

    # from dlt.destinations import filesystem

    unsw_table = sql_table(
        "postgresql://loader:loader@localhost:5432/dlt_data",
        "unsw_flow_7",
        "speed_test",
        # this is ignored by connectorx
        chunk_size=100000,
        backend="connectorx",
        # keep source data types
        detect_precision_hints=True,
        # just to demonstrate how to setup a separate connection string for connectorx
        backend_kwargs={"conn": "postgresql://loader:loader@localhost:5432/dlt_data"},
    )

    pipeline = dlt.pipeline(
        pipeline_name="unsw_download",
        destination='databricks',
        # destination=filesystem(os.path.abspath("../_storage/unsw")),
        progress="log",
        full_refresh=True,
    )

    info = pipeline.run(
        unsw_table,
        dataset_name="speed_test",
        table_name="unsw_flow",
        loader_file_format="parquet",
    )
    print(info)


def test_pandas_backend_verbatim_decimals() -> None:
    pipeline = dlt.pipeline(
        pipeline_name="rfam_cx",
        destination='databricks',
        dataset_name="rfam_data_pandas_2",
    )

    def _double_as_decimal_adapter(table: sa.Table) -> None:
        """Emits decimals instead of floats."""
        for column in table.columns.values():
            if isinstance(column.type, sa.Float):
                column.type.asdecimal = True

    sql_alchemy_source = sql_database(
        "mysql+pymysql://rfamro@mysql-rfam-public.ebi.ac.uk:4497/Rfam?&binary_prefix=true",
        backend="pandas",
        table_adapter_callback=_double_as_decimal_adapter,
        chunk_size=100000,
        # set coerce_float to False to represent them as string
        backend_kwargs={"coerce_float": False, "dtype_backend": "numpy_nullable"},
        # preserve full typing info. this will parse
        detect_precision_hints=True,
    ).with_resources("family", "genome")

    info = pipeline.run(sql_alchemy_source)
    print(info)


if __name__ == "__main__":
    # Load selected tables with different settings
    load_select_tables_from_database()

    # load a table and select columns
    # select_columns()

    # load_entire_database()
    # select_with_end_value_and_row_order()

    # Load tables with the standalone table resource
    # load_standalone_table_resource()

    # Load all tables from the database.
    # Warning: The sample database is very large
    # load_entire_database()

Provided you have set up your credentials, you can run your pipeline like a regular python script with the following command:

python sql_database_pipeline.py

4. Inspecting your load result​

You can now inspect the state of your pipeline with the dlt cli:

dlt pipeline sql_database info

You can also use streamlit to inspect the contents of your Databricks destination for this:

# install streamlit
pip install streamlit
# run the streamlit app for your pipeline with the dlt cli:
dlt pipeline sql_database show

5. Next steps to get your pipeline running in production​

One of the beauties of dlt is, that we are just a plain Python library, so you can run your pipeline in any environment that supports Python >= 3.8. We have a couple of helpers and guides in our docs to get you there:

The Deploy section will show you how to deploy your pipeline to​

  • Deploy with GitHub Actions: Learn how to deploy a dlt pipeline using GitHub Actions for CI/CD. Github Actions
  • Deploy with Airflow and Google Composer: Follow this guide to deploy a dlt pipeline with Airflow and Google Composer. Airflow
  • Deploy with Google Cloud Functions: Discover how to deploy your dlt pipeline using Google Cloud Functions. Google cloud functions
  • Explore other deployment options: Check out additional methods for deploying dlt pipelines. and others...

The running in production section will teach you about:​

  • How to Monitor your pipeline: Learn how to effectively monitor your dlt pipelines to ensure they are running smoothly and efficiently. How to Monitor your pipeline
  • Set up alerts: Discover how to set up alerts for your dlt pipelines to get notified about important events and issues. Set up alerts
  • Set up tracing: Understand how to implement tracing in your dlt pipelines to track their execution and performance. And set up tracing

Additional pipeline guides​

This demo works on codespaces. Codespaces is a development environment available for free to anyone with a Github account. You'll be asked to fork the demo repository and from there the README guides you with further steps.
The demo uses the Continue VSCode extension.

Off to codespaces!

DHelp

Ask a question

Welcome to "Codex Central", your next-gen help center, driven by OpenAI's GPT-4 model. It's more than just a forum or a FAQ hub – it's a dynamic knowledge base where coders can find AI-assisted solutions to their pressing problems. With GPT-4's powerful comprehension and predictive abilities, Codex Central provides instantaneous issue resolution, insightful debugging, and personalized guidance. Get your code running smoothly with the unparalleled support at Codex Central - coding help reimagined with AI prowess.