Skip to main content
Version: 0.5.4 (latest)

Synapse

Install dlt with Synapse​

To install the dlt library with Synapse dependencies:

pip install "dlt[synapse]"

Setup guide​

Prerequisites​

  • Microsoft ODBC Driver for SQL Server

    The Microsoft ODBC Driver for SQL Server must be installed to use this destination. This can't be included with dlt's python dependencies, so you must install it separately on your system. You can find the official installation instructions here.

    Supported driver versions:

    • ODBC Driver 18 for SQL Server

      💡 Older driver versions don't work properly because they don't support the LongAsMax keyword that was introduced in ODBC Driver 18 for SQL Server. Synapse does not support the legacy "long data types", and requires "max data types" instead. dlt uses the LongAsMax keyword to automatically do the conversion.

  • Azure Synapse Workspace and dedicated SQL pool

    You need an Azure Synapse workspace with a dedicated SQL pool to load data into. If you don't have one yet, you can use this quickstart.

Steps​

1. Initialize a project with a pipeline that loads to Synapse by running

dlt init chess synapse

2. Install the necessary dependencies for Synapse by running

pip install -r requirements.txt

This will install dlt with the synapse extra that contains all dependencies required for the Synapse destination.

3. Create a loader user

Execute the following SQL statements to set up the loader user. Change the password and replace yourpool with the name of your dedicated SQL pool:

-- on master database, using a SQL admin account

CREATE LOGIN loader WITH PASSWORD = 'your_loader_password';
-- on yourpool database

CREATE USER loader FOR LOGIN loader;

-- DDL permissions
GRANT CREATE SCHEMA ON DATABASE :: yourpool TO loader;
GRANT CREATE TABLE ON DATABASE :: yourpool TO loader;
GRANT CREATE VIEW ON DATABASE :: yourpool TO loader;

-- DML permissions
GRANT ADMINISTER DATABASE BULK OPERATIONS TO loader; -- only required when loading from staging Storage Account

Optionally, you can create a WORKLOAD GROUP and add the loader user as a member to manage workload isolation. See the instructions on setting up a loader user for an example of how to do this.

4. Enter your credentials into .dlt/secrets.toml.

Example, replace with your database connection info:

[destination.synapse.credentials]
database = "yourpool"
username = "loader"
password = "your_loader_password"
host = "your_synapse_workspace_name.sql.azuresynapse.net"

Equivalently, you can also pass a connection string as follows:

# keep it at the top of your toml file! before any section starts
destination.synapse.credentials = "synapse://loader:your_loader_password@your_synapse_workspace_name.azuresynapse.net/yourpool"

To pass credentials directly you can use the credentials argument of dlt.destinations.synapse(...):

pipeline = dlt.pipeline(
    pipeline_name='chess',
    destination=dlt.destinations.synapse(
        credentials='synapse://loader:your_loader_password@your_synapse_workspace_name.azuresynapse.net/yourpool'
    ),
    dataset_name='chess_data'
)

To use Active Directory Principal, you can use the sqlalchemy.engine.URL.create method to create the connection URL using your Active Directory Service Principal credentials. First create the connection string as:

conn_str = (
    f"DRIVER={{ODBC Driver 18 for SQL Server}};"
    f"SERVER={server_name};"
    f"DATABASE={database_name};"
    f"UID={service_principal_id}@{tenant_id};"
    f"PWD={service_principal_secret};"
    f"Authentication=ActiveDirectoryServicePrincipal"
)

Next, create the connection URL:

connection_url = URL.create(
    "mssql+pyodbc",
    query={"odbc_connect": conn_str}
)

Once you have the connection URL, you can directly use it in your pipeline configuration or convert it to a string.

pipeline = dlt.pipeline(
    pipeline_name='chess',
    destination=dlt.destinations.synapse(
        credentials=connection_url.render_as_string(hide_password=False)
    ),
    dataset_name='chess_data'
)

Write disposition​

All write dispositions are supported.

❗ The staging-optimized replace strategy is not implemented for Synapse.

Data loading​

Data is loaded via INSERT statements by default.

💡 Multi-row INSERT INTO ... VALUES statements are not possible in Synapse, because it doesn't support the Table Value Constructor. dlt uses INSERT INTO ... SELECT ... UNION statements as described here to work around this limitation.

Supported file formats​

Data type limitations​

  • Synapse cannot load TIME columns from parquet files. dlt will fail such jobs permanently. Use the insert_values file format instead, or convert datetime.time objects to str or datetime.datetime, to load TIME columns.
  • Synapse does not have a complex/JSON/struct data type. The dlt complex data type is mapped to the nvarchar type in Synapse.

Table index type​

The table index type of the created tables can be configured at the resource level with the synapse_adapter:

from dlt.destinations.adapters import synapse_adapter

info = pipeline.run(
    synapse_adapter(
        data=your_resource,
        table_index_type="clustered_columnstore_index",
    )
)

Possible values:

❗ Important:

  • Set default_table_index_type to "clustered_columnstore_index" if you want to change the default (see additional destination options).
  • CLUSTERED COLUMNSTORE INDEX tables do not support the varchar(max), nvarchar(max), and varbinary(max) data types. If you don't specify the precision for columns that map to any of these types, dlt will use the maximum lengths varchar(4000), nvarchar(4000), and varbinary(8000).
  • While Synapse creates CLUSTERED COLUMNSTORE INDEXES by default, dlt creates HEAP tables by default. HEAP is a more robust choice because it supports all data types and doesn't require conversions.
  • When using the insert-from-staging replace strategy, the staging tables are always created as HEAP tables—any configuration of the table index types is ignored. The HEAP strategy makes sense for staging tables for reasons explained here.
  • dlt system tables are always created as HEAP tables, regardless of any configuration. This is in line with Microsoft's recommendation that "for small lookup tables, less than 60 million rows, consider using HEAP or clustered index for faster query performance."
  • Child tables, if any, inherit the table index type of their parent table.

Supported column hints​

Synapse supports the following column hints:

  • primary_key - creates a PRIMARY KEY NONCLUSTERED NOT ENFORCED constraint on the column
  • unique - creates a UNIQUE NOT ENFORCED constraint on the column

❗ These hints are disabled by default. This is because the PRIMARY KEY and UNIQUE constraints are tricky in Synapse: they are not enforced and can lead to inaccurate results if the user does not ensure all column values are unique. For the column hints to take effect, the create_indexes configuration needs to be set to True, see additional destination options.

Staging support​

Synapse supports Azure Blob Storage (both standard and ADLS Gen2) as a file staging destination. dlt first uploads Parquet files to the blob container, and then instructs Synapse to read the Parquet file and load its data into a Synapse table using the COPY INTO statement.

Please refer to the Azure Blob Storage filesystem documentation to learn how to configure credentials for the staging destination. By default, dlt will use these credentials for both the write into the blob container, and the read from it to load into Synapse. Managed Identity authentication can be enabled through the staging_use_msi option (see additional destination options).

To run Synapse with staging on Azure Blob Storage:

# Create a dlt pipeline that will load
# chess player data to the snowflake destination
# via staging on Azure Blob Storage
pipeline = dlt.pipeline(
    pipeline_name='chess_pipeline',
    destination='synapse',
    staging='filesystem', # add this to activate the staging location
    dataset_name='player_data'
)

Additional destination options​

The following settings can optionally be configured:

[destination.synapse]
default_table_index_type = "heap"
create_indexes = "false"
staging_use_msi = "false"

[destination.synapse.credentials]
port = "1433"
connect_timeout = 15

port and connect_timeout can also be included in the connection string:

# keep it at the top of your toml file! before any section starts
destination.synapse.credentials = "synapse://loader:your_loader_password@your_synapse_workspace_name.azuresynapse.net:1433/yourpool?connect_timeout=15"

Descriptions:

  • default_table_index_type sets the table index type that is used if no table index type is specified on the resource.
  • create_indexes determines if primary_key and unique column hints are applied.
  • staging_use_msi determines if the Managed Identity of the Synapse workspace is used to authorize access to the staging Storage Account. Ensure the Managed Identity has the Storage Blob Data Reader role (or a higher-privileged role) assigned on the blob container if you set this option to "true".
  • port used for the ODBC connection.
  • connect_timeout sets the timeout for the pyodbc connection attempt, in seconds.

dbt support​

Integration with dbt is currently not supported.

Syncing of dlt state​

This destination fully supports dlt state sync.

Additional Setup 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.